VIDEO BASED DETECTION OF EPILEPTIC SEIZURES USING A THREE-DIMENSIONAL CONVOLUTIONAL NEURAL NETWORK

Seizure detection in epilepsy monitoring units (EMU) is essential for the clinical assessment of drug-resistant epilepsy. Automated video analysis using machine learning provides a promising aid for seizure detection with resultant reduction in the resources required for diagnostic monitoring. We employ a 3D convolutional neural network with fully fine-tuned backbone layers to identify seizures from EMU videos. A two-stream inflated 3D-ConvNet architecture (I3D) classified video clips as a seizure or not a seizure. A pretrained action classification model was fine-tuned on 11 hours of video data containing 49 tonic-clonic seizures from 25 patients monitored at a large academic hospital (site A) using leave-one-patient-out cross-validation. Performance was evaluated by comparing model predictions to ground-truth annotations obtained from video-EEG review by an epileptologist on videos from site A and a separate dataset from a second large academic hospital (site B). The model achieved leave-one subject out cross-validation F1-score of 0.960 ± 0.007 and area under the receiver operating curve (AUC) score of 0.988 ± 0.004 at site A. Evaluation on full videos successfully detected all seizures with median detection latency of 0.0 (0.0, 3.0) seconds from seizure onset. The site A model had an average false alarm rate of 1.81 alarms per hour, though 33 of the 49 videos (67%) had no false alarms. Evaluation at site B demonstrated generalizability of the model architecture and training strategy, though cross-site evaluation (site A model tested on site B data and vice versa) resulted in diminished performance. Our model demonstrates high performance in the detection of epileptic seizures from video data using a fine-tuned I3D model and outperforms prior similar models identified in the literature. This study provides a foundation for future work in real-time EMU seizure monitoring and possibly for reliable and cost-effective at-home detection of tonic-clonic seizures. We evaluate a video-based 3-D CNN for seizure detection in patients undergoing evaluation in an EMU at 2 large academic hospitals. Our video-only model provides highly accurate detection of tonic-clonic seizures with low detection latency. The underlying model architecture requires no video preprocessing and is generalizable across two EMUs.