Delta2Gamma: Band-Wise Self-Supervised EEG Learning Achieves 92% Accuracy in Alzheimer's Detection
Dementia screening faces a big hurdle due to the lack of budget-friendly diagnostic options. While imaging methods can accurately diagnose, they are quite expensive. On the other hand, electroencephalography (EEG) is more affordable and portable, but its noisy data and lack of clinical labels make it tricky for traditional machine learning. To tackle this issue, researchers developed Delta2Gamma, a self-supervised method that learns from unlabeled EEG data by comparing enhanced signal views. It separates recordings into five neural rhythms: delta, theta, alpha, beta, and gamma, each with its own encoder. When tested on the ADFTD cohort, Delta2Gamma reached an impressive 92% accuracy in distinguishing Alzheimer's from cognitively healthy individuals. This research is available on arXiv (identifier 2608.17231) and marks a big step forward in affordable dementia screening.
Key facts
- Delta2Gamma is a self-supervised framework for learning EEG representations from unlabeled data.
- It contrasts augmented views of each signal to learn representations.
- The framework decomposes EEG recordings into five canonical neural rhythms: delta, theta, alpha, beta, and gamma.
- Each frequency band receives its own encoder and projection head.
- Each band also gets an adaptive temperature predicted during contrastive training.
- The ADFTD cohort was used for evaluation.
- A strict leave-one-subject-out protocol was employed.
- Delta2Gamma distinguishes Alzheimer's disease from cognitively normal controls with 92% accuracy.
Entities
Institutions
- arXiv