Memory and Visual Processing EEG for Alcohol Use Disorder Diagnosis with Linear Discriminant Analysis

Computer-aided diagnosis for Alcohol-Use Disorder (AUD) offers a rapid and accurate detection of AUD, preventing further alcohol-related harm. While numerous studies have been proposed aiming for higher accuracy, the reason behind the diagnosis output is mostly not comprehendible, which potentially impedes widespread adoption. Our study aims to present a model of computer-aided diagnosis for AUD classification using an interpretable classifier while minimizing the number of Electroencephalography (EEG) channels. We use the coherence method on selected EEG data associated with processing visual stimuli to derive our features. The resulting connectivity values are then subjected to classification through Linear Discriminant Analysis (LDA). The approach yields discriminant functions with substantial differentiation potential. The most prominent results come from the gamma band features, with a significant p-value < 0.001 and a high canonical correlation of 0.803. Our model achieves a noteworthy classification accuracy of 93.8% and a robust cross-validated accuracy of 90.6%. These outcomes align with prior research indicating that excessive alcohol consumption adversely affects brain function, particularly in visual processing and memory functions tasks.