Institutional Repository
Technical University of Crete
Technical University of Crete
EN
|
EL
| URI: | http://purl.tuc.gr/dl/dias/F8D3A4EA-1D11-4F29-B1B6-114B947295D6 | ||
| Year | 2022 | ||
| Type of Item | Diploma Work | ||
| License |
|
||
| Bibliographic Citation | Athanasios Delatolas, "Electromagnetic brain source analysis with statistical and deep learning approaches", Diploma Work, School of Electrical and Computer Engineering, Technical University of Crete, Chania, Greece, 2022 https://doi.org/10.26233/heallink.tuc.92520 | ||
| Appears in Collections |
Electroencephalography (EEG) is a well-established non-invasive recording method for the brain's functional activity. EEG uses an array of electrodes placed on the scalp to record electrical potential signals. EEG provides high temporal but low spatial resolution of brain activity. To gain insight about the spatial dynamics of the EEG, one has to solve the inverse problem of Source Analysis, that is, to find the neural sources that give rise to the recorded EEG activity. There are many existing numerical methods for solving the inverse problem but most of them strongly rely on priors and require significant amount of computational time. Recently, neural networks have been proposed to resolve these issues but their training is based on suboptimal forward modeling and they cannot localize EEG recordings in various brain anatomies. Here, we present a new CNN architecture which is independent of the modeled brain source space and its training is based on realistic and skull-conductivity calibrated head modeling. The performance of our CNN is validated with simulated EEG data and real EEG somatosensory evoked potentials for the first neurological component at 20 ms (P20/N20 response) from three healthy subjects. Our network has localized the P20/N20 component at the subject-specific Brodmann area 3b. Finally, the results suggest that our CNN outperforms the traditional numerical methods.
Copyright © DIAS 2013
