Contacts
Intranet
SK
Zuzana Rošťáková
National projects
| EDABSS – EEG data analysis by blind source separation methods | |
| Analýza EEG signálu pomocou metód hľadania skrytých zdrojov | |
| Program: | Plán obnovy EÚ |
| Duration: | 1.9.2024 – 31.8.2026 |
| Project leader: | Mgr. Rošťáková Zuzana, PhD. |
| Annotation: | Blind source separation (BSS) approaches are unsupervised machine learning methods focused on the detection of hidden, directly unobservable (latent) structure of real-world data. They play a crucial role in image processing, medical imaging, and music. The proposed project focuses mainly on human electroencephalogram (EEG), for which BSS is beneficial when detecting the narrowband brain oscillations representing brain processes either in health or disease. Two-dimensional BSS methods like principal or independent component analysis are easily applicable and understandable for a broader medical and neurophysiological community. However, the estimated latent component properties are usually incompatible with the real electrophysiological signal character. Consequently, they miss their neurophysiological interpretation. Tensor decomposition is a complex but more flexible mathematical procedure that allows adapting the model structure and constraints to the solution to mimic real-world signal characteristics. The proposed project focuses on tensor decomposition as a tool for i) EEG preprocessing, artefact detection and removal, ii) EEG latent structure analysis using a nonnegative tensor decomposition with block structure allowing to model various relationships between latent components, and iii) post-decomposition analysis of latent component dynamic properties. Obtaining comprehensive information about EEG latent structure and developing novel, user-friendly algorithms is crucial for better understanding brain processes and new methods for treating neurophysiological diseases and disorders. |