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 was expertly annotated by a clinician, who labelled every 5 second timestep with a class of 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 custom Enhanced Residual Fusion layer, where the extracted features are combined into a fused representation. The fused representation is then propagated through two densely connected blocks before being passed through a softmax activation function. The model was trained using k-fold cross validation, with k-1 fold used to train the model and the remaining fold was used to evaluate the model’s performance. The results of these experiments underscore the efficacy of Ictal-Phase Detection, achieving an accuracy and f1-score of 0.8995 and 0.8987. Notably, the model exhibited consistent generalisation, recording a True Positive Rate of 0.9564, 0.8325, and 0.9111 for the Normal, Pre-Ictal, and Ictal classes respectively. These findings were compounded by an average False Positive Rate, recording an overall score of 0.0502. In conclusion, this research introduces a new detection technique and model designed for multimodal seizure detection, with the potential to reduce the false alarm window and differentiate high and low amplitude convulsive movement. We believe the results of this study lay the groundwork for further advancements in non-electroencephalogram seizure detection research.

This research introduces the Open Seizure Database and Toolkit as a novel, publicly accessible resource designed to advance non-electroencephalogram seizure detection research. This paper highlights the scarcity of resources in the non-electroencephalogram domain and establishes the Open Seizure Database as the first openly accessible database containing multimodal sensor data from 49 participants in real-world, in-home environments. The database is comprised of 494 events, encompassing 146 epileptic seizures, collected over a duration of 453 days, presenting the most extensive publicly available non-electroencephalogram seizure data to date. Additionally, the database has 348 labelled false alarms, including 302 common human movements and activities. The Open Seizure Toolkit is designed to facilitate machine and deep learning practices by streamlining data from the Open Seizure Database. Utilising these resources, researchers can rapidly develop and train seizure detection models before deploying them to the Open Seizure Detector Android application. Access to these resources is expected to foster collaborative efforts, ultimately contributing to the establishment of a non-electroencephalogram gold standard and advancing the field of seizure detection. Please view the OSDB at https://ieee-dataport.org/documents/open-seizure-database-v100