Description du projet :
Sleep studies (also called polysomnography) are well-known non-invasive painless, often overnight, tests that allow doctors to monitor patients while asleep to unveil problems related to their brain and body. Sleep studies are a powerful diagnosis technique for a wide range of sleep disorders, such as insomnia, sleep apnoea, narcolepsy, or sleepwalking. Sleep studies lasting 1-2 hours are often used to diagnose different types of epilepsy, since sleep disorders and epilepsy are often comorbid and intertwined in a complex way, greatly hindering diagnosis.
Regardless of the application, sleep studies monitor brain activity by using standard Electroencephalogram (EEG) measuring the electrical activity of brain cells using small electrodes attached to the scalp. Because brain cells keep producing electrical currents even during sleep, EEG aims at capturing changes in the pattern of waveforms. Such changes are specific to different brain areas. Therefore, to capture global brain activity, the EEG electrodes must cover the whole head. The higher the density of the EEG electrodes placement, the higher the resolution of the brain electric activity. Therefore, nowadays, sleep studies employ high density EEGs (hdEEG) using up to 256 channels.
The algorithms are required to pre-process, filter, and analyse hdEEG signals are complex and computationally heavy, as they require feature extraction in the time and frequency domains, thus requiring at least 256-channels Fourier Transforms. Real-time analysis and visualization of hdEEG remains as of today a challenge. Therefore, EEG sleep data are usually recorded together with video of the patients asleep and analysed by doctor or clinician offline and oftentimes manually. However, clinicians will only analyse a set of channels, which greatly diminishes the benefits and purpose of hdEEG and overlooks the global view.
The goal of this project is to tackle this challenge by developing a tool for real-time processing and visualization of hdEEG. We aim at providing clinicians with the possibility to observe important for sleep studies events such as spindles, K-complexes, and slow waves in real-time as well as fluctuation of specific EEG signal features, such as spectral power, entropy, spectral edge frequency and other, associated with electrode location and thus mapped to a scalp. This will not only allow much faster evaluation of sleep related disorders, but also allow a more global and comprehensive view of brain activity, thus opening new avenues of research in much broader domains of applications.
To accomplish this goal, the project puts together expertise in two different domains:
1. The proposal of algorithms for noise and artifact elimination from EEG signals, features extraction and detection of different sleep related events using signal processing techniques and machine learning algorithms will be caried out by the group of prof. Alena Simalatsar, the PI of this project, in HES-SO Valais, Sion.
2. The development hardware and software acceleration techniques by means of using heterogeneous systems composing of CPUs, GPUs and/or FPGAs and adequate workload management and partitioning from the edge (EEG machine) to the server to enable real-time processing and visualization of hdEEG without quality loss will be performed by the group of prof. Marina Zapater, in HEIG-VD, Yverdon-les-Bains.
Equipe de recherche au sein de la HES-SO:
Hennebert Jean
, Brunet Yorick
, Petraglio Enrico
, Chacun Guillaume
, Simalatsar Alena
, Da Rocha Carvalho Bruno
, Extrat Bastien
, Zapater Sancho Marina
, Maillard Philippe
Partenaires académiques: ReDS; FR - EIA - Institut HumanTech
Durée du projet:
01.05.2025 - 30.04.2026
Montant global du projet: 220'000 CHF
Statut: En cours
SWITCH edu-ID
Administration
