Cross-Subject EEG Channel Selection for the Detection of Predisposition to Alcoholism

ElectroEncephaloGram (EEG) is a powerful technology for the early detection, among others, of alcoholism. However, multiple electrodes placed on the scalp to record brain signals may introduce noisy and redundant information, hinder performance and increase computational times in the task of automated decoding of EEG signals. In this work we propose a novel end-to-end Representation Learning-based algorithm to select the most relevant EEG channels to perform detection of predisposition to alcoholism, in a subject-agnostic way. Indeed, EEG signals are characterized by strong subject-specific variance potentially affecting the generalizability of the selection. Results are promising, especially compared to the very limited literature on cross-subject EEG channel selection.