« Back

Depeursinge Adrien

Professeur-e HES Ordinaire

Main contract

Professeur-e HES Ordinaire

Desktop: TP107

HES-SO Valais-Wallis - Haute Ecole de Gestion
Route de la Plaine 2, Case postale 80, 3960 Sierre, CH

Faculty
Economie et services

Main Degree Programme
Informatique de gestion

Research Interests

Personalized medicine via image-based tissue analysis and machine learning for cancer and other diseases (Radiomics)
Building digital models of disease-specific radiological phenotypes at the organ level using localization systems and texture-based computational models of biomedical tissue.
N-dimensional texture analysis with control of scales and orientations
Multimodal fusion
Clinical workflows of image-based computer-aided systems
Web-based tools for image-based personalized medicine research (https://quantimage2.ehealth.hevs.ch/)

No data to display for this section

2024

Making sense of radiomics :

Scientific paper ArODES

insights on human–AI collaboration in medical interaction from an observational user study

Jakub Mlynár, Adrien Depeursinge, John O. Prior, Roger Schaer, Alexandre Martroye de Joly, Florian Evéquoz

Frontiers in Communication, 8

Link to the publication

Summary:

Technologies based on “artificial intelligence” (AI) are transforming every part of our society, including healthcare and medical institutions. An example of this trend is the novel field in oncology and radiology called radiomics, which is the extracting and mining of large-scale quantitative features from medical imaging by machine-learning (ML) algorithms. This paper explores situated work with a radiomics software platform, QuantImage (v2), and interaction around it, in educationally framed hands-on trial sessions where pairs of novice users (physicians and medical radiology technicians) work on a radiomics task consisting of developing a predictive ML model with a co-present tutor. Informed by ethnomethodology and conversation analysis (EM/CA), the results show that learning about radiomics more generally and learning how to use this platform specifically are deeply intertwined. Common-sense knowledge (e.g., about meanings of colors) can interfere with the visual representation standards established in the professional domain. Participants' skills in using the platform and knowledge of radiomics are routinely displayed in the assessment of performance measures of the resulting ML models, in the monitoring of the platform's pace of operation for possible problems, and in the ascribing of independent actions (e.g., related to algorithms) to the platform. The findings are relevant to current discussions about the explainability of AI in medicine as well as issues of machinic agency.

2023

Wide kernels and their DCT compression in convolutional networks for nuclei segmentation

Scientific paper ArODES

Vincent Andrearczyk, Valentin Oreiller, Adrien Depeursinge

Informatics in Medicine Unlocked, 43, 101403

Link to the publication

Summary:

The locality and spatial field of view of image operators have played a major role in image analysis, from hand-crafted to deep learning methods. In Convolutional Neural Networks (CNNs), the field of view is traditionally set to very small values (e.g. 3 × 3 pixels) for individual kernels and grown throughout the network by cascading layers. Automatically learning or adapting the best spatial support of the kernels can be done by using large kernels. Due to the computation requirements of standard CNN architectures, this has been little investigated in the literature. However, if large receptive fields are needed to capture wider contextual information on a given task, it could be learned from the data. Obtaining an optimal receptive field with few layers is very relevant in applications with a limited amount of annotated training data, e.g. in medical imaging. We show that CNNs (2D U-Nets) with large kernels outperform similar models with standard small kernels on the task of nuclei segmentation in histopathology images. We observe that the large kernels mostly capture low-frequency information, which motivates the need for large kernels and their efficient compression via the Discrete Cosine Transform (DCT). Following this idea, we develop a U-Net model with wide and compressed DCT kernels that leads to similar performance and trends to the standard U-Net, with reduced complexity. Visualizations of the kernels in the spatial and frequency domains, as well as the effective receptive fields, provide insights into the models’ behaviors and the learned features.

How far MS lesion detection and segmentation are integrated into the clinical workflow? :

Scientific paper ArODES

a systematic review

Frederico Spagnolo, Adrien Depeursinge, Sabine Schädelin, Aysenur Akbulut, Henning Müller, Muhamed Barakovic, Lester Melie-Garcia, Meritxell Bach Cuadra, Cristina Granziera

NeuroImage: Clinical, 39, 103491

Link to the publication

Summary:

Introduction: Over the past few years, the deep learning community has developed and validated a plethora of tools for lesion detection and segmentation in Multiple Sclerosis (MS). However, there is an important gap between validating models technically and clinically. To this end, a six-step framework necessary for the development, validation, and integration of quantitative tools in the clinic was recently proposed under the name of the Quantitative Neuroradiology Initiative (QNI). Aims: Investigate to what extent automatic tools in MS fulfill the QNI framework necessary to integrate automated detection and segmentation into the clinical neuroradiology workflow. Methods: Adopting the systematic Cochrane literature review methodology, we screened and summarised published scientific articles that perform automatic MS lesions detection and segmentation. We categorised the retrieved studies based on their degree of fulfillment of QNI’s six-steps, which include a tool’s technical assessment, clinical validation, and integration. Results: We found 156 studies; 146/156 (94%) fullfilled the first QNI step, 155/156 (99%) the second, 8/156 (5%) the third, 3/156 (2%) the fourth, 5/156 (3%) the fifth and only one the sixth. Conclusions: To date, little has been done to evaluate the clinical performance and the integration in the clinical workflow of available methods for MS lesion detection/segmentation. In addition, the socio-economic effects and the impact on patients’ management of such tools remain almost unexplored.

QuantImage v2 :

Scientific paper ArODES

a comprehensive and integrated physician-centered cloud platform for radiomics and machine learning research

Daniel Abler, Roger Schaer, Valentin Oreiller, Himanshu Verma, Julien Reichenbach, Orfeas Aidonopoulos, Florian Evéquoz, Mario Jreige, John O. Prior, Adrien Depeursinge

European Radiology Experimental, 7, 16

Link to the publication

Summary:

Background : Radiomics, the field of image-based computational medical biomarker research, has experienced rapid growth over the past decade due to its potential to revolutionize the development of personalized decision support models. However, despite its research momentum and important advances toward methodological standardization, the translation of radiomics prediction models into clinical practice only progresses slowly. The lack of physicians leading the development of radiomics models and insufficient integration of radiomics tools in the clinical workflow contributes to this slow uptake. Methods We propose a physician-centered vision of radiomics research and derive minimal functional requirements for radiomics research software to support this vision. Free-to-access radiomics tools and frameworks were reviewed to identify best practices and reveal the shortcomings of existing software solutions to optimally support physician-driven radiomics research in a clinical environment. Results : Support for user-friendly development and evaluation of radiomics prediction models via machine learning was found to be missing in most tools. QuantImage v2 (QI2) was designed and implemented to address these shortcomings. QI2 relies on well-established existing tools and open-source libraries to realize and concretely demonstrate the potential of a one-stop tool for physician-driven radiomics research. It provides web-based access to cohort management, feature extraction, and visualization and supports “no-code” development and evaluation of machine learning models against patient-specific outcome data. Conclusions : QI2 fills a gap in the radiomics software landscape by enabling “no-code” radiomics research, including model validation, in a clinical environment. Further information about QI2, a public instance of the system, and its source code is available at https://medgift.github.io/quantimage-v2-info/. Key points: - As domain experts, physicians play a key role in the development of radiomics models. - Existing software solutions do not support physician-driven research optimally. - QuantImage v2 implements a physician-centered vision for radiomics research. - QuantImage v2 is a web-based, “no-code” radiomics research platform.

3D-printed iodine-ink CT phantom for radiomics feature extraction :

Scientific paper ArODES

advantages and challenges

Michael Bach, Christoph Aberle, Adrien Depeursinge, Oscar Jimenez-del-Toro, Robert Schaer, Kyriakos Flouris, Ender Konukoglu, Henning Müller, Bram Stieltjes, Markus M. Obmann

Medical Physics, 50, 9, 5682-5697

Link to the publication

Summary:

Background : To test and validate novel CT techniques, such as texture analysis in radiomics, repeat measurements are required. Current anthropomorphic phantoms lack fine texture and true anatomic representation. 3D-printing of iodinated ink on paper is a promising phantom manufacturing technique. Previously acquired or artificially created CT data can be used to generate realistic phantoms. Purpose: To present the design process of an anthropomorphic 3D-printed iodine ink phantom, highlighting the different advantages and pitfalls in its use. To analyze the phantom's X-ray attenuation properties, and the influences of the printing process on the imaging characteristics, by comparing it to the original input dataset. Methods : Two patient CT scans and artificially generated test patterns were combined in a single dataset for phantom printing and cropped to a size of 26 × 19 × 30 cm3. This DICOM dataset was printed on paper using iodinated ink. The phantom was CT-scanned and compared to the original image dataset used for printing the phantom. The water-equivalent diameter of the phantom was compared to that of a patient cohort (N = 104). Iodine concentrations in the phantom were measured using dual-energy CT. 86 radiomics features were extracted from 10 repeat phantom scans and the input dataset. Features were compared using a histogram analysis and a PCA individually and overall, respectively. The frequency content was compared using the normalized spectrum modulus. Results : Low density structures are depicted incorrectly, while soft tissue structures show excellent visual accordance with the input dataset. Maximum deviations of around 30 HU between the original dataset and phantom HU values were observed. The phantom has X-ray attenuation properties comparable to a lightweight adult patient (∼54 kg, BMI 19 kg/m2). Iodine concentrations in the phantom varied between 0 and 50 mg/ml. PCA of radiomics features shows different tissue types separate in similar areas of PCA representation in the phantom scans as in the input dataset. Individual feature analysis revealed systematic shift of first order radiomics features compared to the original dataset, while some higher order radiomics features did not. The normalized frequency modulus |f(ω)| of the phantom data agrees well with the original data. However, all frequencies systematically occur more frequently in the phantom compared to the maximum of the spectrum modulus than in the original data set, especially for mid-frequencies (e.g., for ω = 0.3942 mm−1, |f(ω)|original = 0.09 * |fmax|original and |f(ω)|phantom = 0.12 * |fmax|phantom). Conclusions : 3D-iodine-ink-printing technology can be used to print anthropomorphic phantoms with a water-equivalent diameter of a lightweight adult patient. Challenges include small residual air enclosures and the fidelity of HU values. For soft tissue, there is a good agreement between the HU values of the phantom and input data set. Radiomics texture features of the phantom scans are similar to the input data set, but systematic shifts of radiomics features in first order features, due to differences in HU values, need to be considered. The paper substrate influences the spatial frequency distribution of the phantom scans. This phantom type is of very limited use for dual-energy CT analyses.

2022

Quality assessment for interoperable quantitative CT imaging (QA4IQI) - Open Access to standardized quantitative imaging

Report ArODES

final report - open data project call

Adrien Depeursinge, Roger Schaer

2022, Sierre : HES-SO Valais-Wallis, 11 p.

Link to the publication

Reproducibility of lung cancer radiomics features extracted from data-driven respiratory gating and free-breathing flow imaging in [18F]-FDG PET/CT

Scientific paper ArODES

Daphné Faist, Mario Jreige, Valentin Oreiller, Marie Nicod Lalonde, Niklaus Schaefer, Adrien Depeursinge, John O. Prior

European journal of hybrid imaging, 2022, vol. 6, no. 33

Link to the publication

Summary:

Background : Quality and reproducibility of radiomics studies are essential requirements for the standardisation of radiomics models. As recent data-driven respiratory gating (DDG) [18F]-FDG has shown superior diagnostic performance in lung cancer, we evaluated the impact of DDG on the reproducibility of radiomics features derived from [18F]-FDG PET/CT in comparison to free-breathing flow (FB) imaging. Methods : Twenty four lung nodules from 20 patients were delineated. Radiomics features were derived on FB flow PET/CT and on the corresponding DDG reconstruction using the QuantImage v2 platform. Lin’s concordance factor (Cb) and the mean difference percentage (DIFF%) were calculated for each radiomics feature using the delineated nodules which were also classified by anatomical localisation and volume. Non-reproducible radiomics features were defined as having a bias correction factor Cb < 0.8 and/or a mean difference percentage DIFF% > 10. Results : In total 141 features were computed on each concordance analysis, 10 of which were non-reproducible on all pulmonary lesions. Those were first-order features from Laplacian of Gaussian (LoG)-filtered images (sigma = 1 mm): Energy, Kurtosis, Minimum, Range, Root Mean Squared, Skewness and Variance; Texture features from Gray Level Cooccurence Matrix (GLCM): Cluster Prominence and Difference Variance; First-order Standardised Uptake Value (SUV) feature: Kurtosis. Pulmonary lesions located in the superior lobes had only stable radiomics features, the ones from the lower parts had 25 non-reproducible radiomics features. Pulmonary lesions with a greater size (defined as long axis length > median) showed a higher reproducibility (9 non-reproducible features) than smaller ones (20 non-reproducible features). Conclusion : Calculated on all pulmonary lesions, 131 out of 141 radiomics features can be used interchangeably between DDG and FB PET/CT acquisitions. Radiomics features derived from pulmonary lesions located inferior to the superior lobes are subject to greater variability as well as pulmonary lesions of smaller size.

A global taxonomy of interpretable AI :

Scientific paper ArODES

unifying the terminology for the technical and social sciences

Mara Graziani, Lidia Dutkiewicz, Davide Calvaresi, José Pereira Amorim, Katerina Yordanova, Mor Vered, Rahul Nair, Pedro Henriques Abreu, Tobias Blanke, Valeria Pulignano, John O. Prior, Lode Lauwaert, Wessel Reijers, Adrien Depeursinge, Vincent Andrearczyk, Henning Müller

Artificial intelligence review, To be published

Link to the publication

Summary:

Since its emergence in the 1960s, Artificial Intelligence (AI) has grown to conquer many technology products and their fields of application. Machine learning, as a major part of the current AI solutions, can learn from the data and through experience to reach high performance on various tasks. This growing success of AI algorithms has led to a need for interpretability to understand opaque models such as deep neural networks. Various requirements have been raised from different domains, together with numerous tools to debug, justify outcomes, and establish the safety, fairness and reliability of the models. This variety of tasks has led to inconsistencies in the terminology with, for instance, terms such as interpretable, explainable and transparent being often used interchangeably in methodology papers. These words, however, convey different meanings and are “weighted" differently across domains, for example in the technical and social sciences. In this paper, we propose an overarching terminology of interpretability of AI systems that can be referred to by the technical developers as much as by the social sciences community to pursue clarity and efficiency in the definition of regulations for ethical and reliable AI development. We show how our taxonomy and definition of interpretable AI differ from the ones in previous research and how they apply with high versatility to several domains and use cases, proposing a—highly needed—standard for the communication among interdisciplinary areas of AI.

Steer’n’Detect :

Scientific paper ArODES

fast 2D template detection with accurate orientation estimation

Virginie Uhlmann, Zsuzsanna Püspöki, Adrien Depeursinge, Michael Unser, Daniel Sage, Julien Fageot

Bioinformatics, 2022, vol. 38, no. 11, pp. 3146–3148

Link to the publication

Summary:

Rotated template matching is an efficient and versatile algorithm to analyze microscopy images, as it automates the detection of stereotypical structures, such as organelles that can appear at any orientation. Its performance however quickly degrades in noisy image data. We introduce Steer’n’Detect, an ImageJ plugin implementing a recently published algorithm to detect patterns of interest at any orientation with high accuracy from a single template in 2D images. Steer’n’Detect provides a faster and more robust substitute to template matching. By adapting to the statistics of the image background, it guarantees accurate results even in the presence of noise. The plugin comes with an intuitive user interface facilitating results analysis and further post-processing.

Assessing radiomics feature stability with simulated CT acquisitions

Scientific paper ArODES

Kyriakos Flouris, Oscar Alfonso Jiménez del Toro, Christoph Aberle, Michael Bach, Roger Schaer, Markus M. Obmann, Bram Stieltjes, Henning Müller, Adrien Depeursinge, Ender Konukoglu

Scientific Reports, 2022, vol. 12, article 4732, pp. 1-11

Link to the publication

Summary:

Medical imaging quantitative features had once disputable usefulness in clinical studies. Nowadays, advancements in analysis techniques, for instance through machine learning, have enabled quantitative features to be progressively useful in diagnosis and research. Tissue characterisation is improved via the “radiomics” features, whose extraction can be automated. Despite the advances, stability of quantitative features remains an important open problem. As features can be highly sensitive to variations of acquisition details, it is not trivial to quantify stability and efficiently select stable features. In this work, we develop and validate a Computed Tomography (CT) simulator environment based on the publicly available ASTRA toolbox (www.astra-toolbox.com). We show that the variability, stability and discriminative power of the radiomics features extracted from the virtual phantom images generated by the simulator are similar to those observed in a tandem phantom study. Additionally, we show that the variability is matched between a multi-center phantom study and simulated results. Consequently, we demonstrate that the simulator can be utilised to assess radiomics features’ stability and discriminative power.

Cleaning radiotherapy contours for radiomics studies, is it worth it ? :

Scientific paper ArODES

a head and neck cancer study

Pierre Fontaine, Vincent Andrearczyk, Valentin Oreiller, Daniel Abler, Joël Castelli, Oscar Acosta, Renaud De Crevoisier, Martin Vallières, Mario Jreige, John O. Prior, Adrien Depeursinge

Clinical and translational radiation oncology, 2022, vol. 33, pp. 153-158

Link to the publication

Summary:

A vast majority of studies in the radiomics field are based on contours originating from radiotherapy planning. This kind of delineation (e.g. Gross Tumor Volume, GTV) is often larger than the true tumoral volume, sometimes including parts of other organs (e.g. trachea in Head and Neck, H&N studies) and the impact of such over-segmentation was little investigated so far. In this paper, we propose to evaluate and compare the performance between models using two contour types: those from radiotherapy planning, and those specifically delineated for radiomics studies. For the latter, we modified the radiotherapy contours to fit the true tumoral volume. The two contour types were compared when predicting Progression-Free Survival (PFS) using Cox models based on radiomics features extracted from FluoroDeoxyGlucose-Positron Emission Tomography (FDG-PET) and CT images of 239 patients with oropharyngeal H&N cancer collected from five centers, the data from the 2020 HECKTOR challenge. Using Dedicated contours demonstrated better performance for predicting PFS, where Harell’s concordance indices of 0.61 and 0.69 were achieved for Radiotherapy and Dedicated contours, respectively. Using automatically Resegmented contours based on a fixed intensity range was associated with a C-index of 0.63. These results illustrate the importance of using clean dedicated contours that are close to the true tumoral volume in radiomics studies, even when tumor contours are already available from radiotherapy treatment planning

2021

FDG-PET/CT-based prognostic survival model after surgery for head and neck cancer

Scientific paper ArODES

Gwenaelle Creff, Franck Jegoux, Xavier Palard, Adrien Depeursinge, Ronan Abgral, Remi Marianowski, Jean-Christophe Leclere, Thomas Eugene, Olivier Malard, Renaud De Crevoisier, Anne Devillers, Joel Castelli

Journal of nuclear medicine, 2022, vol. 63, no. 5

Link to the publication

Summary:

Rationale: The aims of this multicenter study were to identify clinical and preoperative PET/CT parameters predicting Overall Survival (OS) and Distant Metastasis Free Survival(DMFS) from a cohort of Head and Neck Squamous Cell Carcinoma (HNSCC) patients treated with surgery, to generate a prognostic model of OS and DMFS and to validate this prognostic model with an independent cohort. Materials and Methods: A total of 382 consecutive HNSCC patients divided into training (n = 318) and validation cohorts (n = 64) were retrospectively included. The following PET/CT parameters were analyzed: clinical parameters, SUVmax, SUVMean, Metabolic Tumor Volume (MTV), Total Lesion Glycolysis (TLG) and distance parameters for the primary tumor and lymph nodes defined by two segmentation methods (relative SUVmax threshold and absolute SUV threshold). Cox analyses were performed for OS and DMFS in the training cohort. The c-index was used to identify highly prognostic parameters. These prognostic parameters were externally tested in the validation cohort. Results: In multivariable analysis, the significant parameters for OS were T stage and Nodal-MTV, achieving a c-index of 0.64 (p<0.001). For DMFS, the significant parameters were T stage, Nodal-MTV and maximal tumor-node distance, with a c-index of 0.76 (p<0.001). These combinations of parameters were externally validated, achieving c-indices of 0.63 (p<0.001) and 0.71 (p<0.001) for OS and DMFS, respectively. Conclusion: The Nodal MTV associated with maximal distance between the primary tumor and the lymph node was significantly correlated with the risk of DMFS. Moreover, this parameter in addition to clinical parameters was associated with higher risk of death. These prognostic factors may be used to tailor individualized treatment.

The discriminative power and stability of radiomics features with computed tomography variations :

Scientific paper ArODES

task-based analysis in an anthropomorphic 3D-printed CT phantom

Oscar Alfonso Jiménez del Toro, Christoph Aberle, Michael Bach, Roger Schaer, Markus Obmann, Kyriakos Flouris, Ender Konukoglu, Bram Stieltjes, Henning Müller, Adrien Depeursinge

Investigative radiology, 2021, vol. 56, no. 12, pp. 820-825

Link to the publication

Summary:

Objectives : The aims of this study were to determine the stability of radiomics features against computed tomography (CT) parameter variations and to study their discriminative power concerning tissue classification using a 3D-printed CT phantom based on real patient data. Materials and Methods : A radiopaque 3D phantom was developed using real patient data and a potassium iodide solution paper-printing technique. Normal liver tissue and 3 lesion types (benign cyst, hemangioma, and metastasis) were manually annotated in the phantom. The stability and discriminative power of 86 radiomics features were assessed in measurements taken from 240 CT series with 8 parameter variations of reconstruction algorithms, reconstruction kernels, slice thickness, and slice spacing. Pairwise parameter group and pairwise tissue class comparisons were performed using Wilcoxon signed rank tests. Results : In total, 19,264 feature stability tests and 8256 discriminative power tests were performed. The 8 CT parameter variation pairwise group comparisons had statistically significant differences on average in 78/86 radiomics features. On the other hand, 84% of the univariate radiomics feature tests had a successful and statistically significant differentiation of the 4 classes of liver tissue. The 86 radiomics features were ranked according to the cumulative sum of successful stability and discriminative power tests. Conclusions : The differences in radiomics feature values obtained from different types of liver tissue are generally greater than the intraclass differences resulting from CT parameter variations.

Head and neck tumor segmentation in PET/CT :

Scientific paper ArODES

The HECKTOR challenge

Valentin Oreiller, Vincent Andrearczyk, Mario Jreige, Sarah Boughdad, Hesham Elhalawanic, Joel Castelli, Martin Vallières, Simeng Zhu, Juanying Xie, Ying Peng, Andrei Iantsen, Matthieu Hatt, Yading Yuan, Jun Ma, Xiaoping Yang, Chinmay Rao, Suraj Pai, Kanchan Ghimire, Xue Feng, Mohamed A. Naser, Clifton D. Fuller, Fereshteh Yousefirizi, Arman Rahmim, Huai Chen, Lisheng Wang, John O. Prior, Adrien Depeursinge

Medical image analysis, 2022, vol. 77, article no. 102336, pp. 1-16

Link to the publication

Summary:

This paper relates the post-analysis of the first edition of the HEad and neCK TumOR (HECKTOR) challenge. This challenge was held as a satellite event of the 23rd International Conference on Medical Image Computing and Computer-Assisted Intervention (MICCAI) 2020, and was the first of its kind focusing on lesion segmentation in combined FDG-PET and CT image modalities. The challenge’s task is the automatic segmentation of the Gross Tumor Volume (GTV) of Head and Neck (H&N) oropharyngeal primary tumors in FDG-PET/CT images. To this end, the participants were given a training set of 201 cases from four different centers and their methods were tested on a held-out set of 53 cases from a fifth center. The methods were ranked according to the Dice Score Coefficient (DSC) averaged across all test cases. An additional inter-observer agreement study was organized to assess the difficulty of the task from a human perspective. 64 teams registered to the challenge, among which 10 provided a paper detailing their approach. The best method obtained an average DSC of 0.7591, showing a large improvement over our proposed baseline method and the inter-observer agreement, associated with DSCs of 0.6610 and 0.61, respectively. The automatic methods proved to successfully leverage the wealth of metabolic and structural properties of combined PET and CT modalities, significantly outperforming human inter-observer agreement level, semi-automatic thresholding based on PET images as well as other single modality-based methods. This promising performance is one step forward towards large-scale radiomics studies in H&N cancer, obviating the need for error-prone and time-consuming manual delineation of GTVs.

On the scale invariance in state of the art CNNs trained on ImageNet

Scientific paper ArODES

Mara Graziani, Thomas Lompech, Henning Müller, Adrien Depeursinge, Vincent Andrearczyk

Machine learning and knowledge extraction, 2021, vol. 3, no. 2, pp. 374-391

Link to the publication

Summary:

The diffused practice of pre-training Convolutional Neural Networks (CNNs) on large natural image datasets such as ImageNet causes the automatic learning of invariance to object scale variations. This, however, can be detrimental in medical imaging, where pixel spacing has a known physical correspondence and size is crucial to the diagnosis, for example, the size of lesions, tumors or cell nuclei. In this paper, we use deep learning interpretability to identify at what intermediate layers such invariance is learned. We train and evaluate different regression models on the PASCAL-VOC (Pattern Analysis, Statistical modeling and ComputAtional Learning-Visual Object Classes) annotated data to (i) separate the effects of the closely related yet different notions of image size and object scale, (ii) quantify the presence of scale information in the CNN in terms of the layer-wise correlation between input scale and feature maps in InceptionV3 and ResNet50, and (iii) develop a pruning strategy that reduces the invariance to object scale of the learned features. Results indicate that scale information peaks at central CNN layers and drops close to the softmax, where the invariance is reached. Our pruning strategy uses this to obtain features that preserve scale information. We show that the pruning significantly improves the performance on medical tasks where scale is a relevant factor, for example for the regression of breast histology image magnification. These results show that the presence of scale information at intermediate layers legitimates transfer learning in applications that require scale covariance rather than invariance and that the performance on these tasks can be improved by pruning off the layers where the invariance is learned. All experiments are performed on publicly available data and the code is available on GitHub.

Principled design and implementation of steerable detectors

Scientific paper ArODES

Julien Fageot, Virginie Uhlmann, Zsuzsanna Püspöki, Benjamin Beck, Michael Unser, Adrien Depeursinge

IEEE transactions on image processing, 2021, vol. 30, pp. 4465-4478

Link to the publication

Summary:

We provide a complete pipeline for the detection of patterns of interest in an image. In our approach, the patterns are assumed to be adequately modeled by a known template, and are located at unknown positions and orientations that we aim at retrieving. We propose a continuous-domain additive image model, where the analyzed image is the sum of the patterns to localize and a background with self-similar isotropic power-spectrum. We are then able to compute the optimal filter fulfilling the SNR criterion based on one single template and background pair: it strongly responds to the template while being optimally decoupled from the background model. In addition, we constrain our filter to be steerable, which allows for a fast template detection together with orientation estimation. In practice, the implementation requires to discretize a continuous-domain formulation on polar grids, which is performed using quadratic radial B-splines. We demonstrate the practical usefulness of our method on a variety of template approximation and pattern detection experiments. We show that the detection performance drastically improves when we exploit the statistics of the background via its power-spectrum decay, which we refer to as spectral-shaping. The proposed scheme outperforms state-of-the-art steerable methods by up to 50% of absolute detection performance.

2020

The importance of feature aggregation in radiomics :

Scientific paper ArODES

a head and neck cancer study

Pierre Fontaine, Oscar Acosta, Joël Castelli, Renaud De Crevoisier, Henning Müller, Adrien Depeursinge

Scientific reports, 2020, vol. 10, article no. 19679 , pp. 1-11

Link to the publication

Summary:

In standard radiomics studies the features extracted from clinical images are mostly quantified with simple statistics such as the average or variance per Region of Interest (ROI). Such approaches may smooth out any intra-region heterogeneity and thus hide some tumor aggressiveness that may hamper predictions. In this paper we study the importance of feature aggregation within the standard radiomics workflow, which allows to take into account intra-region variations. Feature aggregation methods transform a collection of voxel values from feature response maps (over a ROI) into one or several scalar values that are usable for statistical or machine learning algorithms. This important step has been little investigated within the radiomics workflows, so far. In this paper, we compare several aggregation methods with standard radiomics approaches in order to assess the improvements in prediction capabilities. We evaluate the performance using an aggregation function based on Bags of Visual Words (BoVW), which allows for the preservation of piece-wise homogeneous information within heterogeneous regions and compared with standard methods. The different models are compared on a cohort of 214 head and neck cancer patients coming from 4 medical centers. Radiomics features were extracted from manually delineated tumors in clinical PET-FDG and CT images were analyzed. We compared the performance of standard radiomics models, the volume of the ROI alone and the BoVW model for survival analysis. The average concordance index was estimated with a five fold cross-validation. The performance was significantly better using the BoVW model 0.627 (95% CI: 0.616–0.637) as compared to standard radiomics0.505 (95% CI: 0.499–0.511), mean-var. 0.543 (95% CI: 0.536–0.549), mean0.547 (95% CI: 0.541–0.554), var.0.530 (95% CI: 0.524–0.536) or volume 0.577 (95% CI: 0.571–0.582). We conclude that classical aggregation methods are not optimal in case of heterogeneous tumors. We also showed that the BoVW model is a better alternative to extract consistent features in the presence of lesions composed of heterogeneous tissue.

Local rotation invariance in 3D CNNs

Scientific paper ArODES

Vincent Andrearczyk, Julien Fageot, Valentin Oreiller, Xavier Montet, Adrien Depeursinge

Medical image analysis, 2020, vol. 65, article 101756, pp. 1-11

Link to the publication

Summary:

Locally Rotation Invariant (LRI) image analysis was shown to be fundamental in many applications and in particular in medical imaging where local structures of tissues occur at arbitrary rotations. LRI constituted the cornerstone of several breakthroughs in texture analysis, including Local Binary Patterns (LBP), Maximum Response 8 (MR8) and steerable filterbanks. Whereas globally rotation invariant Convolutional Neural Networks (CNN) were recently proposed, LRI was very little investigated in the context of deep learning. LRI designs allow learning filters accounting for all orientations, which enables a drastic reduction of trainable parameters and training data when compared to standard 3D CNNs. In this paper, we propose and compare several methods to obtain LRI CNNs with directional sensitivity. Two methods use orientation channels (responses to rotated kernels), either by explicitly rotating the kernels or using steerable filters. These orientation channels constitute a locally rotation equivariant representation of the data. Local pooling across orientations yields LRI image analysis. Steerable filters are used to achieve a fine and efficient sampling of 3D rotations as well as a reduction of trainable parameters and operations, thanks to a parametric representations involving solid Spherical Harmonics (SH),which are products of SH with associated learned radial profiles. Finally, we investigate a third strategy to obtain LRI based on rotational invariants calculated from responses to a learned set of solid SHs. The proposed methods are evaluated and compared to standard CNNs on 3D datasets including synthetic textured volumes composed of rotated patterns, and pulmonary nodule classification in CT. The results show the importance of LRI image analysis while resulting in a drastic reduction of trainable parameters, outperforming standard 3D CNNs trained with rotational data augmentation.

The image biomarker standardization initiative :

Scientific paper ArODES

standardized quantitative radiomics for high-throughput image-based phenotyping

Alex Zwanenburg, Martin Vallières, Mahmoud A. Abdalah, Hugo J. W. L. Aerts, Vincent Andrearczyk, Adrien Depeursinge, Henning Müller

Radiology, 2020, vol. 295, no. 2, pp. 328-338

Link to the publication

Summary:

The Image Biomarker Standardization Initiative validated consensus-based reference values for 169 radiomics features, thus enabling calibration and verification of radiomics software. Radiomic features may quantify characteristics present in medical imaging. However, the lack of standardized definitions and validated reference values have hampered clinical use. Radiomic features were assessed in three phases. In phase I, 487 features were derived from the basic set of 174 features. Twenty-five research teams with unique radiomics software implementations computed feature values directly from a digital phantom, without any additional image processing. In phase II, 15 teams computed values for 1347 derived features using a CT image of a patient with lung cancer and predefined image processing configurations. In both phases, consensus among the teams on the validity of tentative reference values was measured through the frequency of the modal value and classified as follows: less than three matches, weak; three to five matches, moderate; six to nine matches, strong; 10 or more matches, very strong. In the final phase (phase III), a public data set of multimodality images (CT, fluorine 18 fluorodeoxyglucose PET, and T1-weighted MRI) from 51 patients with soft-tissue sarcoma was used to prospectively assess reproducibility of standardized features. Consensus on reference values was initially weak for 232 of 302 features (76.8%) at phase I and 703 of 1075 features (65.4%) at phase II. At the final iteration, weak consensus remained for only two of 487 features (0.4%) at phase I and 19 of 1347 features (1.4%) at phase II. Strong or better consensus was achieved for 463 of 487 features (95.1%) at phase I and 1220 of 1347 features (90.6%) at phase II. Overall, 169 of 174 features were standardized in the first two phases. In the final validation phase (phase III), most of the 169 standardized features could be excellently reproduced (166 with CT; 164 with PET; and 164 with MRI). A set of 169 radiomics features was standardized, which enabled verification and calibration of different radiomics software.

Impact of a Gaussian filter applied to post-reconstruction PET images on radiomic features to predict complete pathological response in breast cancer

Scientific paper ArODES

Sarah Boughdad, Valentin Oreiller, Marie Meyer, Niklaus Schaefer, Marie Nicod-Lalonde, Mario Jreige, Adrien Depeursinge, John O. Prior

Journal of nuclear medicine, 2020, vol. 61, no. supplement 1, article 606

Link to the publication

Impact of a Gaussian filter applied to post-reconstruction PET on radiomic features in assessing tumor heterogeneity in breast cancer

Scientific paper ArODES

Sarah Boughdad, Valentin Oreiller, Marie Meyer, Niklaus Schaefer, Marie Nicod-Lalonde, Mario Jreige, Adrien Depeursinge, John O. Prior

Journal of nuclear medicine, 2020, vol. 61, no. supplement 1, article 612

Link to the publication

Evaluation of the prognostic value of FDG PET/CT parameters for patients with surgically treated head and neck cancer

Scientific paper ArODES

Gwenaelle Creff, Anne Deviller, Adrien Depeursinge, Xavier Palard-Novello, Oscar Acosta, Franck Jejoux, Joel Castelli

JAMA Otolaryngology–Head Neck Surgery, 2020, vol. 146, No. 5, pp. 471-479

Link to the publication

Summary:

Objective : To report the available evidence on the value of quantitative parameters of fluorodeoxyglucose F 18–labeled positron emission tomography and computed tomography (FDG-PET/CT) performed before surgical treatment of HNSCC to estimate overall survival (OS), disease-free survival (DFS), and distant metastasis (DM) and to discuss their limitations.

Integrating radiomics into holomics for personalised oncology :

Scientific paper ArODES

from algorithms to bedside

Roberto Gatta, Adrien Depeursinge, Osman Ratib, Olivier Michielin, Antoine Leimgruber

European radiology experimental, 2020, vol. 4, article 11, pp. 1-9

Link to the publication

Summary:

Radiomics, artificial intelligence, and deep learning figure amongst recent buzzwords in current medical imaging research and technological development. Analysis of medical big data in assessment and follow-up of personalised treatments has also become a major research topic in the area of precision medicine. In this review, current research trends in radiomics are analysed, from handcrafted radiomics feature extraction and statistical analysis to deep learning. Radiomics algorithms now include genomics and immunomics data to improve patient stratification and prediction of treatment response. Several applications have already shown conclusive results demonstrating the potential of including other “omics” data to existing imaging features. We also discuss further challenges of data harmonisation and management infrastructure to shed a light on the much-needed integration of radiomics and all other “omics” into clinical workflows. In particular, we point to the emerging paradigm shift in the implementation of big data infrastructures to facilitate databanks growth, data extraction and the development of expert software tools. Secured access, sharing, and integration of all health data, called “holomics”, will accelerate the revolution of personalised medicine and oncology as well as expand the role of imaging specialists.

2019

Texture-driven parametric snakes for semi-automatic image segmentation

Scientific paper ArODES

Anaïs Badoual, Michael Unser, Adrien Depeursinge

Computer vision and image understanding, 2019, vol. 188, article 102793, pp. 1-11

Link to the publication

Summary:

We present a texture-driven parametric snake for semi-automatic segmentation of a single and closed structure in an image. We propose a new energy functional that combines intensity and texture information. The two types of image information are balanced using Fisher’s linear discriminant analysis. The framework can be used with any filter-based texture features. The parametric representation of the snake allows for easy and friendly user interaction while the framework can be trained on-the-fly from pixel collections provided by the user. We demonstrate the efficiency of the snake through an extensive validation on synthetic as well as on real data. Additionally, we show that the proposed snake is robust to noise and that it improves the segmentation performance when compared to an intensity-only scheme.

Fusing learned representations from Riesz Filters and Deep CNN for lung tissue classification

Scientific paper ArODES

Ranveer Joyseeree, Sebastian Otálora, Henning Müller, Adrien Depeursinge

Medical image analysis, August 2019, vol. 56, pp. 172-183

Link to the publication

Summary:

A novel method to detect and classify several classes of diseased and healthy lung tissue in CT (Computed Tomography), based on the fusion of Riesz and deep learning features, is presented. First, discriminative parametric lung tissue texture signatures are learned from Riesz representations using a one–versus–one approach. The signatures are generated for four diseased tissue types and a healthy tissue class, all of which frequently appear in the publicly available Interstitial Lung Diseases (ILD) dataset used in this article. Because the Riesz wavelets are steerable, they can easily be made invariant to local image rotations, a property that is desirable when analyzing lung tissue micro–architectures in CT images. Second, features from deep Convolutional Neural Networks (CNN) are computed by fine–tuning the Inception V3 architecture using an augmented version of the same ILD dataset. Because CNN features are both deep and non–parametric, they can accurately model virtually any pattern that is useful for tissue discrimination, and they are the de facto standard for many medical imaging tasks. However, invariance to local image rotations is not explicitly implemented and can only be approximated with rotation–based data augmentation. This motivates the fusion of Riesz and deep CNN features, as the two techniques are very complementary. The two learned representations are combined in a joint softmax model for final classification, where early and late feature fusion schemes are compared. The experimental results show that a late fusion of the independent probabilities leads to significant improvements in classification performance when compared to each of the separate feature representations and also compared to an ensemble of deep learning approaches.

Neural network training for cross-protocol radiomic feature standardization in computed tomography

Scientific paper ArODES

Vincent Andrearczyk, Adrien Depeursinge, Henning Müller

Journal of Medical Imaging, 2019, vol. 6, no. 2

Link to the publication

Summary:

Radiomics has shown promising results in several medical studies, yet it suffers from a limited discrimination and informative capability as well as a high variation and correlation with the tomographic scanner types, pixel spacing, acquisition protocol, and reconstruction parameters. We propose and compare two methods to transform quantitative image features in order to improve their stability across varying image acquisition parameters while preserving the texture discrimination abilities. In this way, variations in extracted features are representative of true physiopathological tissue changes in the scanned patients. A first approach is based on a two-layer neural network that can learn a nonlinear standardization transformation of various types of features including handcrafted and deep features. Second, domain adversarial training is explored to increase the invariance of the transformed features to the scanner of origin. The generalization of the proposed approach to unseen textures and unseen scanners is demonstrated by a set of experiments using a publicly available computed tomography texture phantom dataset scanned with various imaging devices and parameters.

PET-based prognostic survival model after radiotherapy for head and neck cancer :

Scientific paper ArODES

responsibility, individual consumption and demand-side policies in European deep decarbonisation pathways

Joël Castelli, Adrien Depeursinge, A Devillers, B. Campillo-Gimenez, Y Dicente, J. O. Prior, E. Chajon, F. Jegoux, C. Sire, O. Acosta, E. Gherga, X. Sun, B. De Bari, R. de Crevoisier

European journal of nuclear medicine and molecular imaging, March 2019, vol. 46, no. 3, pp. 638–649

Link to the publication

Summary:

Purpose The aims of this multicentre retrospective study of locally advanced head and neck cancer (LAHNC) treated with definitive radiotherapy were to (1) identify positron emission tomography (PET)- 18 F-fluorodeoxyglucose ( 18 F-FDG) parameters correlated with overall survival (OS) in a training cohort, (2) compute a prognostic model, and (3) externally validate this model in an independent cohort. Materials and methods A total of 237 consecutive LAHNC patients divided into training ( n = 127) and validation cohorts ( n = 110) were retrospectively analysed. The following PET parameters were analysed: SUV Max , metabolic tumour volume (MTV), total lesion glycolysis (TLG), and SUV Mean for the primary tumour and lymph nodes using a relative SUV Max threshold or an absolute SUV threshold. Cox analyses were performed on OS in the training cohort. The c-index was used to identify the highly prognostic parameters. A prognostic model was subsequently identified, and a nomogram was generated. The model was externally tested in the validation cohort. Results In univariate analysis, the significant PET parameters for the primary tumour included MTV (relative thresholds from 6 to 83% and absolute thresholds from 1.5 to 6.5) and TLG (relative thresholds from 1 to 82% and absolute thresholds from 0.5 to 4.5). For the lymph nodes, the significant parameters included MTVand TLG regardless of the threshold value. In multivariate analysis, tumour site, p16 status, MTV35% of the primary tumour, and MTV44% of the lymph nodes were independent predictors of OS. Based on these four parameters, a prognostic model was identified with a c-index of 0.72. The corresponding nomogram was generated. This prognostic model was externally validated, achieving a c-index of 0.66. Conclusions A prognostic model of OS based on primary tumour and lymph node MTV, tumour site, and p16 status was proposed and validated. The corresponding nomogram may be used to tailor individualized treatment

Revealing tumor habitats from texture heterogeneity analysis for classification of lung cancer malignancy and aggressiveness

Scientific paper ArODES

Dmitri Cherezo, Dmitri Goldgof, Lawrence Hall, Robert Gillies, Matthew Schabath, Henning Müller, Adrien Depeursinge

Scientific reports, Mars 2019, vol. 9, article no. 4500

Link to the publication

Summary:

We propose an approach for characterizing structural heterogeneity of lung cancer nodules using Computed Tomography Texture Analysis (CTTA). Measures of heterogeneity were used to test the hypothesis that heterogeneity can be used as predictor of nodule malignancy and patient survival. To do this, we use the National Lung Screening Trial (NLST) dataset to determine if heterogeneity can represent differences between nodules in lung cancer and nodules in non-lung cancer patients. 253 participants are in the training set and 207 participants in the test set. To discriminate cancerous from non-cancerous nodules at the time of diagnosis, a combination of heterogeneity and radiomic features were evaluated to produce the best area under receiver operating characteristic curve (AUROC) of 0.85 and accuracy 81.64%. Second, we tested the hypothesis that heterogeneity can predict patient survival. We analyzed 40 patients diagnosed with lung adenocarcinoma (20 short-term and 20 long-term survival patients) using a leave-one-out cross validation approach for performance evaluation. A combination of heterogeneity features and radiomic features produce an AUROC of 0.9 and an accuracy of 85% to discriminate long- and short-term survivors.

2018

Rotation invariance and directional sensitivity :

Book chapter ArODES

spherical harmonics versus radiomics features

Adrien Depeursinge, Julien Fageot, Vincent Andrearczyk, John Paul Ward, Michael Unser

Dans Liu, Minigxia, Shi, Yinghuan, Suk, Heung-Il, Machine learning in medical imaging (pp. 107-115). 2018, Cham : Springer

Link to the publication

Summary:

We define and investigate the Local Rotation Invariance (LRI) and Directional Sensitivity (DS) of radiomics features. Most of the classical features cannot combine the two properties, which are antagonist in simple designs. We propose texture operators based on spherical harmonic wavelets (SHW) invariants and show that they are both LRI and DS. An experimental comparison of SHW and popular radiomics operators for classifying 3D textures reveals the importance of combining the two properties for optimal pattern characterization.

(18 F)-FDG PET/CT parameters to predict survival and recurrence in patients with locally advanced cervical cancer treated with chemoradiotherapy

Scientific paper ArODES

N. Scher, J. Castelli, Adrien Depeursinge, J. Bourhis, J. O. Prior, F. G. Herrera, M. Ozsahin

Cancer/Radiothérapie, May 2018, vol. 22, no. 3, pp. 229-235

Link to the publication

Summary:

Purpose To identify predictive (18F)-fluorodeoxyglucose (FDG) positron emission tomography (PET)/computed tomography (CT)-based parameters for locoregional control, disease-free survival and overall survival, by testing different thresholds of metabolic tumor volume and total lesion glycolysis in patients with locally-advanced cervical cancer. Patients and methods Thirty-seven patients treated with standard chemoirradiation underwent a pretreatment (18F)-FDG-PET/CT. Using different thresholds of maximum standardized uptake value, the following PET parameters were computed: maximum standardized uptake value, mean standardized uptake value, metabolic tumor volume and total lesion glycolysis for primary tumor and lymph nodes and a new parameter combining the metabolic tumor volume and the distance between lymph nodes and the primary tumor, namely metabolic node distance. Correlation between PET and clinical parameters with clinical outcome (overall survival, disease-free survival, and locoregional control) was assessed using univariate and multivariate analyses (Cox model). Results In univariate analyses, PET/CT parameters associated with overall survival and disease-free survival were: metabolic tumor volume and total lesion glycolysis of the primary tumor, total lesion glycolysis of lymph nodes and metabolic node distance. The most predictive threshold segmentation for metabolic tumor volume and total lesion glycolysis was 48% of maximum standardized uptake value for the primary tumor and 30% for the lymph nodes. In multivariate Cox analysis, the total lesion glycolysis of primary tumor 48% and metabolic node distance were the two independent risk factors for overall survival (P < 0.01), disease-free survival (P < 0.01) and locoregional control (P = 0.046). Conclusion Total lesion glycolysis of primary tumor and distance between the invaded positive lymph node and the primary tumor seem to have the highest predictive value when compared to classical clinical prognostic parameters and may be useful to identify high risk groups at time of diagnosis and to tailor the therapeutic approach in locally-advanced cervical cancer.

Rotation-covariant tissue analysis for interstitial lung diseases using learned steerable filters :

Scientific paper ArODES

performance evaluation and relevance for diagnostic aid

Ranveer Joyseeree, Henning Müller, Adrien Depeursinge

Computerized medical imaging and graphics, March 2018, vol. 64, pp. 1-11

Link to the publication

Summary:

A novel method to detect and classify several classes of diseased and healthy lung tissue of interstitial lung diseases is presented, as these diseases are hard to diagnose and differentiate. Local organizations of image directions at several scales drive the process of creating discriminative lung tissue texture signatures using spatial and Fourier domain information extracted from the images. The signatures are generated for four diseased tissue classes and healthy tissue, all of which appear in the Interstitial Lung Disease (ILD) database, using a novel one-versus-one approach for learning discriminative filter signatures. A multiclass tissue classification accuracy of 80.31% is observed using Radial Basis Function (RBF) Support Vector Machines (SVMs). The presented method compares well against a variety of state-of-the-art approaches. Another strong feature of our approach is the ability to access the individual class probabilities before a final classification decision is made. This enables an analysis of the causes of misclassification in this paper. We also make the case against total reliance on the accuracy of the ground truth given that the ILD database only contains a single label for a specific region and sometimes more than one pattern can be present, particularly for regions classified as healthy tissue. Measures to address misclassifications in this context are also proposed.

2017

Signature of survival :

Scientific paper ArODES

a 18F-FDG PET based whole-liver radiomic analysis predicts survival after 90Y-TARE for hepatocellular carcinoma

Paul Blanc-Durand, Axel Van Der Gucht, Mario Jreige, Marie Nicod-Lalonde, Marina Silva-Monteiro, John O. Prior, Alban Denys, Adrien Depeursinge, Niklaus Schaefer

Oncotarget, 2017, vol. 9, no. 4, pp. 4549-4558

Link to the publication

Summary:

Methods: Pretreatment 18F-FDG PET of 47 consecutive patients undergoing 90Y-TARE for uHCC (31 resin spheres, 16 glass spheres) were retrospectively analyzed. For each patient, based on PET radiomics signature from whole-liver semi-automatic segmentation, PFS and OS predictive PET-radiomics scores (pPET-RadScores) were obtained using LASSO Cox regression. Using X-tile software, the optimal score to predict PFS (PFS-pPET-RadScore) and OS (OS-pPET-RadScore) served as cutoff to separate high and low-risk patients. Survival curves were estimated using the Kaplan-Meier method. The prognostic value of PFS and OS-pPET-RadScore, Barcelona-Clinic Liver Cancer staging system and serum alpha-fetoprotein level was analyzed to predict PFS and OS in multivariate analysis.

Assessing treatment response in triple-negative breast cancer from quantitative image analysis in perfusion magnetic resonance imaging

Scientific paper ArODES

Imon Banerjee, Sadhika Malladi, Daniela Lee, Adrien Depeursinge, Melinda Telli, Jafi Lipson, Daniel Golden, Daniel L. Rubin

Journal of medical imaging, January 2018, vol. 5, no. 1

Link to the publication

Summary:

Dynamic contrast-enhanced magnetic resonance imaging (DCE-MRI) is sensitive but not specific to determining treatment response in early stage triple-negative breast cancer (TNBC) patients. We propose an efficient computerized technique for assessing treatment response, specifically the residual tumor (RT) status and pathological complete response (pCR), in response to neoadjuvant chemotherapy. The proposed approach is based on Riesz wavelet analysis of pharmacokinetic maps derived from noninvasive DCE-MRI scans, obtained before and after treatment. We compared the performance of Riesz features with the traditional gray level co-occurrence matrices and a comprehensive characterization of the lesion that includes a wide range of quantitative features (e.g., shape and boundary). We investigated a set of predictive models ( ∼96 ) incorporating distinct combinations of quantitative characterizations and statistical models at different time points of the treatment and some area under the receiver operating characteristic curve (AUC) values we reported are above 0.8. The most efficient models are based on first-order statistics and Riesz wavelets, which predicted RT with an AUC value of 0.85 and pCR with an AUC value of 0.83, improving results reported in a previous study by ∼13% . Our findings suggest that Riesz texture analysis of TNBC lesions can be considered a potential framework for optimizing TNBC patient care.

Fundamentals of texture processing for biomedical image analysis :

Book chapter ArODES

a general definition and problem formulation

Adrien Depeursinge, Julien Fageot, Omar Al-Kadi

Biomedical texture analysis : fundamentals, tools and challenges (pp. 7-35). 2017, London : Academic Press

Link to the publication

Summary:

This chapter aims to provide an overview of the foundations of texture processing for biomedical image analysis. Its purpose is to define precisely what biomedical texture is, how is it different from general texture information considered in computer vision, and what is the general problem formulation to translate 2Dand 3Dtextured patterns from biomedical images to visually and biologically relevant measurements. First, a formal definition of biomedical tex-ture information is proposed from both perceptual and mathematical point of views. Second, a general problem formulation for biomedical texture analysis is introduced, considering that any approach can be characterized as a set of local texture operators and regional aggre-gation functions. The operators allow locally isolating desired texture information in terms of spatial scales and directions of a texture image. The type of desirable operator invariances are discussed, and are found to be different from photographic image analysis. Scalar-valued texture measurements are obtained by aggregating operator’s response maps over regions of interest.

Web-based tools for exploring the potential of quantitative imaging biomarkers in radiology :

Book chapter ArODES

intensity and texture analysis on the ePAD platform

Roger Schaer, Yashin Dicente Cid, Emel Alkim, Sheryl John, Daniel L. Rubin, Adrien Depeursinge

Biomedical texture analysis : fundamentals, tools and challenges (pp. 379–410). 2017, [S. l.] : Elsevier

Link to the publication

Summary:

Recent papers in the field of radiomics showed strong evidence that novel image biomarkers based on structural tissue properties have the potential to complement and even surpass invasive and costly biopsy-based molecular assays in certain clinical contexts. To date, very few translations of these research results to clinical practive have been carried out. In addition, a majority of the identified imaging biomarkers are perceived as black boxes by end-users, hindering their acceptance in clinical and research environments. We present a suite of plugins called Quantitative Feature Explore (QFExplore) for the open-access cloud-based ePAD platform enabling the exploration and validation of new imaging biomarkers in a clinical environment. The latter include the extraction, visualization and comparison of intensity- and texture-based quantitative imaging features, regional division of regions of interest to reveal tissue diversity, as well as the construction, use and sharing of user-personalized statistical machine learning models. No software installation is required and the platform can be accessed through any web browser. The relevance of the developed tools is demonstrated in the context of various clinical use-cases. The software is available online.

Biomedical texture operators and aggregation functions :

Book chapter ArODES

a methodological review and user's guide

Adrien Depeursinge, Julien Fageot

Biomedical texture analysis : fundamentals, tools and challenges (pp. 55–94). 2017, London : Academic Press

Link to the publication

Summary:

This chapter reviews most popular texture analysis approaches under novel comparison axes that are specific to biomedical imaging. A concise checklist is proposed as a user guide to assess the relevance of each approach for a particular medical or biological task in hand. We revealed that few approaches are regrouping most of the desirable properties for achieving optimal performance. In particular, moving frames texture representations based on learned steerable operators showed to enable data-specific and rigid-transformation-invariant characterization of local directional patterns, the latter being a fundamental property of biomedical textures. Potential limitations of having recourse to data augmentation and transfer learning for deep convolutional neural networks and dictionary learning approaches to palliate the lack of large annotated training collections in biomedical imaging are mentioned. We conclude by summarizing the strengths and limitations of current approaches, providing insights on key aspects required to build the next generation of biomedical texture analysis approaches.

QuantImage :

Book chapter ArODES

an online tool for high-throughput 3D radiomics feature extraction in PET-CT

Yashin Dicente Cid, Joël Castelli, Roger Schaer, Nathaniel Scher, Anastasia Pomoni, John O. Prior, Adrien Depeursinge

Biomedical texture analysis : fundamentals, tools and challenges (pp. 365-394). 2017, London : Academic Press

Link to the publication

Summary:

Theprocessesofradiomicsconsistofimage-basedpersonalizedtumorphenotypingforpre-cision medicine. They complement slow, costly and invasive molecular analysis of tumoral tissue. Whereastherelevanceofalargevarietyofquantitativeimagingbiomarkershasbeen demonstrated for various cancer types, most studies were based on 2D image analysis of relatively small patient cohorts. In this work, we propose an online tool for automatically ex-tracting 3D state-of-the-art quantitative imaging features from large batches of patients. The developed platform is called QuantImage and can be accessed from any web browser. Its use is straightforward and can be further parameterized for refined analyses. It relies on a robust 3D processing pipeline allowing normalization across patients and imaging protocols. Theusercansimplydrag-and-dropalargezipfilecontainingallimagedataforabatchofpa-tients and the platform returns a spreadsheet with the set of quantitative features extracted for each patient. It is expected to enable high-throughput reproducible research and the validation of radiomics imaging parameters to shape the future of non-invasive personalized medicine.

Multi-scale and multi-directional biomedical texture analysis :

Book chapter ArODES

finding the needle in the haystack

Adrien Depeursinge

Biomedical texture analysis : fundamentals, tools and challenges (pp. 37-62). 2017, London : Academic Press

Link to the publication

Summary:

This chapter clarifies the important aspects of biomedical texture analysis under the general framework introduced in Chapter 1. It was proposed that any approach can be character-ized as the combination of local texture operators and regional aggregation functions. The type of scale and directional information that can or cannot be modeled by categories of texture processing methods is revealed through theoretic analyses and experimental valida-tions. Several key aspects are found to be commonly overlooked in the literature and are highlighted. First, we demonstrate the risk of using regions of interest for aggregation that are regrouping tissue types of different natures. Second, a detailed study of the type of directional information important for biomedical texture characterization suggests that fun-damental properties lie in the local organization of image directions. In addition, it was found that most approaches cannot efficiently characterize the latter, and even fewer can do it with invariance to local rotations. We conclude by deriving novel comparison axes to evalu-ate the relevance of biomedical texture analysis methods in a specific medical or biological applicative context.

Biomedical texture analysis :

Book ArODES

fundamentals, tools and challenges

Adrien Depeursinge, Omar S. Al-Kadi, J.Ross Mitchell

2017, London : Academic Press, 430 p.

Link to the publication

Summary:

Biomedical Texture Analysis: Fundamentals, Applications, Tools and Challenges describes the fundamentals and applications of biomedical texture analysis (BTA) for precision medicine. It defines what biomedical textures (BTs) are and why they require specific image analysis design approaches when compared to more classical computer vision applications. The fundamental properties of BTs are given to highlight key aspects of texture operator design, providing a foundation for biomedical engineers to build the next generation of biomedical texture operators. Examples of novel texture operators are described and their ability to characterize BTs are demonstrated in a variety of applications in radiology and digital histopathology. Recent open-source software frameworks which enable the extraction, exploration and analysis of 2D and 3D texture-based imaging biomarkers are also presented. This book provides a thorough background on texture analysis for graduate students and biomedical engineers from both industry and academia who have basic image processing knowledge. Medical doctors and biologists with no background in image processing will also find available methods and software tools for analyzing textures in medical images.

Metabolic tumor volume and total lesion glycolysis in oropharyngeal cancer treated with definitive radiotherapy :

Scientific paper ArODES

which treshold is the best predictor of local control ?

Joël Castelli, Adrien Depeursinge, Berardino De Bari, Anne Devilliers, Renaud De Crevoisier, Jean Bourhis, John O. Prior

Clinical nuclear medicine, 2017, vol. 42, no. 6, pp. 281-285

Link to the publication

Text- and content-based medical image retrievals in the VISCERAL retrieval benchmark

Book chapter ArODES

Fan Zhang, Yang Song, Weidong Cai, Adrien Depeursinge, Henning Müller

Cloud-based benchmarking of medical image analysis (pp. 237-249). 2017, Cham : Springer

Link to the publication

Summary:

Text- and content-based retrieval are the most widely used approaches for medical image retrieval. They capture the similarity between the images from different perspectives: text-based methods rely on manual textual annotations or captions associated with images; content-based approaches are based on the visual content of the images themselves such as colours and textures. Text-based retrieval can better meet the high-level expectations of humans but is limited by the time-consuming annotations. Content-based retrieval can automatically extract the visual features for high-throughput processing; however, its performance is less favourable than the text-based approaches due to the gap between low-level visual features and high-level human expectations. In this chapter, we present the participation from our joint research team of USYD/HES-SO in the VISCERAL retrieval task. Five different methods are introduced, of which two are based on the anatomy–pathology terms, two are based on the visual image content and the last one is based on the fusion of the aforementioned methods. The comparison results, given the different methods indicated that the text-based methods outperformed the content-based retrieval and the fusion of text and visual contents, generated the best performance overall.

3D solid texture classification using locally-oriented wavelet transforms

Scientific paper ArODES

Yashin Dicente Cid, Henning Müller, Alexandra Platon, Pierre–Alexandre Poletti, Adrien Depeursinge

IEEE transactions on image processing, 2017, vol. 26, no. 4, pp. 1899-1910

Link to the publication

Summary:

Many image acquisition techniques used in biomedical imaging, material analysis, and structural geology are capable of acquiring 3D solid images. Computational analysis of these images is complex but necessary, since it is difficult for humans to visualize and quantify their detailed 3D content. One of the most common methods to analyze 3D data is to characterize the volumetric texture patterns. Texture analysis generally consists of encoding the local organization of image scales and directions, which can be extremely diverse in 3D. Current state-of-the-art techniques face many challenges when working with 3D solid texture, where most approaches are not able to consistently characterize both scale and directional information. 3D Riesz-wavelets can deal with both properties. One key property of Riesz filterbanks is steerability, which can be used to locally align the filters and compare textures with arbitrary (local) orientations. This paper proposes and compares three novel local alignment criteria for higher-order 3D Riesz-wavelet transforms. The estimations of local texture orientations are based on higher-order extensions of regularized structure tensors. An experimental evaluation of the proposed methods for the classification of synthetic 3D solid textures with alterations (such as rotations and noise) demonstrated the importance of local directional information for robust and accurate solid texture recognition. These alignment methods achieved an accuracy of 0.95 in the rotated data, three times more than the unaligned Riesz descriptor that achieved 0.32. The accuracy obtained is better than all other techniques that are published and tested on the same database.

Steerable wavelet machines (SWM) :

Scientific paper ArODES

learning moving frames

Adrien Depeursinge, Zsuzsanna Püspöki, John Paul Ward, Michael Unser

IEEE transactions on image processing, April 2017, vol. 26, issue 4, pp. 1626-1636

Link to the publication

Summary:

We present texture operators encoding class-specific local organizations of image directions (LOIDs) in a rotation-invariant fashion. The LOIDs are key for visual understanding, and are at the origin of the success of the popular approaches, such as local binary patterns (LBPs) and the scale-invariant feature transform (SIFT). Whereas, LBPs and SIFT yield hand-crafted image representations, we propose to learn data-specific representations of the LOIDs in a rotation-invariant fashion. The image operators are based on steerable circular harmonic wavelets (CHWs), offering a rich and yet compact initial representation for characterizing natural textures. The joint location and orientation required to encode the LOIDs is preserved by using moving frames (MFs) texture representations built from locally-steered image gradients that are invariant to rigid motions. In a second step, we use support vector machines to learn a multi-class shaping matrix for the initial CHW representation, yielding data-driven MFs called steerable wavelet machines (SWMs). The SWM forward function is composed of linear operations (i.e., convolution and weighted combinations) interleaved with non-linear steermax operations. We experimentally demonstrate the effectiveness of the proposed operators for classifying natural textures. Our scheme outperforms recent approaches on several test suites of the Outex and the CUReT databases.

A PET-based nomogram for oropharyngeal cancers

Scientific paper ArODES

Joël Castelli, Adrien Depeursinge, V. Ndoh, et al.

European journal of cancer, April 2017, vol. 75, pp. 222-230

Link to the publication

Summary:

Purpose In the context of locally advanced oropharyngeal cancer (LAOC) treated with definitive radiotherapy (RT) (combined with chemotherapy or cetuximab), the aims of this study were: (1) to identify PET-FDG parameters correlated with overall survival (OS) from a first cohort of patients; then (2) to compute a prognostic score; and (3) finally to validate this scoring system in a second independent cohort of patients. Materials and methods A total of 76 consecutive patients (training cohort from Rennes) treated with chemoradiotherapy or RT with cetuximab for LAOC were used to build a predictive model of locoregional control (LRC) and OS based on PET-FDG parameters. After internal calibration and validation of this model, a nomogram and a scoring system were developed and tested in a validation cohort of 46 consecutive patients treated with definitive RT for LAOC in Lausanne. Results In multivariate analysis, the metabolic tumour volume (MTV) of the primary tumour and the lymph nodes were independent predictive factors for LRC and OS. Internal calibration showed a very good adjustment between the predicted OS and the observed OS at 24 months. Using the predictive score, two risk groups were identified (median OS 42 versus 14 months, p < 0.001) and confirmed in the validation cohort from Lausanne (median OS not reached versus 26 months, p = 0.008). Conclusions This is the first report of a PET-based nomogram in oropharyngeal cancer. Interestingly, it appeared stronger than the classical prognostic factors and was validated in independent cohorts markedly diverging in many aspects, which suggest that the observed signal was robust.

2016

A 3-D Riesz-covariance texture model for prediction of nodule recurrence in Lung CT

Scientific paper ArODES

Henning Müller, Adrien Depeursinge, et al.

IEEE Transactions on medical imaging, 2016, vol. 35, issue 12, pp. 2620 - 2630

Link to the publication

Summary:

This paper proposes a novel imaging biomarker of lung cancer relapse from 3–D texture analysis of CT images. Three–dimensional morphological nodular tissue properties are described in terms of 3–D Riesz–wavelets. The responses of the latter are aggregated within nodular regions by means of feature covariances, which leverage rich intra– and inter– variations of the feature space dimensions. When compared to the classical use of the average for feature aggregation, feature covariances preserve spatial co–variations between features. The obtained Riesz–covariance descriptors lie on a manifold governed by Riemannian geometry allowing geodesic measurements and differentiations. The latter property is incorporated both into a kernel for support vector machines (SVM) and a manifold–aware sparse regularized classifier. The effectiveness of the presented models is evaluated on a dataset of 110 patients with non–small cell lung carcinoma (NSCLC) and cancer recurrence information. Disease recurrence within a timeframe of 12 months could be predicted with an accuracy of 81.3–82.7%. The anatomical location of recurrence could be discriminated between local, regional and distant failure with an accuracy of 78.3–93.3%. The obtained results open novel research perspectives by revealing the importance of the nodular regions used to build the predictive models.

Optimized distributed hyperparameter search and simulation for Lung texture classification in CT Usi

Scientific paper ArODES

Roger Schaer, Henning Müller, Adrien Depeursinge

Journal of imaging, 2016, vol. 2, no. 2

Link to the publication

Summary:

Many medical image analysis tasks require complex learning strategies to reach a quality of image-based decision support that is sufficient in clinical practice. The analysis of medical texture in tomographic images, for example of lung tissue, is no exception. Via a learning framework, very good classification accuracy can be obtained, but several parameters need to be optimized. This article describes a practical framework for efficient distributed parameter optimization. The proposed solutions are applicable for many research groups with heterogeneous computing infrastructures and for various machine learning algorithms. These infrastructures can easily be connected via distributed computation frameworks. We use the Hadoop framework to run and distribute both grid and random search strategies for hyperparameter optimization and cross-validations on a cluster of 21 nodes composed of desktop computers and servers. We show that significant speedups of up to 364 compared to a serial execution can be achieved using our in-house Hadoop cluster by distributing the computation and automatically pruning the search space while still identifying the best-performing parameter combinations. To the best of our knowledge, this is the first article presenting practical results in detail for complex data analysis tasks on such a heterogeneous infrastructure together with a linked simulation framework that allows for computing resource planning. The results are directly applicable in many scenarios and allow implementing an efficient and effective strategy for medical (image) data analysis and related learning approaches.

Automated classification of brain tumor type in whole-slide digital pathology images using local representative tiles

Scientific paper ArODES

Jocelyn Barker, Assaf Hoogi, Adrien Depeursinge, Daniel L. Rubin

Medical image analysis, May 2016, vol.30, pp.60–71

Link to the publication

Summary:

Computerized analysis of digital pathology images offers the potential of improving clinical care (e.g. automated diagnosis) and catalyzing research (e.g. discovering disease subtypes). There are two key challenges thwarting computerized analysis of digital pathology images: first, whole slide pathology images are massive, making computerized analysis inefficient, and second, diverse tissue regions in whole slide images that are not directly relevant to the disease may mislead computerized diagnosis algorithms. We propose a method to overcome both of these challenges that utilizes a coarse-to-fine analysis of the localized characteristics in pathology images. An initial surveying stage analyzes the diversity of coarse regions in the whole slide image. This includes extraction of spatially localized features of shape, color and texture from tiled regions covering the slide. Dimensionality reduction of the features assesses the image diversity in the tiled regions and clustering creates representative groups. A second stage provides a detailed analysis of a single representative tile from each group. An Elastic Net classifier produces a diagnostic decision value for each representative tile. A weighted voting scheme aggregates the decision values from these tiles to obtain a diagnosis at the whole slide level. We evaluated our method by automatically classifying 302 brain cancer cases into two possible diagnoses (glioblastoma multiforme (N = 182) versus lower grade glioma (N = 120)) with an accuracy of 93.1 % (p << 0.001). We also evaluated our method in the dataset provided for the 2014 MICCAI Pathology Classification Challenge, in which our method, trained and tested using 5-fold cross validation, produced a classification accuracy of 100% (p << 0.001). Our method showed high stability and robustness to parameter variation, with accuracy varying between 95.5% and 100% when evaluated for a wide range of parameters. Our approach may be useful to automatically differentiate between the two cancer subtypes.

2015

Using MapReduce for large-scale medical image analysis

Scientific paper ArODES

Dimitrios Markonis, Roger Schaer, Ivan Eggel, Henning Müller, Adrien Depeursinge

ArXiv,

Link to the publication

Summary:

The growth of the amount of medical image data produced on a daily basis in modern hospitals forces the adaptation of traditional medical image analysis and indexing approaches towards scalable solutions. The number of images and their dimensionality increased dramatically during the past 20 years. We propose solutions for large-scale medical image analysis based on parallel computing and algorithm optimization. The MapReduce framework is used to speed up and make possible three large-scale medical image processing use-cases: (i) parameter optimization for lung texture segmentation using support vector machines, (ii) content-based medical image indexing, and (iii) three-dimensional directional wavelet analysis for solid texture classification. A cluster of heterogeneous computing nodes was set up in our institution using Hadoop allowing for a maximum of 42 concurrent map tasks. The majority of the machines used are desktop computers that are also used for regular office work. The cluster showed to be minimally invasive and stable. The runtimes of each of the three use-case have been significantly reduced when compared to a sequential execution. Hadoop provides an easy-to-employ framework for data analysis tasks that scales well for many tasks but requires optimization for specific tasks.

Automated classification of usual interstitial pneumonia using regional volumetric texture analysis in high-resolution computed tomography

Scientific paper ArODES

Adrien Depeursinge, Anne C. Chin, Ann N. Leung, Donato Terrone, Michael Bristow, Glenn Rosen, Daniel L. Rubin

Investigative Radiology, avril 2015, vol. 50, no. 4, pp. 261–267

Link to the publication

Summary:

Objectives:We propose a novel computational approach for the automated classification of classic versus atypical usual interstitial pneumonia (UIP). Materials and Methods: Thirty-three patients with UIP were enrolled in this study. They were classified as classic versus atypical UIP by a consensus of 2 thoracic radiologists with more than 15 years of experience using the American Thoracic Society evidence–based guidelines for computed tomography diagnosis of UIP. Two cardiothoracic fellows with 1 year of subspecialty training provided independent readings. The system is based on regional characterization of the morphological tissue properties of lung using volumetric texture analysis of multiple-detector computed tomography images. A simple digital atlas with 36 lung subregions is used to locate texture properties, from which the responses of multidirectional Riesz wavelets are obtained. Machine learning is used to aggregate and to map the regional texture attributes to a simple score that can be used to stratify patients with UIP into classic and atypical subtypes. Results: We compared the predictions on the basis of regional volumetric texture analysis with the ground truth established by expert consensus. The area under the receiver operating characteristic curve of the proposed score was estimated to be 0.81 using a leave-one-patient-out cross-validation, with high specificity for classic UIP. The performance of our automated method was found to be similar to that of the 2 fellows and to the agreement between experienced chest radiologists reported in the literature. However, the errors of our method and the fellows occurred on different cases, which suggests that combining human and computerized evaluations may be synergistic. Conclusions: Our results are encouraging and suggest that an automated system may be useful in routine clinical practice as a diagnostic aid for identifying patients with complex lung disease such as classic UIP, obviating the need for invasive surgical lung biopsy and its associated risks.

Predicting adenocarcinoma recurrence using computational texture models of nodule components in lung CT

Scientific paper ArODES

Adrien Depeursinge, Masahiro Yanagawa, Ann N. Leung, Daniel L. Rubin

Medical Physics, April 2015, vol. 42, no. 4, pp. 2054-2063

Link to the publication

Summary:

Purpose: To investigate the importance of presurgical computed tomography (CT) intensity and texture information from ground-glass opacities (GGO) and solid nodule components for the prediction of adenocarcinoma recurrence. Methods: For this study, 101 patients with surgically resected stage I adenocarcinoma were selected. During the follow-up period, 17 patients had disease recurrence with six associated cancer-related deaths. GGO and solid tumor components were delineated on presurgical CT scans by a radiologist. Computational texture models of GGO and solid regions were built using linear combinations of steerable Riesz wavelets learned with linear support vector machines (SVMs). Unlike other traditional texture attributes, the proposed texture models are designed to encode local image scales and directions that are specific to GGO and solid tissue. The responses of the locally steered models were used as texture attributes and compared to the responses of unaligned Riesz wavelets. The texture attributes were combined with CT intensities to predict tumor recurrence and patient hazard according to disease-free survival (DFS) time. Two families of predictive models were compared: LASSO and SVMs, and their survival counterparts: Cox-LASSO and survival SVMs. Results: The best-performing predictive model of patient hazard was associated with a concordance index (C-index) of 0.81±0.02 and was based on the combination of the steered models and CT intensities with survival SVMs. The same feature group and the LASSO model yielded the highest area under the receiver operating characteristic curve (AUC) of 0.8±0.01 for predicting tumor recurrence, although no statistically significant difference was found when compared to using intensity features solely. For all models, the performance was found to be significantly higher when image attributes were based on the solid components solely versus using the entire tumors (p < 3.08×10−5). Conclusions: This study constitutes a novel perspective on how to interpret imaging information from CT examinations by suggesting that most of the information related to adenocarcinoma aggressiveness is related to the intensity and morphological properties of solid components of the tumor. The prediction of adenocarcinoma relapse was found to have low specificity but very high sensitivity. Our results could be useful in clinical practice to identify patients for which no recurrence is expected with a very high confidence using a presurgical CT scan only. It also provided an accurate estimation of the risk of recurrence after a given duration t from surgical resection (i.e., C-index = 0.81±0.02).

2024

FLAIR vs MPRAGE contribution to white matter lesion automatic segmentation in MS using localized saliency maps

Conference ArODES

Frederico Spagnolo, Roger Schaer, Vincent Andrearczyk, Nataliia Molchanova, Mara Graziani, Henning Müller, Meritxell Bach Cuadra, Cristina Granziera, Adrien Depeursinge

Proceedings of BIAS 2023

Link to the conference

2023

Deep learning uncertainty quantification of cortical lesions in MP2RAGE for missed lesions discovery

Conference ArODES

Nataliia Molchanova, Alessandro Cagol, Mario Ocampo-Pineda, Xinjie Chen, Matthias Weigel, Adrien Depeursinge, Henning Müller, Mara Graziani, Cristina Granziera, Meritxell Bach Cuadra

Multiple Sclerosis Journal

Link to the conference

Identification of paramagnetic rim lesions using conventional MRI :

Conference ArODES

a deep learning approach

Frederico Spagnolo, Adrien Depeursinge, Vincent Andrearczyk, Pascal Benkert, Pascal Benkert, Stefanie Müller, Lutz Achtnichts, Jochen Vehoff, Giulio Disanto, Olivier Findling, Andrew Chan, Anke Salmen, Caroline Pot, Claire Bridel, Chiara Zecca, Tobias Johannes Derfuss, Johanna M. Lieb, Luca Remonda, Franca Wagner, Maria Isabel Vargas Gomez, Nataliia Molchanova

Multiple Sclerosis Journal

Link to the conference

Evaluation of PAIRE PET/CT segmentation software as fully-automated cancerous lesion contouring tool in image annotation workflows for research studies

Conference ArODES

Daniel Abler, Mario Jreige, Olivier Michielin, Niklaus Schaefer, John O. Prior, Adrien Depeursinge, Michel A. Cuendet, Paul Blanc-Durand

EANM’23 Abstract Book Congress

Link to the conference

Why is the winner the best?

Conference ArODES

M. Eisenmann, A. Reinke, V. Weru, M. D. Tizabi, F. Isensee, T. J. Adler, S. Ali, Andrearczyk. Vincent, M. Aubreville, U. Baid, S. Bakas, N. Balu, S. Bano, J. Bernal, S. Bodenstedt, A. Casella, V. Cheplygina, M. Daum, M. De Bruijne, Adrien Depeursinge, R. Dorent, J. Egger, D. G. Ellis, S. Engelhardt, M. Ganz, N. Ghatwary, G. Girard, P. Godau, A. Gupta, L. Hansen, K. Harada, M. Heinrich, N. Heller, A. Hering, A. Huaulmé, P. Jannin, A. E. Kavur, O. Kodym, M. Kozubek, J. Li, H. Li, J. Ma, C. Martín-Isla, B. Menze, A. Noble, Valentin Oreiller, N. Padoy, S. Pati, K. Payette, T. Rädsch, J. Rafael-Patiño, V. Singh Bawa, S. Speidel, C. H. Sudre, K. Van Wijnen, M. Wagner, D. Wei, A. Yamlahi, M. H. Yap, C. Yuan, M. Zenk, A. Zia, D. Zimmerer, D. Aydogan, B. Bhattarai, L. Bloch, R. Brüngel, J. Cho, C. Choi, Q. Dou, I. Ezhov, C. M. Friedrich, C. Fuller, R. R. Gaire, A. Galdran, Á. García Faura, M. Grammatikopoulou, S. Hong, M. Jahanifar, I. Jang, A. Kadkhodamohammadi, I. Kang, F. Kofler, S. Kondo, H. Kuijf, M. Li, M. Luu, T. Martincic, P. Morais, M. A. Naser, B. Oliveira, D. Owen, S. Pang, J. Park, S. Park, S. Plotka, E. Puybareau, N. Rajpoot, K. Ryu, N. Saeed, A. Shephard, P. Shi, D. Štepec, R. Subedi, G. Tochon, H. R. Torres, H. Urien, J. L. Vilaça, K. A. Wahid, H. Wang, J. Wang, L. Wang, X. Wang, B. Wiestler, Marek Wodzinski

2023 IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR)

Link to the conference

Summary:

International benchmarking competitions have become fundamental for the comparative performance assessment of image analysis methods. However, little attention has been given to investigating what can be learnt from these competitions. Do they really generate scientific progress? What are common and successful participation strategies? What makes a solution superior to a competing method? To address this gap in the literature, we performed a multicenter study with all 80 competitions that were conducted in the scope of IEEE ISBI 2021 and MICCAI 2021. Statistical analyses performed based on comprehensive descriptions of the submitted algorithms linked to their rank as well as the underlying participation strategies revealed common characteristics of winning solutions. These typically include the use of multi-task learning (63%) and/or multi-stage pipelines (61%), and a focus on augmentation (100%), image preprocessing (97%), data curation (79%), and post-processing (66%). The “typical” lead of a winning team is a computer scientist with a doctoral degree, five years of experience in biomedical image analysis, and four years of experience in deep learning. Two core general development strategies stood out for highly-ranked teams: the reflection of the metrics in the method design and the focus on analyzing and handling failure cases. According to the organizers, 43% of the winning algorithms exceeded the state of the art but only 11% completely solved the respective domain problem. The insights of our study could help researchers (1) improve algorithm development strategies when approaching new problems, and (2) focus on open research questions revealed by this work.

Guide for radiologists and nuclear medicine physicians for a standardized radiomics analysis

Conference ArODES

Camille Noirot, Mario Jreige, Adrien Depeursinge, John O. Prior, Clarisse Dromain

Online Abstract Book of the Swiss Congress of Radiology 2023

Link to the conference

Rethinking the role of AI with physicians in oncology :

Conference ArODES

revealing perspectives from clinical and research workflows

Himanshu Verma, Jakub Mlynar, Roger Schaer, Julien Reichenbach, Mario Jreige, John Prior, Florian Evéquoz, Adrien Depeursinge

CHI '23: Proceedings of the 2023 CHI : Conference on Human Factors in Computing Systems

Link to the conference

Summary:

Significant and rapid advancements in cancer research have been attributed to Artificial Intelligence (AI). However, AI’s role and impact on the clinical side has been limited. This discrepancy manifests due to the overlooked, yet profound, differences in the clinical and research practices in oncology. Our contribution seeks to scrutinize physicians’ engagement with AI by interviewing 7 medical-imaging experts and disentangle its future alignment across the clinical and research workflows, diverging from the existing “one-size-fits-all” paradigm within Human-Centered AI discourses. Our analysis revealed that physicians’ trust in AI is less dependent on their general acceptance of AI, but more on their contestable experiences with AI. Contestability, in clinical workflows, underpins the need for personal supervision of AI outcomes and processes, i.e., clinician-in-the-loop. Finally, we discuss tensions in the desired attributes of AI, such as explainability and control, contextualizing them within the divergent intentionality and scope of clinical and research workflows.

Overview of the HECKTOR Challenge at MICCAI 2022 :

Conference ArODES

automatic head and neck tumor segmentation and outcome prediction in PET/CT

Vincent Andrearczyk, Oreillerm Valentin, Moamen Abobakr, Azadeh Akhavanallaf, Panagiotis Balermpas, Sarah Boughdad, Leo Capriotti, Joel Castelli, Catherine Cheze Le Rest, Pierre Decazes, Ricardo Correia, Dina El-Habashy, Hesham Elhalawani, Clifton D. Fuller, Mario Jreige, Yomna Khamis, Agustina La Greca, Abdallah Mohamed, Mohamed Naser, John O. Prior, Su Ruan, Stephanie Tanadini-Lang, Olena Tankyevych, Yazdan Salimi, Martin Vallières, Pierre Vera, Dimitri Visvikis, Kareem Wahid, Habib Zaidi, Mathieu Hatt, Adrien Depeursinge

Head and neck tumor segmentation and outcome prediction

Link to the conference

Summary:

This paper presents an overview of the third edition of the HEad and neCK TumOR segmentation and outcome prediction (HECKTOR) challenge, organized as a satellite event of the 25th International Conference on Medical Image Computing and Computer Assisted Intervention (MICCAI) 2022. The challenge comprises two tasks related to the automatic analysis of FDG-PET/CT images for patients with Head and Neck cancer (H &N), focusing on the oropharynx region. Task 1 is the fully automatic segmentation of H &N primary Gross Tumor Volume (GTVp) and metastatic lymph nodes (GTVn) from FDG-PET/CT images. Task 2 is the fully automatic prediction of Recurrence-Free Survival (RFS) from the same FDG-PET/CT and clinical data. The data were collected from nine centers for a total of 883 cases consisting of FDG-PET/CT images and clinical information, split into 524 training and 359 test cases. The best methods obtained an aggregated Dice Similarity Coefficient (DSCagg) of 0.788 in Task 1, and a Concordance index (C-index) of 0.682 in Task 2.

Preface of Head and neck tumor segmentation and outcome prediction

Conference ArODES

Vincent Andrearczyk, Valentin Oreiller, Matthieu Hatt, Adrien Depeursinge

Head and neck tumor segmentation and outcome prediction : third Challenge, HECKTOR 2022 held in conjunction with MICCAI 2022 Singapore, september 22, 2022 proceedings

Link to the conference

2022

HEad and neCK TumOR segmentation and outcome prediction :

Conference ArODES

the HECKTOR challenge

Vincent Andrearczyk, Adrien Depeursinge

European Congress of Radiology (ECR 2022) : book of abstracts

Link to the conference

Segmentation and classification of head and neck nodal metastases and primary tumors in PET/CT

Conference ArODES

Vincent Andrearczyk, Valentin Oreiller, Mario Jreige, Joël Castelli, John O. Prior, Adrien Depeursinge

Proceedings of the 44th Annual International Conference of the IEEE Engineering in Medicine & Biology Society (EMBC)

Link to the conference

Summary:

The prediction of cancer characteristics, treatment planning and patient outcome from medical images generally requires tumor delineation. In Head and Neck cancer (H&N), the automatic segmentation and differentiation of primary Gross Tumor Volumes (GTVt) and malignant lymph nodes (GTVn) is a necessary step for large-scale radiomics studies to predict patient outcome such as Progression Free Survival (PFS). Detecting malignant lymph nodes is also a crucial step for Tumor-Node-Metastases (TNM) staging and to support the decision to resect the nodes. In turn, automatic TNM staging and patient outcome prediction can greatly benefit patient care by helping clinicians to find the best personalized treatment. We propose the first model to automatically individually segment GTVt and GTVn in PET/CT images. A bi-modal 3D U-Net model is trained for multi-class and multi-components segmentation on the multi-centric HECKTOR 2020 dataset containing 254 cases. The dataset has been specifically re-annotated by experts to obtain ground truth GTVn contours. The results show promising segmentation performance for the automation of radiomics pipelines and their validation on large-scale studies for which manual annotations are not available. An average test Dice Similarity Coefficients (DSC) of 0.717 is obtained for the segmentation of GTVt. The GTVn segmentation is evaluated with an aggregated DSC to account for the cases without GTVn, which is estimated at 0.729 on the test set.

Robust multi-organ nucleus segmentation using a locally rotation invariant bispectral U-Net

Conference ArODES

Valentin Oreiller, Julien Fagot, Vincent Andrearczyk, John O. Prior, Adrien Depeursinge

Proceedings of Medical Imaging with Deep Learning

Link to the conference

Summary:

Locally Rotation Invariant (LRI) operators have shown great potential to robustly identify biomedical textures where discriminative patterns appear at random positions and orientations. We build LRI operators through the local projection of the image on circular harmonics followed by the computation of the bispectrum, which is LRI by design. This formulation allows to avoid the discretization of the orientations and does not require any criterion to locally align the descriptors. This operator is used in a convolutional layer resulting in LRI Convolutional Neural Networks (LRI CNN). To evaluate the relevance of this approach, we used it to segment cellular nuclei in histopathological images. We compared the proposed bispectral LRI layer against a standard convolutional layer in a U-Net architecture. While they performed equally in terms of F-score, the LRI CNN provided more robust segmentation with respect to orientation, even when rotational data augmentation was used. This robustness is essential when the relevant pattern may vary in orientation, which is often the case in medical images.

QuantImage v2 :

Conference ArODES

a clinician-in-the-loop cloud platform for radiomics research

Roger Schaer, Valentin Oreiller, Daniel Abler, Himanshu Verma, Julien Reichenbach, Florian Evéquoz, Mario Jreige, John O. Prior, Adrien Depeursinge

ECR 2022 Book of Abstracts

Link to the conference

Comparison of MR preprocessing strategies and sequences for radiomics-Based MGMT prediction

Conference ArODES

Daniel Abler, Vincent Andrearczyk, Valentin Oreiller, Javier Barranco Garcia, Diem Vuong, Stéphanie Tanadini-Lang, Matthias Guckenberger, Mauricio Reyes, Adrien Depeursinge

Proceedings of International MICCAI BrainLes Workshop, 2022

Link to the conference

Summary:

Hypermethylation of the O6-methylguanine-DNA-methyltransferase (MGMT) promoter in glioblastoma (GBM) is a predictive biomarker associated with improved treatment outcome. In clinical practice, MGMT methylation status is determined by biopsy or after surgical removal of the tumor. This study aims to investigate the feasibility of non-invasive medical imaging based “radio-genomic” surrogate markers of MGMT methylation status. The imaging dataset of the RSNA-ASNR-MICCAI Brain Tumor Segmentation (BraTS) challenge allows exploring radiomics strategies for MGMT prediction in a large and very heterogeneous dataset that represents a variety of real-world imaging conditions including different imaging protocols and devices. To characterize and optimize MGMT prediction strategies under these conditions, we examined different image preprocessing approaches and their effect on the average prediction performance of simple radiomics models. We found features derived from FLAIR images to be most informative for MGMT prediction, particularly if aggregated over the entire (enhancing and non-enhancing) tumor with or without inclusion of the edema. Our results also indicate that the imaging characteristics of the tumor region can distort MR-bias-field correction in a way that negatively affects the prediction performance of the derived models.

Impact of deep learning segmentation methods on the robustness of MR glioblastoma radiomics

Conference ArODES

Diem Vuong, Katrin Daetwyler, Marta Bogowicz, Piotr Radojewski, Raphael Meier, Mauricio Reyes, Natalia Saltybaeva, Adrien Depeursinge, Michael Bach, Marco Piccirelli, Matthias Guckenberger, Stéphanie Tanadini-Lang, Roland Wies

Radiotherapy Oncology

Link to the conference

Deep learning classifier for MGMT promoter methylation status in glioblastoma cancer

Conference ArODES

Javier Barranco Garcia, Daniel Abler, Mauricio Reyes, Diem Voung, Matthias Guckenberger, Stéphanie Tanadini-Lang, Adrien Depeursinge

Radiotherapy Oncology

Link to the conference

Overview of the HECKTOR challenge at MICCAI 2021 :

Conference ArODES

automatic head and neck tumor segmentation and outcome prediction in PET/CT images

Vincent Andrearczyk, Valentin Oreiller, Sarah Boughdad, Catherine Chez Le Rest, Hesham Elhalawani, Mario Jreige, John O. Prior, Martin Vallières, Dimitris Visvikis, Mathieu Hatt, Adrien Depeursinge

Head and Neck Tumor Segmentation and Outcome Prediction : second challenge, HECKTOR 2021, held in conjunction with MICCAI 2021, Strasbourg, France, September 27, 2021, Proceedings

Link to the conference

Summary:

This paper presents an overview of the second edition of the HEad and neCK TumOR (HECKTOR) challenge, organized as a satellite event of the 24th International Conference on Medical Image Computing and Computer Assisted Intervention (MICCAI) 2021. The challenge is composed of three tasks related to the automatic analysis of PET/CT images for patients with Head and Neck cancer (HN), focusing on the oropharynx region. Task 1 is the automatic segmentation of HN primary Gross Tumor Volume (GTVt) in FDG-PET/CT images. Task 2 is the automatic prediction of Progression Free Survival (PFS) from the same FDG-PET/CT. Finally, Task 3 is the same as Task 2 with ground truth GTVt annotations provided to the participants. The data were collected from six centers for a total of 325 images, split into 224 training and 101 testing cases. The interest in the challenge was highlighted by the important participation with 103 registered teams and 448 result submissions. The best methods obtained a Dice Similarity Coefficient (DSC) of 0.7591 in the first task, and a Concordance index (C-index) of 0.7196 and 0.6978 in Tasks 2 and 3, respectively. In all tasks, simplicity of the approach was found to be key to ensure generalization performance. The comparison of the PFS prediction performance in Tasks 2 and 3 suggests that providing the GTVt contour was not crucial to achieve best results, which indicates that fully automatic methods can be used. This potentially obviates the need for GTVt contouring, opening avenues for reproducible and large scale radiomics studies including thousands potential subjects.

2021

Multi-task deep segmentation and radiomics for automatic prognosis in head and neck cancer

Conference ArODES

Vincent Andrearczyk, Pierre Fontaine, Valentin Oreiller, Joel Castelli, Mario Jreige, John O. Prior, Adrien Depeursinge

Predictive Intelligence in Medicine : 4th International Workshop, PRIME 2021, Held in Conjunction with MICCAI 2021, Strasbourg, France, October 1, 2021, Proceedings

Link to the conference

Summary:

We propose a novel method for the prediction of patient prognosis with Head and Neck cancer (H&N) from FDG-PET/CT images. In particular, we aim at automatically predicting Disease-Free Survival (DFS) for patients treated with radiotherapy or both radiotherapy and chemotherapy. We design a multi-task deep UNet to learn both the segmentation of the primary Gross Tumor Volume (GTVt) and the outcome of the patient from PET and CT images. The motivation for this approach lies in the complementarity of the two tasks and the shared visual features relevant to both tasks. A multi-modal (PET and CT) 3D UNet is trained with a combination of survival and Dice losses to jointly learn the two tasks. The model is evaluated on the HECKTOR 2020 dataset consisting of 239 H&N patients with PET, CT, GTVt contours and DFS data (five centers). The results are compared with a standard Cox PET/CT radiomics model. The proposed multi-task CNN reaches a C-index of 0.723, outperforming both the deep radiomics model without segmentation (C-index of 0.650) and the standard radiomics model (C-index of 0.695). Besides the improved performance in outcome prediction, the main advantage of the proposed multi-task approach is that it can predict patient prognosis without a manual delineation of the GTVt, a tedious and time-consuming process that hinders the validation of large-scale radiomics studies. The code will be shared for reproducibility on our GitHub repository.

Fully automatic head and neck cancer prognosis prediction in PET/CT

Conference ArODES

Pierre Fontaine, Vincent Andrearczyk, Valentin Oreiller, Joël Castelli, Mario Jreige, John O. Prior, Adrien Depeursinge

Multimodal Learning for Clinical Decision Support : 11th International Workshop, ML-CDS 2021, Held in Conjunction with MICCAI 2021, Strasbourg, France, October 1, 2021, Proceedings

Link to the conference

Summary:

Several recent PET/CT radiomics studies have shown promising results for the prediction of patient outcomes in Head and Neck (H&N) cancer. These studies, however, are most often conducted on relatively small cohorts (up to 300 patients) and using manually delineated tumors. Recently, deep learning reached high performance in the automatic segmentation of H&N primary tumors in PET/CT. The automatic segmentation could be used to validate these studies on larger-scale cohorts while obviating the burden of manual delineation. We propose a complete PET/CT processing pipeline gathering the automatic segmentation of primary tumors and prognosis prediction of patients with H&N cancer treated with radiotherapy and chemotherapy. Automatic contours of the primary Gross Tumor Volume (GTVt) are obtained from a 3D UNet. A radiomics pipeline that automatically predicts the patient outcome (Disease Free Survival, DFS) is compared when using either the automatically or the manually annotated contours. In addition, we extract deep features from the bottleneck layers of the 3D UNet to compare them with standard radiomics features (first- and second-order as well as shape features) and to test the performance gain when added to them. The models are evaluated on the HECKTOR 2020 dataset consisting of 239 H&N patients with PET, CT, GTVt contours and DFS data available (five centers). Regarding the results, using Hand-Crafted (HC) radiomics features extracted from manual GTVt achieved the best performance and is associated with an average Concordance (C) index of 0.672. The fully automatic pipeline (including deep and HC features from automatic GTVt) achieved an average C index of 0.626, which is lower but relatively close to using manual GTVt (p-value = 0.20). This suggests that large-scale studies could be conducted using a fully automatic pipeline to further validate the current state of the art H&N radiomics. The code will be shared publicly for reproducibility.

Assessment of the stability and discriminative power of radiomics features in liver lesions using an anthropomorphic 3D-printed CT phantom

Conference ArODES

Oscar Alfonso Jiménez del Toro, C. Aberle, M. Bach, Roger Schaer, M. Obmann, K. Flouris, E. Konukoglu, B. Stieltjes, Henning Müller, Adrien Depeursinge

Proceedings of the SCR'21 (Swiss Congress of Radiology 2021)

Link to the conference

QuantImage v2 :

Conference ArODES

an open-source and web-based integrated platform for clinical radiomics research

Roger Schaer, Valentin Oreiller, O. Aidonopoulos, John O. Prior, Adrien Depeursinge

Proceedings of the SCR'21 (Swiss Congress of Radiology 2021)

Link to the conference

Revealing the most suitable CT radiomics features for retrospective studies with heterogeneous datasets

Conference ArODES

Oscar Alfonso Jiménez del Toro, C. Aberle, M. Bach, Roger Schaer, M. Obmann, K. Flouris, E. Konukoglu, B. Stieltjes, Henning Müller, Adrien Depeursinge

Proceedings of the 2021 European Congress of Radiology (ECR 2021)

Link to the conference

Preface to "Head and neck tumor segmentation"

Conference ArODES

Vincent Andrearczyk, Valentin Oreiller, Adrien Depeursinge

Head and Neck Tumor Segmentation : First Challenge, HECKTOR 2020, Held in Conjunction with MICCAI 2020, Lima, Peru, October 4, 2020, Proceedings

Link to the conference

Head and neck tumor segmentation :

Conference ArODES

first challenge, HECKTOR 2020, held in conjunction with MICCAI 2020, Lima, Peru, October 4, 2020, Proceedings

Vincent Andrearczyk, Valentin Oreiller, Adrien Depeursinge

HECKTOR: 3D Head and Neck Tumor Segmentation in PET/CT Challenge

Link to the conference

Summary:

This book constitutes the First 3D Head and Neck Tumor Segmentation in PET/CT Challenge, HECKTOR 2020, which was held in conjunction with the 23rd International Conference on Medical Image Computing and Computer-Assisted Intervention, MICCAI 2020, in Lima, Peru, in October 2020. The challenge took place virtually due to the COVID-19 pandemic. The 2 full and 8 short papers presented together with an overview paper in this volume were carefully reviewed and selected form numerous submissions. This challenge aims to evaluate and compare the current state-of-the-art methods for automatic head and neck tumor segmentation. In the context of this challenge, a dataset of 204 delineated PET/CT images was made available for training as well as 53 PET/CT images for testing. Various deep learning methods were developed by the participants with excellent results.

2020

Interpretable CNN pruning for preserving scale-covariant features in medical imaging

Conference ArODES

Mara Graziani, Thomas Lompech, Henning Müller, Adrien Depeursinge, Vincent Andrearczyk

Proceedings of Workshop on Interpretability of Machine Intelligence in Medical Image Computing at MICCAI 2020

Link to the conference

Summary:

Image scale carries crucial information in medical imaging, e.g. the size and spatial frequency of local structures, lesions, tumors and cell nuclei. With feature transfer being a common practice, scale-invariant features implicitly learned from pretraining on ImageNet tend to be preferred over scale-covariant features. The pruning strategy in this paper proposes a way to maintain scale covariance in the transferred features. Deep learning interpretability is used to analyze the layer-wise encoding of scale information for popular architectures such as InceptionV3 and ResNet50. Interestingly, the covariance of scale peaks at central layers and decreases close to softmax. Motivated by these results, our pruning strategy removes the layers where invariance to scale is learned. The pruning operation leads to marked improvements in the regression of both nuclei areas and magnification levels of histopathology images. These are relevant applications to enlarge the existing medical datasets with open-access images as those of PubMed Central. All experiments are performed on publicly available data and the code is shared on GitHub.

Overview of the HECKTOR challenge at MICCAI 2020 :

Conference ArODES

automatic head and neck tumor segmentation in PET/CT

Vincent Andrearczyk, Valentin Oreiller, Mario Jreige, Martin Vallières, Joël Castelli, Hesham Elhalawani, Sarah Boughdad, John O. Prior, Adrien Depeursinge

Head and Neck Tumor Segmentation : First Challenge, HECKTOR 2020, Held in Conjunction with MICCAI 2020, Lima, Peru, October 4, 2020, Proceedings

Link to the conference

Summary:

This paper presents an overview of the first HEad and neCK TumOR (HECKTOR) challenge, organized as a satellite event of the 23rd International Conference on Medical Image Computing and Computer Assisted Intervention (MICCAI) 2020. The task of the challenge is the automatic segmentation of head and neck primary Gross Tumor Volume in FDG-PET/CT images, focusing on the oropharynx region. The data were collected from five centers for a total of 254 images, split into 201 training and 53 testing cases. The interest in the task was shown by the important participation with 64 teams registered and 18 team submissions. The best method obtained a Dice Similarity Coefficient (DSC) of 0.7591, showing a large improvement over our proposed baseline method with a DSC of 0.6610 as well as inter-observer DSC agreement reported in the literature (0.69).

Consistency of scale equivariance in internal representations of CNNs

Conference ArODES

Vincent Andrearczyk, Mara Graziani, Henning Müller, Adrien Depeursinge

Proceedings of the Irish Machine Vision and Image Processing Conference (IMVIP) 2020

Link to the conference

Summary:

Despite the approximate invariance to scale learned in deep Convolutional Neural Networks (CNNs) trained on natural images, intermediate layers have been shown to contain information of scale while the invariance is only obtained in the final layers. In this paper, we experimentally analyze how this scale information is encoded in the hidden layers. Linear regression of scale is used to (i) evaluate whether scale information can be encoded, at a given layer, by individual response maps or a combination of many of them is necessary; (ii) evaluate whether the encoding of scale is shared among classes. If we can find a direction representative of scale variations in the hidden space, is this consistent across the data manifold? Or is it rather encoded locally within class-specific neighborhoods? We observe that scale information is encoded as a combination of a few response maps (around 3%) and that the encoding is relatively consistent across classes, with some amount of class-specific encoding.

Oropharynx detection in PET-CT for tumor segmentation

Conference ArODES

Vincent Andrearczyk, Valentin Oreiller, Adrien Depeursinge

Proceedings of the 2020 Irish Machine Vision and Image Processing Conference (IMVIP 2020)

Link to the conference

Summary:

We propose an automatic detection of the oropharyngeal area in PET-CT images. This detection can be used to preprocess images for efficient segmentation of Head and Neck (H&N) tumors in the cropped regions by a Convolutional Neural Network (CNN) for treatment planning and large-scale radiomics studies (e.g. prognosis prediction). The developed method is based on simple image processing steps to segment the brain on the PET image and retrieve a fixed size bounding box of the extended oropharyngeal region. We evaluate the results by measuring whether the primary Gross Tumor Volume (GTV) is fully contained in the bounding box. 194 out of 201 regions (96.5%) are correctly detected. The code is available on our GitHub repository.

Comparison of feature selection in radiomics for the prediction of overall survival after radiotherapy for hepatocellular carcinoma

Conference ArODES

Pierre Fontaine, François-Georges Riet, Joël Castelli, Khemara Gnep, Adrien Depeursinge, Oscar Acosta

Proceedings of the 42nd Annual International Conference of the IEEE Engineering in Medicine & Biology Society (EMBC)

Link to the conference

Summary:

Hepatocellular carcinoma (HCC) is the sixth more frequent cancer worldwide. This type of cancer has a poor overall survival rate mainly due to underlying cirrhosis and risk of recurrence outside the treated lesion. Quantitative imaging within a radiomics workflow may help assessing the probability of survival and potentially may allow tailoring personalized treatments. In radiomics a large amount of features can be extracted, which may be correlated across a population and very often can be surrogates of the same physiopathology. This issues are more pronounced and difficult to tackle with imbalanced data. Feature selection strategies are therefore required to extract the most informative with the increased predictive capabilities. In this paper, we compared different unsupervised and supervised strategies for feature selection in presence of imbalanced data and optimize them within a machine learning framework. Multi-parametric Magnetic Resonance Images from 81 individuals (19 deceased) treated with stereotactic body radiation therapy (SBRT) for inoperable (HCC) were analyzed. Pre-selection of a reduced set of features based on Affinity Propagation clustering (non supervised) achieved a significant improvement in AUC compared to other approaches with and without feature pre-selection. By including the synthetic minority over-sampling technique (SMOTE) for imbalanced data and Random Forest classification this workflow emerges as an appealing feature selection strategy for survival prediction within radiomics studies.

Automatic segmentation of head and neck tumors and nodal metastases in PET-CT scans

Conference ArODES

Vincent Andrearczyk, Valentin Oreiller, Martin Vallières, Joel Castelli, Hesham Elhalawani, Mario Jreige, Sarah Boughdad, John O. Prior, Adrien Depeursinge

Proceedings of International conference on medical imaging with deep learning

Link to the conference

Summary:

Radiomics, the prediction of disease characteristics using quantitative image biomarkers from medical images, relies on expensive manual annotations of Regions of Interest (ROI) to focus the analysis. In this paper, we propose an automatic segmentation of Head and Neck (H&N) tumors and nodal metastases from FDG-PET and CT images. A fully-convolutional network (2D and 3D V-Net) is trained on PET-CT images using ground truth ROIs that were manually delineated by radiation oncologists for 202 patients. The results show the complementarity of the two modalities with a statistically significant improvement from 48.7% and 58.2% Dice Score Coefficients (DSC) with CT- and PET-only segmentation respectively, to 60.6% with a bimodal late fusion approach. We also note that, on this task, a 2D implementation slightly outperforms a similar 3D design (60.6% vs 59.7% for the best results respectively). The data is publicly available and the code will be shared on our GitHub repository.

An exploration of uncertainty information for segmentation quality assessment

Conference ArODES

Katharina Hoebel, Vincent Andrearczyk, Andrew Beers, Jay Patel, Ken Chang, Adrien Depeursinge, Henning Müller, Kalpathy-Cramer. Jayashree

Proceedings of Medical Imaging 2020: Image Processing

Link to the conference

Summary:

Including uncertainty information in the assessment of a segmentation of pathologic structures on medical images, offers the potential to increase trust into deep learning algorithms for the analysis of medical imaging. Here, we examine options to extract uncertainty information from deep learning segmentation models and the influence of the choice of cost functions on these uncertainty measures. To this end we train conventional UNets without dropout, deep UNet ensembles, and Monte-Carlo (MC) dropout UNets to segment lung nodules on low dose CT using either soft Dice or weighted categorical cross-entropy (wcc) as loss functions. We extract voxel-wise uncertainty information from UNet models based on softmax maximum probability and from deep ensembles and MC dropout UNets using mean voxel-wise entropy. Upon visual assessment, areas of high uncertainty are localized in the periphery of segmentations and are in good agreement with incorrectly labelled voxels. Furthermore, we evaluate how well uncertainty measures correlate with segmentation quality (Dice score). Mean uncertainty over the segmented region (Ulabelled) derived from conventional UNet models does not show a strong quantitative relationship with the Dice score (Spearman correlation coefficient of -0.45 for the soft Dice vs -0.64 for the wcc model respectively). By comparison, image-level uncertainty measures derived from soft Dice as well as wcc MC UNet and deep UNet ensemble models correlate well with the Dice score. In conclusion, using uncertainty information offers ways to assess segmentation quality fully automatically without access to ground truth. Models trained using weighted categorical cross-entropy offer more meaningful uncertainty information on a voxel-level.

2019

Exploring local rotation invariance in 3D CNNs with steerable filters

Conference ArODES

Vincent Andrearczyk, Julien Fageot, Valentin Oreiller, Xavier Montet, Adrien Depeursinge

Proceedings of the International Conference on Medical Imaging with Deep Learning (MIDL) 2019

Link to the conference

Summary:

Locally Rotation Invariant (LRI) image analysis was shown to be fundamental in many applications and in particular in medical imaging where local structures of tissues occur at arbitrary rotations. LRI constituted the cornerstone of several breakthroughs in texture analysis, including Local Binary Patterns (LBP), Maximum Response 8 (MR8) and steerable lterbanks. Whereas globally rotation invariant Convolutional Neural Networks (CNN) were recently proposed, LRI was very little investigated in the context of deep learning. We use trainable 3D steerable lters in CNNs in order to obtain LRI with directional sensitivity, i.e. non-isotropic. Pooling across orientation channels after the rst convolution layer releases the constraint on nite rotation groups as assumed in several recent works. Steerable lters are used to achieve a ne and ecient sampling of 3D rotations. We only convolve the input volume with a set of Spherical Harmonics (SHs) modulated by trainable radial supports and directly steer the responses, resulting in a drastic reduction of trainable parameters and of convolution operations, as well as avoiding approximations due to interpolation of rotated kernels. The proposed method is evaluated and compared to standard CNNs on 3D texture datasets including synthetic volumes with rotated patterns and pulmonary nodule classication in CT. The results show the importance of LRI in CNNs and the need for a ne rotation sampling.

A lung graph model for the classification of interstitial lung diseases on CT images

Conference ArODES

Guillaume Vanoost, Yashin Dicente Cid, Daniel Rubin, Adrien Depeursinge

Proceedings of the Medical Imaging 2019: Computer-Aided Diagnosis

Link to the conference

Summary:

Diagnosing Interstitial Lung Diseases (ILD) is a difficult task. It requires experienced chest radiologists that may not be available in less-specialized health centers. Moreover, a correct diagnosis is needed to decide for an appropriate treatment and prognostic. In this paper, we focus on the classification of 3 common subtypes of ILDs: Usual Interstitial Pneumonia (UIP), Non-Specific Interstitial Pneumonia (NSIP) and Chronic Hypersensitivity Pneumonitis (CHP). We propose a graph model of the lungs built from a large dataset. The structure of the graph is inspired from medical knowledge of disease predominance, where the nodes correspond to 24 distinct regions obtained from lateral, anterior-posterior and vertical splits of the images. The adjacency matrix is built from distances between intensity distributions of distinct regions. Graphs models are interpretable and were successfully used in neuroimaging. However, to the best of our knowledge, this is the first attempt to use a graph model of the lungs for classifying ILDs. In the particular case of ILDs, graph methods are relevant for the following reasons. In order to differentiate between the subtypes, not only the types of local patterns of the disease are important but also their anatomical location. Therefore, we hypothesize that the comparison between regional distributions of Hounsfield Unit (HU) values is relevant to discriminate between the considered ILD subtypes. For instance, typical UIP shows a spatial predominance of reticular abnormalities and honeycombing in the peripheral regions of the lung bases. Therefore, we expect a marked difference of HU distributions between the central and peripheral regions of the lung bases. Moreover, the construction of the graph leads to an interpretable patient descriptor. The descriptor led to encouraging area under the Receiver Operating Characteristic (ROC) curve in 0.6-0.8 for one-versus-one classification configurations, which also showed to outperform feature sets based on a simple concatenation of regional HU distributions.

Learning cross-protocol radiomics and deep feature standardization from CT images of texture phantoms

Conference ArODES

Vincent Andrearczyk, Adrien Depeursinge, Henning Müller

Proceedings of SPIE medical Imaging 2019: imaging informatics for healthcare, research, and applications

Link to the conference

Summary:

Radiomics has shown promising results in several medical studies, yet it suers from a limited discrimination and informative capability as well as a high variation and correlation with the tomographic scanner types, pixel spacing, acquisition protocol and reconstruction parameters. This paper introduces a new method to transform image features in order to improve their stability across scanners. This method is based on a two-layer neural network that can learn a non-linear standardization transformation of various types of features including handcrafted and deep features. In this setting, variations in extracted features will be representative of true physiopathological tissue changes in the scanned patients. This approach uses a publicly available texture phantom dataset and can be applied to both hand-crafted radiomic and deep features.

2018

Holographic visualisation and interaction of fused CT, PET and MRI volumetric medical imaging data using dedicated remote GPGPU ray casting

Conference ArODES

Magali Fröhlich, Christophe Bolinhas, Adrien Depeursinge, Antoine Widmer, Nicolas Chevrey, Patric Hagmann, Christian Simon, Vivianne Kokje, Stéphane Gobron

Proceedings of the 21st international Conference on MICCAI 2018

Link to the conference

Summary:

Medical experts commonly use imaging including Computed Tomography (CT), Positron-Emission Tomography (PET) and Magnetic Resonance Imaging (MRI) for diagnosis or to plan a surgery. These scans give a highly detailed representation of the patient anatomy, but the usual Three-Dimensional (3D) separate visualisations on screens does not provide an convenient and performant understanding of the real anatom-ical complexity. This paper presents a computer architecture allowing medical staﬀ to visualise and interact in real-time holographic fused CT, PET, MRI of patients. A dedicated workstation with a wireless connec-tion enables real-time General-Purpose Processing on Graphics Process-ing Units (GPGPU) ray casting computation through the mixed reality (MR) headset. The hologram can be manipulated with hand gestures and voice commands through the following interaction features: instan-taneous visualisation and manipulation of 3D scans with a frame rate of 30 fps and a delay lower than 120 ms. These performances give a seamless interactive experience for the user.

2017

Nouveaux paramètres métaboliques du FDG-PET/TDM pour prédire la récurrence et la survie des cancers du col utérin traité par radio-chimiothérapie

Conference ArODES

N. Scher, F. Herrera, Adrien Depeursinge, Yashin Dicente Cid, Roger Schaer, T. Breuneval, J. Bourhis, J.O. Prior, M. Ozsahin, J. Castelli

Actes du 28ème congrès de la SFRO

Link to the conference

2016

Multidimensional texture analysis for improved prediction of ultrasound liver tumor response to chemotherapy treatment

Conference ArODES

Adrien Depeursinge, Omar S. Al-Kadi, Dimitri Van De Ville

Proceedings of MICCAI 2016

Link to the conference

Summary:

The number density of scatterers in tumor tissue contribute to a heterogeneous ultrasound speckle pattern that can be diffcult to discern by visual observation. Such tumor stochastic behavior becomes even more challenging if the tumor texture heterogeneity itself is investigated for changes related to response to chemotherapy treatment. Here we define a new tumor texture heterogeneity model for evaluating response to treatment. The characterization of the speckle patterns is performed via state-of-the-art multi-orientation and multi-scale circular harmonic wavelet (CHW) frames analysis of the envelope of the radio-frequency signal. The lacunarity measure -corresponding to scatterer number density- is then derived from fractal dimension texture maps within the CHW decomposition, leading to a localized quantitative assessment of tumor texture heterogeneity. Results indicate that evaluating tumor heterogeneity in a multidimensional texture analysis approach could potentially impact on designing an early and effective chemotherapy treatment.

Is tumor heterogeneity quantified by 3D texture analysis of MRI able to predict non-response to NAC in breast cancer ?

Conference ArODES

Nicolas Michoux, Laurent Bollondi, Adrien Depeursinge, Latifa Fellah, Isabelle Leconte, Henning Müller

Proceedings of the ESMRMB 2016, 33rd Annual Scientific Meeting

Link to the conference

GPU-accelerated texture analysis using steerable Riesz wavelets

Conference ArODES

Ranveer Joyseeree, Adrien Depeursinge, Henning Müller, et al.

Proceedings of the 24th Euromicro International Conference on Parallel, Distributed, and Network-Based Processing (PDP) 2016

Link to the conference

Summary:

Visual pattern recognition is a key research topic in the field of image processing and computer vision with many applications including medical diagnosis, identification and classification tasks. Texture analysis based on steerable Riesz wavelets is powerful, but requires computing pixel–wise operations resulting in a run time in the order of days when large volumes of data are processed. To overcome this limitation we propose a Graphics Processing Unit (GPU) based solution. A standard CPU version is used as starting point for the development of baseline GPU versions. To further increase the performance, and to overcome compute and memory limitations we apply a series of optimization techniques, leading to five versions in total. The best performing GPU solution ensures a speed–up of 93x for the parallelized section of the application and of 29.6x for the entire application. Furthermore, we show that a higher Riesz order and/or a higher image resolution further increases the speed–up.

2015

Quantitative image texture analysis predicts malignancy on multiparametric prostate MRI

Conference ArODES

Adrien Depeursinge, Geoffrey Sonn, Lewis Hahn, Richard Fan, Jocelyn Barker, Daniel Rubin

Proceedings of the 91st Annual Meeting of the Western Section of American Urological Association (WSAUA) 2015

Link to the conference

Summary:

This study demonstrates the feasibility of using quantitative Riesz texture analysis to predict whether a suspicious lesion on mpMRI (PI-RADS ≥3) is cancerous. Particularly notable is the 85% correct prediction for PI-RADS 4 lesions, given the tremendous variability of biopsy results for PI-RADS 4 lesions among different radiologists. Validation of this approach in a larger dataset is ongoing. In the future, we expect that quantitative image analysis will be incorporated into grading systems to refine prostate mpMRI image interpretation, enable greater reproducibility across radiologists, and improve patient counselling and decision making about prostate biopsy.

Holistic classification of CT attenuation patterns for interstitial lung diseases via deep convolutional neural networks

Conference ArODES

Mingchen Gao, Ulas Bagci, Adrien Depeursinge, et al.

Proceedings of MICCAI 2015

Link to the conference

Summary:

Interstitial lung diseases (ILD) involve several abnormal imaging patterns observed in computed tomography (CT) images. Accurate classification of these patterns plays a significant role in precise clinical decision making of the extent and nature of the diseases. Therefore it is important for developing automated pulmonary computer-aided detection (CAD) systems. Conventionally, this task relies on experts’ manual identification of regions of interest (ROIs) as a prerequisite to diagnose potential diseases. This protocol is time consuming and inhibits fully automatic assessment. In this paper, we present a new method to classify ILD imaging patterns on CT images. The main difference is that the proposed algorithm uses the entire image as a holistic input. By circumventing the prerequisite of manually input ROIs, our problem setup is significantly more difficult than previous work but can better address the clinical workflow. Qualitative and quantitative results using a publicly available ILD database demonstrates state-of-the-art classification accuracy under the patch based classification and shows the potential of predicting the ILD type using holistic image.

3D Riesz-wavelet based Covariance descriptors for texture classification of lung nodule tissue in CT

Conference ArODES

Henning Müller, Pol Cirujeda, Daniel Rubin, Todd Aguilera, Billy Loo Jr., Maximilian Diehn, Xavier Binefa, Adrien Depeursinge

Proceedings of the 37th Annual International Conference of the IEEE Engineering in Medicine and Biology Society (EMBC) 2015

Link to the conference

Summary:

In this paper we present a novel technique for characterizing and classifying 3D textured volumes belonging to different lung tissue types in 3D CT images. We build a volume-based 3D descriptor, robust to changes of size, rigid spatial transformations and texture variability, thanks to the integration of Riesz-wavelet features within a Covariance-based descriptor formulation. 3D Riesz features characterize the morphology of tissue density due to their response to changes in intensity in CT images. These features are encoded in a Covariance-based descriptor formulation: this provides a compact and flexible representation thanks to the use of feature variations rather than dense features themselves and adds robustness to spatial changes. Furthermore, the particular symmetric definite positive matrix form of these descriptors causes them to lay in a Riemannian manifold. Thus, descriptors can be compared with analytical measures, and accurate techniques from machine learning and clustering can be adapted to their spatial domain. Additionally we present a classification model following a “Bag of Covariance Descriptors” paradigm in order to distinguish three different nodule tissue types in CT: solid, ground-glass opacity, and healthy lung. The method is evaluated on top of an acquired dataset of 95 patients with manually delineated ground truth by radiation oncology specialists in 3D, and quantitative sensitivity and specificity values are presented.

Hierarchic anatomical structure segmentation guided by spatial correlations (AnatSeg-Gspac) :

Conference ArODES

VISCERAL Anatomy3

Henning Müller, Oscar Alfonso Jiménez del Toro, Yashin Dicente Cid, Adrien Depeursinge

Workshop proceedings of the VISCERAL Anatomy3 organ segmentation challenge co-located with IEEE 12th International Symposium on Biomedical Imaging (ISBI) 2015

Link to the conference

Summary:

Medical image analysis techniques require an initial localization and segmentation of the corresponding anatomical structures. As part of the VISCERAL Anatomy segmentation benchmarks, a hierarchical multi-atlas multi-structure segmentation approach guided by anatomical correlations is proposed (AnatSeg-Gspac). The method defines a global alignment of the images and renes locally the anatomical regions of interest for the smaller structures. In this paper, the method is evaluated in the VISCERAL Anatomy3 benchmark in twenty anatomical structures in both contrast-enhanced and non-enhanced computed tomography (CT) scans. AnatSeg-Gspac obtained the lowest average Hausdor distance in 19 out of the 40 possible structure scores in the test set CT scans.

USYD/HES-SO in the VISCERAL Retrieval Benchmark

Conference ArODES

Henning Müller, Fan Zhang, Yang Song, Weidong Cai, Adrien Depeursinge

Proceedings of the 37th European Conference on Information Retrieval (ECIR) Workshop on Multimodal Retrieval in the Medical Domain 2015

Link to the conference

Summary:

This report presents the participation of our joint research team in the VISCERAL retrieval task. Given a query case, the cases with highest similarities in the database were retrieved. 5 runs were submitted for the 10 queries provided in the task, of which two were based on the anatomy-pathology terms, two were based on the visual image content, and the last one was based on the fusion of the aforementioned four runs.

Efficient and fully automatic segmentation of the lungs in CT volumes

Conference ArODES

Henning Müller, Yashin Dicente Cid, Oscar Alfonso Jiménez del Toro, Adrien Depeursinge

Workshop proceedings of the VISCERAL Anatomy3 organ Segmentation Challenge co-located with IEEE 12th International Symposium on Biomedical Imaging (ISBI) 2015

Link to the conference

Summary:

The segmentation of lung volumes constitutes the first step for most computer-aided systems for lung diseases. CT (Computed Tomography) is the most common imaging technique used by these systems, so fast and accurate methods are needed to for allow early and reliable analysis. In this paper, an efficient and fully automatic method for the segmentation of the lung volumes in CT is presented. This method deals with the initial segmentation of the respiratory system, the posterior extraction of the air tracks, and the final identification of the tow lungs with 3 novel approaches. The system relies only on anatomical assumptions and was evaluated in the context of the VISCERAL Anatomy3 Challenge, achieving one of the best results.

Optimized steerable wavelets for texture analysis of lung tissue in 3-D CT: classification of usual interstitial pneumonia :

Conference ArODES

classification of usual interstitial pneumonia

Adrien Depeursinge, Pedram Pad, Anne S. Chin, Ann N. Leung, Daniel L. Rubin, Henning Müller, Michael Unser

Proceedings of the IEEE 12th International Symposium on Biomedical Imaging 2015 (ISBI)

Link to the conference

Summary:

Our aim is to optimize wavelet–based feature extraction for differentiating between the classical versus atypical pattern of usual interstitial pneumonia (UIP) in volumetric CT. Our proposal is to act on the bandwidth of steerable wavelets while maintaining their tight frame property. To that end, we designed a family of maximally localized wavelet pyramids in 3–D for a continuously adjustable radial bandwidth [ ; ], 2 [=4; =2]. The proposed wavelets are coupled with a rotation–covariant directional operator based on the Riesz transform, which provides characterizations of the organization image directions independently from their local orientations. The influence of the wavelet bandwidth on the classification performance was found to be large with area under the receiver operating characteristic curve (AUC) values in [0:784; 0:921]. This demonstrated the importance of finding the minimum spatial support of the wavelet required to leverage the wealth of morphological tissue properties in the vicinity of the lung boundaries.

Predicting non-response to NAC in patients with breast cancer using 3D texture analysis

Conference ArODES

Nicolas Michoux, Laurent Bollondi, Adrien Depeursinge, Antoine Geissbuhler, Latifa Fellah, Henning Müller, I. Leconte

Proceedings of European Congress of Radiology 2015

Link to the conference

Pulmonary embolism detection using localized vessel-based features in dual energy CT

Conference ArODES

Yashin Dicente Cid, Adrien Depeursinge, Antonio Foncubierta-Rodriguez, Henning Müller, Alexandra Platon, Pierre-Alexandre Poletti

Proceedings of SPIE medical imaging 2015

Link to the conference

Summary:

Pulmonary embolism (PE) aects up to 600,000 patients and contributes to at least 100,000 deaths every year in the United States alone. Diagnosis of PE can be dicult as most symptoms are unspecic and early diagnosis is essential for successful treatment. Computed Tomography (CT) images can show morphological anomalies that suggest the existance of PE. Various image-based procedures have been proposed for improving computer-aided diagnosis of PE. We propose a novel method for detecting PE based on localized vessel-based features computed in Dual Energy CT (DECT) images. DECT provides 4D data indexed by the three spatial coordinates and the energy level. The proposed features encode the variation of the Hounseld Units across the dierent levels and the CT attenuation related to the amount of iodine contrast in each vessel. A local classication of the vessels is obtained through the classication of these features. Moreover, the localization of the vessel in the lung provides better comparison between patients. Results show that the simple features designed are able to classify pulmonary embolism patients with an AUC (area under the receiver operating curve) of 0.71 on a lobe basis. Prior segmentation of the lung lobes is not necessary because an automatic atlas-based segmentation obtains similar AUC levels (0.65) for the same dataset. The automatic atlas reaches 0.80 AUC in a larger dataset with more control cases.

Texture classification of anatomical structures using a context-free machine learning approach

Conference ArODES

Oscar Alfonso Jimenez del Toro, Antonio Foncubierta-Rodriguez, Adrien Depeursinge, Henning Müller

Proceedings of SPIE medical imaging 2015

Link to the conference

Summary:

Medical images contain a large amount of visual information about structures and anomalies in the human body. To make sense of this information, human interpretation is often essential. On the other hand, computer-based approaches can exploit information contained in the images by numerically measuring and quantifying specic visual features. Annotation of organs and other anatomical regions is an important step before computing numerical features on medical images. In this paper, a texture-based organ classication algorithm is presented, which can be used to reduce the time required for annotating medical images. The texture of organs is analyzed using a combination of state{of{the{art techniques: the Riesz transform and a bag of meaningful visual words. The eect of a meaningfulness transformation in the visual word space yields two important advantages that can be seen in the results. The number of descriptors is enormously reduced down to 10% of the original size, whereas classication accuracy is improved by up to 25% with respect to the baseline approach.

2014

A semantic framework for the retrieval of similar radiological images based on medical annotations

Conference ArODES

Adrien Depeursinge, Camille Kurtz, Sandy Napel, Christopher Beaulieu, Daniel Rubin

Proceedings of the 21st IEEE International Conference on Image Processing (ICIP) 2014

Link to the conference

Summary:

Image retrieval approaches can assist radiologists by finding similar images in databases as a means to providing decision support. In general, images are indexed using low-level imaging features, and a distance function is used to find the best matches in the feature space. However, using low-level features to capture the appearance of diseases in images is challenging and the semantic gap between these features and the high-level visual concepts in radiology may impair the system performance. We present a semantic framework that enables retrieving similar images based on high-level semantic image annotations. This framework relies on (1) an automatic approach to predict the annotations as semantic terms from Riesz texture image features and (2) a distance function to compare images considering both texture-based and radiodensity-based similarities among image annotations. Experiments performed on CT images emphasize the relevance of this framework.

Achievements

PEOPLE@HES-SO Directory and Skills inventory

Depeursinge Adrien

Professeur-e HES Ordinaire

Professeur-e HES Ordinaire

Research Interests

PEOPLE@HES-SO
Directory and Skills inventory