Understanding Patients' Behavior: Vision-Based Analysis of Seizure Disorders

Area of Science:

• Neurology
• Biomedical Engineering
• Computer Science

Background:

• Functional neurological disorders (FND) are frequently misdiagnosed as epileptic seizures (ES) due to overlapping semiology.
• Diagnostic errors can lead to inappropriate treatments and complications.
• Current diagnostic methods struggle with differentiating FND and ES, especially when electrophysiological changes are absent in FND or present in some epilepsy types.

Purpose of the Study:

• To develop and compare marker-free deep learning models for distinguishing between FND and ES using video recordings.
• To quantify semiology through advanced video analysis techniques.
• To assess the efficacy of computer vision in a clinical setting for seizure disorder assessment.

Main Methods:

• Proposed two marker-free deep learning models: a landmark-based and a region-based approach.
• Quantified semiology using a fusion of reference points and flow fields, or complete body analysis.
• Utilized video recordings from 35 patients and employed leave-one-subject-out cross-validation.

Main Results:

• The region-based deep learning model achieved a higher average accuracy of 79.6% compared to the landmark-based model's 68.1%.
• Demonstrated the potential of video analytics for automated semiology identification in challenging hospital environments.
• Highlighted the benefit of computer vision in overcoming limitations of traditional sensor-based systems.

Conclusions:

• Marker-free deep learning models show promise in differentiating functional neurological disorders from epileptic seizures based on video analysis.
• Automated video analysis can support clinical diagnosis, reducing misdiagnosis rates and improving patient care.
• This approach offers a non-invasive and robust method for seizure disorder assessment in clinical settings.