SeizAI: A Secure AI-Based Seizure Detection via Homomorphic EEG Encryption

Epileptic seizures, triggered by irregular electrical activity in the brain, pose serious health challenges for millions worldwide. Accurate, secure and timely detection of seizures is vital to improve patient outcomes and quality of life. This work presents a secure, real-time seizure detection framework driven by artificial intelligence (AI) utilizing machine learning algorithms and homomorphic encryption (HE). This framework ensures data privacy and compliance with healthcare regulatory requirements while enabling computational analysis directly on encrypted data. To address class imbalance and improve model performance, methods including fixed-length signal segmentation and sliding window approaches were applied. The widely used CHB-MIT and Bonn EEG datasets were used for validation. Among the models tested, XGBoost achieved the best results, with an average accuracy of 90%, a precision of 87%, and a sensitivity of 92%. The proposed approach underscores the potential of integrating machine learning with encryption technologies to develop secure and accurate seizure detection systems, paving the way for innovative and privacy-preserving solutions in healthcare.