Leveraging pretrained vision transformers for classifying alcohol use disorder using raw resting-state EEG

Alcohol Use Disorder (AUD) is a prevalent neuropsychiatric condition affecting about 28 million adults in the USA, with few objective biomarkers to assist in its clinical diagnosis. In this study, we investigated the potential of deep learning to classify individuals with AUD using raw resting-state electroencephalogram (EEG) data. EEG recordings were obtained from the Collaborative Study on the Genetics of Alcoholism (COGA). The initial cohort included a total of 5402 recordings from 2710 participants (aged 13-83, mean age 24; 1512 males and 1198 females). Minimal preprocessing was applied to preserve the raw EEG features. We utilized EEGViT, a hybrid deep learning architecture that combines convolutional patch embedding with a Vision Transformer (ViT) pretrained on ImageNet, thereby enabling end-to-end learning directly from raw EEG input. The analysis groups were matched and stratified by age and sex. The architecture was evaluated independently on Cannabis Use Disorder (CUD) and Opioid Use Disorder (OUD). Results for the AUD model showed a classification accuracy of approximately 57% in the overall dataset, 55% for males, and 59% for females. The CUD model showed an accuracy of about 64%, with 58% for females and 67% for males. The OUD model showed an accuracy of about 63%, with 61% for females and 65% for males. Temporal analysis indicated higher accuracy in later minutes than in earlier ones. While modest, these findings provide preliminary evidence that transformer-based models can be applied to psychiatric classification using raw EEG data and establish a foundation for future research on EEG-based approaches.