Detection of Tonic-Clonic Seizures by Video Analysis
Abstract number :
2.433
Submission category :
4. Clinical Epilepsy / 4B. Clinical Diagnosis
Year :
2023
Submission ID :
1319
Source :
www.aesnet.org
Presentation date :
12/3/2023 12:00:00 AM
Published date :
Authors :
Presenting Author: Satsuki Watanabe, MD, PhD – Saitama Medical University
Yuichi Tanaka, Ph.D. – Professor, Graduate School of Engineering, Osaka University; Yoshiko Murata, M.D., Ph.D. – Department of Psychiatry – Saitama Medical University; Koji Matsuo, M.D., Ph.D. – Professor, Department of Psychiatry, Saitama Medical University
Rationale:
Seizure detection system has the potential to improve treatment of patients with epilepsy or prevent sudden unexpected death in epilepsy. In previous studies, a variety of modalities such as EEG, ECG, EMG, and accelerometry were applied to detect seizures. While these methods require attachment of electrodes or sensors to patients’ bodies, video-based detection is contactless. However, the existing video-based methods have problems of accuracy and/or computational cost. We have developed a new video-based seizure detection algorithm for tonic-clonic seizures. In this study, we examined the accuracy of the seizure detection algorithm. This study was conducted with the approval of the Saitama Medical University Hospital IRB. All participants or their legal guardians provided written informed consent.
Methods:
We used fifteen videos of tonic-clonic seizures which were recorded during long-term video-EEG (VEEG) monitoring in Saitama Medical University Hospital. Video resolution was 320 x 240 pixels and 30 frames/second. The video analysis was done as follows. 1) Each frame was divided into 40 x 40 blocks and their average luminance was calculated. 2) Difference between the averages at the neighboring frames in the same spatial location was calculated. 3) These difference values were transformed to the frequency domain by a short-time Fourier transform and spectrograms were obtained. 4) If a component of 1 to 6 Hz was larger than other frequency components and this situation lasted over one second, a video was judged as containing a seizure. We evaluated seizure detection accuracy. All experiments were implemented with MATLAB R2016b, and they were run on a PC with 3.6 GHz Intel Core i7-6950X processor with and 32 GB RAM.
Results:
The fifteen seizures were recorded from eight adult patients with epilepsy. Six patients had focal epilepsy (three temporal, three frontal) and two patients had Lennox-Gastaut syndrome. All the tonic-clonic seizures were focal to bilateral tonic-clonic seizures. Sensitivity of detection of seizures was 0.93 and false positive length was 64.8 min/24 hours. Processing speed was 109-110 frames/sec for offline processing, and 30 frames/sec for online processing. Average time from onset of the clonic phase of the seizures to detection of seizures was 9.6 seconds. Mean length of the seizures was 80.5 sec and the seizures were detected 50.4 sec from a seizure onset on average.
Conclusions:
Our algorithm detected tonic-clonic seizures with high accuracy and low complexity. We conclude that it is possible to use our algorithm to alert and record tonic-clonic seizures with a real time seizure detection system. In the future study, we will test portable seizure detection devices in VEEG monitoring units and residential settings.
Funding: This study was funded by JSPS KAKENHI Grant Number JP 18K12160.
Clinical Epilepsy