Task-guided Spatiotemporal Network for EEG-based Dementia Diagnosis

A study published on arXiv introduces a task-guided spatiotemporal network (TGSN) enhanced by diffusion augmentation for diagnosing dementia and predicting MMSE scores through EEG analysis. This technique tackles feature entanglement in multi-task learning by employing a multi-band feature fusion module that extracts complementary spectral information from EEG signals. Additionally, a pre-trained data augmentation module utilizing a diffusion process enhances the diversity of samples. To model intricate EEG patterns, a gated spatiotemporal attention mechanism is implemented. The goal of this method is to simultaneously diagnose dementia and forecast Mini-Mental State Examination (MMSE) scores based on EEG data.