Patient-specific Video-based Detection of Epileptic Seizures

2 de dezembro de 2025

Epilepsy is a neurological disease characterized by recurrent seizures, affecting more than 50 million individuals worldwide. For those at risk of severe motor seizures—such as tonic–clonic events—timely detection is crucial to enable rapid caregiver intervention and reduce the risk of complications, including physical injury and Sudden Unexpected Death in Epilepsy (SUDEP). In this context, detection models that achieve low latency together with near-zero false alarm rates can provide substantial clinical value, even with moderate sensitivity, particularly when used alongside more sensitive but slower monitoring systems. This work presents a video-based framework for detecting tonic–clonic seizures using only RGB recordings acquired under real clinical monitoring conditions. To extract discriminative motion features from raw video, a human pose estimation (HPE) pipeline was developed and evaluated on data from two Epilepsy Monitoring Units. A custom annotated dataset was created to fine-tune state-of-the-art HPE models, revealing that architecture choice and dataset homogeneity are essential for reliable keypoint localization in challenging in-bed environments. A transformer-based model (PVT v2) achieved the strongest performance and served as the foundation for downstream analysis. Using spatial, temporal, and frequency features derived from pose sequences, a gradient-boosted classifier was trained to identify seizure events. The proposed system achieved zero false alarm rates and average detection latencies below 33 s, outperforming comparable unimodal video-based approaches. These findings highlight the potential of pose-based video analysis as a privacy-preserving, interpretable, and clinically viable solution for continuous seizure monitoring.

Autores da comunidade :

• 

  M

  Mariana Madruga Bernardino

  ist195956

Orientadores desta instituição:

• 

  V

  Vicente Mayer Mendonça de Lima Garção

  ist186810

• 

  H

  Hugo Humberto Plácido da Silva

  ist46129

industrial-biotechnology - Biotecnologia Industrial

eng - Inglês

Instituto Superior Técnico