Predicting Resting-State fMRI Network Dynamics from Concurrent EEG Using Deep Learning Models

The simultaneous recording and analysis of EEG and fMRI have gained considerable attention due to their complementary strengths and distinct physiological perspectives on brain function. In this work, several deep learning models are implemented to perform EEG-to-fMRI prediction. Specifically, multi-channel EEG signals are used to predict the BOLD signals of the seven resting-state networks using deep learning models, owing to their ability to operate without explicit feature engineering. The primary model employed is based on EEGNet, a CNN adapted for regression to perform EEG-to-BOLD fMRI regression. ​