THREE-DIMENSIONAL CONVOLUTIONAL NEURAL NETWORK BASED DETECTION OF EPILEPTIC SEIZURES FROM VIDEO DATA

Objective

Seizure detection in epilepsy monitoring units (EMU) is essential for successful seizure characterization and 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 novel approach, using a 3D convolutional neural network to identify seizures from EMU videos.

Methods

A two-stream inflated 3D-ConvNet architecture 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 on a separate dataset from a second large academic hospital (site B).

Results

The model classified previously unseen videos from site A with a mean accuracy of 0.944 ± 0.064 and area under the receiver operating curve (AUC) score of 0.991 ± 0.015. The mean latency between seizure onset and seizure detection was 0.61 ± 2.01 seconds, which surpasses the prior state-of-the-art. A mean accuracy of 0.964 ± 0.034 and AUC score of 0.984 ± 0.049 were achieved by the model trained and tested at site B. Cross-site evaluation (site A model tested on site B data and vice versa) demonstrated generalizability of the model with a mean decrease in AUC score of −0.156 from same-site evaluation.

Significance

Our model demonstrates high performance in detection of epileptic seizures from video data using spatiotemporal 3D-CNN models 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.

KEY POINTS

• 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.