Synthetic EEG Signal Generation and Alpha Band Feature Extraction

The Individual Alpha Frequency (IAF) is a key neurophysiological biomarker for studying cognitive functioning and brain aging. However, Gold Standard Alpha peak extraction approaches fail approximately 20% of the time, classifying those subjects as non-responders. This creates a situation where precious clinical EEG data are completely discarded and lost. This study seeks to develop a synthetic EEG simulator able to reproduce faithfully a real EEG signal, in order to validate an innovative trial-by-trial Alpha peak extraction approach able to overcome this limitation. Since the exact value of IAF is unknown when you deal with a real EEG signal, the main focus of this study was related to the building of a simulator able to faithfully reproduce a realistic EEG signal. Thus, synthetic EEG signal values allow the creation of a controlled environment in which the Ground Truth parameters are known, so the effectiveness of the innovative extraction approach can be validated. The synthetic EEG signal generation model was implemented in the MATLAB environment, integrating different toolboxes, each one specialized in different applications. FieldTrip was used to address the signal projection at level of electrodes, while EEGLAB was used for topographic visualization and comparison with real data. Qualitative analysis shown that the model created is able to reproduce real EEG activity quite accurately, though inevitable approximations are present. The results of the quantitative analysis indicated that the innovative trial-by-trial approach suffers the Spectral Leakage Phenomenon but is able to correctly select corrupted epochs, suggesting its future applications as an automated epoch screening mechanism to improve efficiency and precision. In the end, the synthetic EEG signal generation model proposed in this study is recommended as an innovative tool for data augmentation with the purpose of training neural networks for research on neurodegenerative diseases.