A Study of Detection of Paroxysmal Events Utilizing Computer Vision and Machine Learning - Nelli
A Study of Detection of Paroxysmal Events Utilizing Computer Vision and Machine Learning - Nelli
Nelli is a video-based non-EEG physiological seizure monitoring system. This study is a blinded comparison of Nelli's identified events to gold-standard video EEG review in at-rest pediatric subjects with suspected motor seizures.
Automated analysis of video recordings to detect seizures, assisted by modern methods of machine learning, holds great promise to address this issue. Increased computational power has made it possible to implement complex image recognition tasks and machine learning in everyday use. Nelli® software is designed to use computer vision and machine learning-based algorithms to automatically detect seizure events. This study will provide evidence that Nelli software can identify seizure events and deliver objective data to clinicians for evaluation of seizure management.
This study is being conducted to validate the Nelli Software's ability to identify periods of audio
/video data that contain recordings of patients experiencing seizures (or seizure-like events) during periods of rest. The software's performance will be compared to the gold standard, expert review of video EEG data.
Nelli Software will review the audio and video data and independently identify events with positive motor manifestations. The outcomes of event identification will be compared between epileptologists and the Nelli Software. For each category of event captured the positive percent agreement will be calculated using the exact binomial method. The primary endpoint of this study is to demonstrate that Nelli is able to identify seizures that have a positive motor component with a sensitivity of >70% (lower 95% CI) and with a false discovery rate (FDR) comparable to similar devices on the market.
Inclusion Criteria:
Exclusion Criteria: