Researchers of a scoping review of 67 studies up to December 2023 mapped artificial intelligence (AI)-driven wearable technologies for epilepsy, focusing on device types, biosignals, algorithms, and validation. Research activity has surged since 2021, led by India, the United States, and China. Both commercial and prototype devices were common (each ~46%); Empatica smart bands appeared most often. Frequently monitored signals included activity (54%), cardiovascular metrics (49%), brain activity (36%), and skin conductance (34%). Popular models were support vector machines (42%), random forests (21%), and convolutional neural networks (15%), with most work centered on seizure detection (81%) rather than prediction (21%). Multimodal sensor fusion and lightweight deep models are emerging but remain early-stage.
Sensitivity was the predominant metric reported, but robust clinical validation remained limited. Video-electroencephalography was the primary reference standard (~52%), and leave-one-out or k-fold cross-validation were typical. Closed-source datasets dominated (~66%), constraining generalizability. Overall, wearable AI shows strong promise for real-time, continuous monitoring and early detection. To translate into practice, the field needs standardized validation protocols, greater open-data sharing, stronger prospective/real-world studies (especially for prediction), and energy-efficient, on-device algorithms suitable for everyday deployment. Future work should also prioritize interoperability with electronic health record/caregiver alert systems and rigorous human-factors testing to ensure usability for patients and clinicians.
Reference: Aziz S, A M Ali A, Aslam H, et al. Wearable Artificial Intelligence for Epilepsy: Scoping Review. J Med Internet Res. 2025;27:e73593. doi: 10.2196/73593.
