Particle.news
Download on the App Store

Örebro AI Models Spot Dementia From EEG With Reported High Accuracy

The peer-reviewed work emphasizes explainability and privacy-preserving federated training, with broader validation planned before clinical use.

Overview

  • Researchers report two EEG-based AI approaches that distinguish healthy individuals from patients with Alzheimer’s disease or frontotemporal dementia.
  • An explainable framework combining temporal convolutional networks and LSTM classified Alzheimer’s, frontotemporal dementia and healthy controls with over 80% accuracy.
  • A compact hybrid‑fusion EEGNetv4 model trained via federated learning kept patient data local, measured under one megabyte in size and achieved over 97% accuracy.
  • Explainable AI highlights which EEG signal segments and frequency bands—such as alpha, beta and gamma—contributed to each output to aid clinical interpretation.
  • The studies, published in Frontiers journals, were conducted with international partners, and the team plans larger, more diverse datasets and added dementia types before any deployment.