Suche

Graziani Mara

Assistant-e post-doc

Hauptkompetenzen

Assistant-e post-doc

Telefon-Nummer: +41 58 606 90 21

Büro: TP116

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

Es müssen keine Daten für diesen Abschnitt angezeigt werden.

Laufend

PROCESS: Exascale Medical Imaging Use Case

Rolle: Hauptgesuchsteller/in

Description du projet :

Exascale learning in medical image data aims at improving the performance of decision support in the process of cancer detection, localisation and staging/grading to optimize and personalize treatment planning.

Forschungsteam innerhalb von HES-SO: Graziani Mara

Durée du projet: 01.11.2017

Url des Projektstandortes: https://www.process-project.eu/exascale-learning-on-medical-image-data/

Statut: Laufend

2021

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

Scientifique ArODES

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

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

Link zur Publikation

Zusammenfassung:

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.

Process data infrastructure and data services

Scientifique ArODES

Cushing Reginald, Valkering Onno, Belloum Adam, Madougou Souley, Bobak Martin, Habala Ondrej, Tran Viet, Meizner Jan, Nowakowski Piotr, Graziani Mara, Müller Henning

Computers and informatics, 2020, vol. 39, no. 4, pp. 724-756

Link zur Publikation

Zusammenfassung:

Due to energy limitation and high operational costs, it is likely that exascale computing will not be achieved by one or two datacentres but will require many more. A simple calculation, which aggregates the computation power of the 2017 Top500 supercomputers, can only reach 418 petaflops. Companies like Rescale, which claims 1.4 exaflops of peak computing power, describes its infrastructure as composed of 8 million servers spread across 30 datacentres. Any proposed solution to address exascale computing challenges has to take into consideration these facts and by design should aim to support the use of geographically distributed and likely independent datacentres. It should also consider, whenever possible, the co-allocation of the storage with the computation as it would take 3 years to transfer 1 exabyte on a dedicated 100 Gb Ethernet connection. This means we have to be smart about managing data more and more geographically dispersed and spread across different administrative domains. As the natural settings of the PROCESS project is to operate within the European Research Infrastructure and serve the European research communities facing exascale challenges, it is important that PROCESS architecture and solutions are well positioned within the European computing and data management landscape namely PRACE, EGI, and EUDAT. In this paper we propose a scalable and programmable data infrastructure that is easy to deploy and can be tuned to support various data-intensive scientific applications.

Breast histopathology with high-performance computing and deep learning

Scientifique ArODES

Graziani Mara, Eggel Ivan, Deligand François, Bobák Martin, Andrearczyk Vincent, Müller Henning

Computing and informatics, 2020, vol. 39, no. 4, pp. 780-807

Link zur Publikation

Zusammenfassung:

The increasingly intensive collection of digitalized images of tumor tissue over the last decade made histopathology a demanding application in terms of computational and storage resources. With images containing billions of pixels, the need for optimizing and adapting histopathology to large-scale data analysis is compelling. This paper presents a modular pipeline with three independent layers for the detection of tumoros regions in digital specimens of breast lymph nodes with deep learning models. Our pipeline can be deployed either on local machines or high-performance computing resources with a containerized approach. The need for expertise in high-performance computing is removed by the self-su_cient structure of Docker containers, whereas a large possibility for customization is left in terms of deep learning models and hyperparameters optimization. We show that by deploying the software layers in di_erent infrastructures we optimize both the data preprocessing and the network training times, further increasing the scalability of the application to datasets of approximatively 43 million images. The code is open source and available on Github.

Reference exascale architecture (extended version)

Scientifique ArODES

Bobák Martin, Hluchý Ladislav, Habala Ondrej, Tran Viet, Cushing Reginald, Valkering Onno, Belloum Adam, Graziani Mara, Müller Henning, Madougou Souley, Maassen Jason

Computing and informatics, 2020, vol. 34, no. 4, pp. 644-677

Link zur Publikation

Zusammenfassung:

While political commitments for building exascale systems have been made, turning these systems into platforms for a wide range of exascale applications faces several technical, organisational and skills-related challenges. The key technical challenges are related to the availability of data. While the first exascale machines are likely to be built within a single site, the input data is in many cases impossible to store within a single site. Alongside handling of extreme-large amount of data, the exascale system has to process data from different sources, support accelerated computing, handle high volume of requests per day, minimize the size of data flows, and be extensible in terms of continuously increasing data as well as an increase in parallel requests being sent. These technical challenges are addressed by the general reference exascale architecture. It is divided into three main blocks: virtualization layer, distributed virtual file system, and manager of computing resources. Its main property is modularity which is achieved by containerization at two levels: 1) application containers - containerization of scientific workflows, 2) micro-infrastructure - containerization of extreme-large data service-oriented infrastructure. The paper also presents an instantiation of the reference architecture - the architecture of the PROCESS project (PROviding Computing solutions for ExaScale ChallengeS) and discusses its relation to the reference exascale architecture. The PROCESS architecture has been used as an exascale platform within various exascale pilot applications. This paper also presents performance modelling of exascale platform with its validation.

2020

Concept attribution :

Scientifique ArODES

explaining CNN decisions to physicians

Graziani Mara, Andrearczyk Vincent, Marchand-Maillet Stéphane, Müller Henning

Computers in biology and medicine, 2020, vol. 123, Article 103865, pp. 1-11

Link zur Publikation

Zusammenfassung:

Deep learning explainability is often reached by gradient-based approaches that attribute the network output to perturbations of the input pixels. However, the relevance of input pixels may be difficult to relate to relevant image features in some applications, e.g. diagnostic measures in medical imaging. The framework described in this paper shifts the attribution focus from pixel values to user-defined concepts. By checking if certain diagnostic measures are present in the learned representations, experts can explain and entrust the network output. Being post-hoc, our method does not alter the network training and can be easily plugged into the latest state-of-the-art convolutional networks. This paper presents the main components of the framework for attribution to concepts, in addition to the introduction of a spatial pooling operation on top of the feature maps to obtain a solid interpretability analysis. Furthermore, regularized regression is analyzed as a solution to the regression overfitting in high-dimensionality latent spaces. The versatility of the proposed approach is shown by experiments on two medical applications, namely histopathology and retinopathy, and on one non-medical task, the task of handwritten digit classification. The obtained explanations are in line with clinicians’ guidelines and complementary to widely used visualization tools such as saliency maps.

2019

Visualizing and interpreting feature reuse of pretrained CNNs for histopathology

Chapter ArODES

Graziani Mara, Andrearczyk Vincent, Müller Henning

Dans Courtney, Jane, Deegan, Catherine, Leamy, Paul, MVIP 2019: Irish Machine Vision and Image Processing Conference Proceedings (4 p.). 2019, Dublin, Ireland : Irish Pattern Recognition and Classification Society

Link zur Publikation

Zusammenfassung:

Reusing the parameters of networks pretrained on large scale datasets of natural images, such as ImageNet, is a common technique in the medical imaging domain. The large variability of objects and classes is, however, drastically reduced in most medical applications where images are dominated by repetitive patterns with, at times, subtle differences between the classes. This paper takes the example of finetuning a pretrained convolutional network on a histopathology task. Because of the reduced visual variability in this application domain, the network mostly learns to detect textures and simple patterns. As a result, the complex structures that maximize the channel activations of deep layers in the pretrained network are not present after finetuning. The learned features seem to be used by the network to spot atypical nuclei in the images, as shown by class activation maps. Finally, texture measures appear discriminative after finetuning, as shown by accurate Regression Concept Vectors.

Heterogeneous exascale computing

Chapter ArODES

Hluchý Ladislav, Bobák Martin, Müller Henning, Graziani Mara, Heikkurinen Matti, et al.

Dans Haidegger, Tamás, Kovács, Levente, Szakál, Anikó, Recent advances in intelligent engineering : volume dedicated to Imre J. Rudas’ Seventieth Birthday (pp. 81-110). 2019, Cham : Springer

Link zur Publikation

Zusammenfassung:

Exascale services bring new unique challenges that the current computational, big data and workflow solutions are unable to meet. The chapter includes a detailed description of selected exascale services with known state of the art in extreme date solutions. The integration of requirements and the analysis of the state of the art in the exascale field is centered in on a description of a high-level architectural approach. The next main contribution of the paper is the description of the architecture capable to handle heterogeneous exascale services coming from both academic as well as industrial sphere. Those two models represent a (conceptual, and technological) design of a platform that addresses the requirements of the use cases. The resulting architecture will help us provide computing solutions to exascale challenges within the H2020 project PROCESS.

2017

Semi-automatic training of an object recognition system in scene camera data using gaze tracking and accelerometers

Chapter ArODES

Cognolato Matteo, Graziani Mara, Giordaniello Francesca, Saetta Gianluca, Bassetto , Brugger Peter, Caputo Barbara, Müller Henning, Atzori Manfredo

Computer Vision Systems : 11th International Conference, ICVS 2017, Shenzhen, China, July 10-13, 2017 (pp. 175-184). 2017, Cham : Springer

Link zur Publikation

Zusammenfassung:

Object detection and recognition algorithms usually require large, annotated training sets. The creation of such datasets requires expensive manual annotation. Eye tracking can help in the annotation procedure. Humans use vision constantly to explore the environment and plan motor actions, such as grasping an object. In this paper we investigate the possibility to semi-automatically train object recognition with eye tracking, accelerometer in scene camera data, learning from the natural hand-eye coordination of humans. Our approach involves three steps. First, sensor data are recorded using eye tracking glasses that are used in combination with accelerometers and surface electromyography that are usually applied when controlling prosthetic hands. Second, a set of patches are extracted automatically from the scene camera data while grasping an object. Third, a convolutional neural network is trained and tested using the extracted patches. Results show that the parameters of eye-hand coordination can be used to train an object recognition system semi-automatically. These can be exploited with proper sensors to fine-tune a convolutional neural network for object detection and recognition. This approach opens interesting options to train computer vision and multi-modal data integration systems and lays the foundations for future applications in robotics. In particular, this work targets the improvement of prosthetic hands by recognizing the objects that a person may wish to use. However, the approach can easily be generalized.

2021

Sharpening local interpretable model-agnostic explanations for histopathology :

Conference ArODES

improved understandability and reliability

Henning Müller, Mara Graziani, Vincent Andrearczyk

Medical Image Computing and Computer Assisted Intervention – MICCAI 2021 : 24th International Conference, Strasbourg, France, September 27–October 1, 2021, Proceedings, Part III

Link zur Konferenz

Zusammenfassung:

Being accountable for the signed reports, pathologists may be wary of high-quality deep learning outcomes if the decision-making is not understandable. Applying off-the-shelf methods with default configurations such as Local Interpretable Model-Agnostic Explanations (LIME) is not sufficient to generate stable and understandable explanations. This work improves the application of LIME to histopathology images by leveraging nuclei annotations, creating a reliable way for pathologists to audit black-box tumor classifiers. The obtained visualizations reveal the sharp, neat and high attention of the deep classifier to the neoplastic nuclei in the dataset, an observation in line with clinical decision making. Compared to standard LIME, our explanations show improved understandability for domain-experts, report higher stability and pass the sanity checks of consistency to data or initialization changes and sensitivity to network parameters. This represents a promising step in giving pathologists tools to obtain additional information on image classification models. The code and trained models are available on GitHub.

Evaluation and comparison of CNN visual explanations for histopathology

Conference ArODES

Henning Müller, Mara Graziani, Vincent Andrearczyk

Proceedings of the 35th AAAI Conference on Artificial Intelligence

Link zur Konferenz

Zusammenfassung:

Visualization methods for Convolutional Neural Networks (CNNs) are spreading within the medical community to obtain explainable AI (XAI). The sole qualitative assessment of the explanations is subject to a risk of confirmation bias. This paper proposes a methodology for the quantitative evaluation of common visualization approaches for histopathology images, i.e. Class Activation Mapping and Local-Interpretable ModelAgnostic Explanations. In our evaluation, we propose to assess four main points, namely the alignment with clinical factors, the agreement between XAI methods, the consistency and repeatability of the explanations. To do so, we compare the intersection over union of multiple visualizations of the CNN attention with the semantic annotation of functionally different nuclei types. The experimental results do not show stronger attributions to the multiple nuclei types than those of a randomly initialized CNN. The visualizations hardly agree on salient areas and LIME outputs have particularly unstable repeatability and consistency. The qualitative evaluation alone is thus not sufficient to establish the appropriateness and reliability of the visualization tools. The code is available on GitHub at bit.ly/2K4BHKz.

2020

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

Conference ArODES

Adrien Depeursinge, Henning Müller, Mara Graziani, Vincent Andrearczyk

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

Link zur Konferenz

Zusammenfassung:

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.

Consistency of scale equivariance in internal representations of CNNs

Conference ArODES

Adrien Depeursinge, Henning Müller, Mara Graziani, Vincent Andrearczyk

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

Link zur Konferenz

Zusammenfassung:

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.

2019

Improved interpretability for computer-aided severity assessment of retinopathy of prematurity

Conference ArODES

Henning Müller, Mara Graziani, Vincent Andrearczyk

Proceedings of the Medical Imaging 2019: Computer-Aided Diagnosis (SPIE)

Link zur Konferenz

Zusammenfassung:

Computer-aided diagnosis tools for Retinopathy of Prematurity (ROP) base their decisions on handcrafted retinal features that highly correlate with expert diagnoses, such as arterial and venous curvature, tortuosity and dilation. Deep learning leads to performance comparable to those of expert physicians, albeit not ensuring that the same clinical factors are learned in the deep representations. In this paper, we investigate the relationship between the handcrafted and the deep learning features in the context of ROP diagnosis. Average statistics on the handcrafted features for each input image were expressed as retinal concept measures. Three disease severity grades, i.e. normal, pre-plus and plus, were classified by a deep convolutional neural network. Regression Concept Vectors (RCV) were computed in the network feature space for each retinal concept measure. Relevant concept measures were identified by bidirectional relevance scores for the normal and plus classes. Results show that the curvature, diameter and tortuosity of the segmented vessels are indeed relevant to the classification. Among the potential applications of this method, the analysis of borderline cases between the classes and of network faults, in particular, can be used to improve the performance.

2018

Regression concept vectors for bidirectional explanations in histopathology

Conference ArODES

Henning Müller, Mara Graziani, Vincent Andrearczyk

Proceedings of the 21st International Conference on MICCAI, 2018

Link zur Konferenz

Zusammenfassung:

Explanations for deep neural network predictions in terms of domain-related concepts can be valuable in medical applications, where justifications are important for confidence in the decision-making. In this work, we propose a methodology to exploit continuous concept measures as Regression Concept Vectors (RCVs) in the activation space of a layer. The directional derivative of the decision function along the RCVs rep- resents the network sensitivity to increasing values of a given concept measure. When applied to breast cancer grading, nuclei texture emerges as a relevant concept in the detection of tumor tissue in breast lymph node samples. We evaluate score robustness and consistency by statisti- cal analysis.

Errungenschaften

Suche

Graziani Mara

Assistant-e post-doc

Hauptkompetenzen

Deep Learning

Machine Learning

Imagerie médicale

Assistant-e post-doc