Örebro University researchers develop explainable, privacy-preserving EEG AI to classify dementia
Researchers at Örebro University developed two explainable AI models that classify EEG recordings to distinguish healthy controls, Alzheimer's disease, and frontotemporal dementia. One model combines temporal convolutional networks with LSTM and achieves over 80% accuracy while highlighting EEG segments that drive decisions. A second, sub-megabyte hybrid EEGNetv4 model uses federated learning to preserve patient privacy and reported over 97% dementia-classification accuracy. Both approaches analyze alpha, beta, and gamma bands and emphasize interpretability for clinical use.
Key Points
- 1Core technical detail: The team combined temporal convolutional networks and LSTM, decomposed EEG into alpha/beta/gamma bands, and applied explainable-AI; a separate hybrid-fusion EEGNetv4 ran under 1 MB and was trained via federated learning.
- 2Business implication: High-accuracy, privacy-preserving models enable edge deployable diagnostic tools and multi-institution collaborations without raw-data sharing, reducing regulatory and infrastructure barriers for healthcare partners.
- 3Future impact: If validated in large, multi-center clinical trials, these methods could accelerate early, distributed dementia screening and drive adoption of federated, explainable clinical ML pipelines; however, wider generalization and clinical validation remain required.
Sources
Public references used for this report.
Practice interview problems based on real data
1,625 SQL & Python problems across 15 industry datasets — the exact type of data you work with.
Try 250 free problems
