Enhanced Non-EEG Multimodal Seizure Detection: A Real-World Model for Identifying Generalised Seizures Across the Ictal State.

Non-electroencephalogram seizure detection models hold promise for the early detection of generalised onset seizures. However, these models often experience high false alarm rates and difficulties in distinguishing normal movements from seizure manifestations. To address this, we were granted exclusive access to the newly developed Open Seizure Database, from which a representative dataset of 94 events was selected (42 generalised tonic-clonic seizures, 19 auras/focal seizures, and 33 seizures labelled as Other), with a combined duration of approximately 5 hours and 29 minutes. Each event contains acceleration and heart rate data which have been expertly annotated by a clinician, who labelled each 5-second timestep as Normal, Pre-Ictal, or Ictal. We then introduced the AMBER (Attention-guided Multi-Branching-pipeline with Enhanced Residual fusion) model. AMBER constructs multiple branches to form independent feature extraction pipelines for each sensing modality. The outputs of each branch are passed to our Residual Fusion layer, where the extracted features are combined into a fused representation and propagated through two densely connected blocks. The dataset was split by event, ensuring no overlap between events in the training and testing subsets. The model was trained using -fold cross-validation, where -1 folds were used for training and the remaining fold for validation. The results of these experiments highlight the effectiveness of Ictal-Phase Detection, with the model achieving an accuracy and -score of 0.9027 and 0.9035, respectively, on unseen test data. Notably, the model exhibited consistent generalisation, recording a True Positive Rate of 0.8342, 0.9485 and 0.9118 for the Normal, Pre-Ictal, and Ictal classes respectively, and an average False Positive Rate of 0.0502. In conclusion, this study introduces a new multimodal seizure detection technique and model that reduces the false alarm window and differentiates high and low-amplitude convulsive movements, laying the groundwork for further advancements in non-EEG-based seizure detection research.