Brain-inspired computational approach to multivariate analysis of EEG for dementia diagnostics
Reference number | |
Coordinator | KUNGLIGA TEKNISKA HÖGSKOLAN - KTH/CSC Dept. Computational Science and Technology |
Funding from Vinnova | SEK 150 000 |
Project duration | July 2017 - August 2018 |
Status | Completed |
Important results from the project
The project is concerned with novel computational approaches to analysis of multivariate EEG signals with the aim of identifying new biomarkers of dementia. The backbone of the proposed method of exploratory search for spatio-spectro-temporal EEG patterns is derived from a brain-inspired computing algorithm proposed by the KTH group to study neural information processing principles in the hierarchical networks in the brain. We investigated the suitability of the proposed network algorithms as a method for both semi-supervised and unsupervised analysis of time-varying EEG.
Expected long term effects
As expected, the proposed brain-inspirted network model performed well as a feature extractor, classifier and clustering method (unsupervised mode) for EEG patterns. We also demonstrated the computational scalability of our methods. The results suggested that certain level of heterogeneity reflected in the EEG clustering within Alzheimer’s patient group could offer the basis for identifying new groups of clinical relevance. Unfortunately, the hypothesis could not be tested as there was no opportunity to access necessary clinical data within the project timeframe.
Approach and implementation
The project was implemented to large extent as a collaboration between KTH and MentisCura. Following the initial phase of benchmarking EEG features extracted with our new method on a limited dataset, in the next stage we focused on advancing and validation of our network algorithms on more challenging EEG datasets with multiple diagnostic categories of dementia, provided by MentisCura. In the final phase, KTH partner building upon earlier developments applied the method to extract new data representations and cluster them within a broad class of Alzheimer’s patients