EEG-Based Classification of Neurodegenerative Diseases Using Modern Hopfield Networks

Accurate classification of electroencephalography (EEG) signals plays a key role in the early detection of neurodegenerative diseases such as Alzheimer’s disease and frontotemporal dementia. Traditional machine learning and deep learning methods often struggle to capture the associative memory processes that influence brain activity in these conditions. This study investigates the application of modern Hopfield networks, which are neural architectures inspired by associative memory and conceptually related to attention mechanisms, for classifying EEG data from individuals with AD, FTD, and healthy controls. A publicly available EEG dataset was used, and both spectral and temporal features were extracted using a standardized preprocessing procedure. Four neural models incorporating Hopfield mechanisms were developed and tested through stratified k-fold cross-validation and hold-out evaluation. The results indicate that integrating Hopfield networks significantly enhances classification accuracy. The best-performing model achieved more than 96% accuracy in both binary and multiclass classification tasks. These findings show that Hopfield networks can strengthen EEG feature representation and offer a promising memory-based approach for the classification of neurodegenerative diseases.