Summary
This project aimed at a sensitive and specific seizure detection with properties which allow an implementation in a closed-loop device to treat human epilepsy. In collaboration between the epilepsy center, informatics and microsystems technology, the following major milestones were achieved:
- Implementation of a large EEG-derived feature set to extract key biomarkers of ictal epileptic activity
- Application of machine learning classifiers on long-term intracranial EEG time series, including support vector machines and random forest classifiers
- Successful early seizure detection within a period of 3-4 seconds after the emergence of an ictal discharge as determined by intracranial EEG recordings in the area of the seizure onset zone
- Markedly improved specificity of detections as compared to the state of the art as implemented in the Neuropace device
- Identification of seizure patterns representing different seizure types within the first 5 seconds of an emerging seizure pattern, allowing for a seizure-type dependent interventional strategy
- Implementation of deep learning strategies for the identification of ictal EEG segments
- Realization of seizure detection on a microprocessor with low energy demands suitable for an intracranial implant in the human.
These results allow for an implementation of significantly improved seizure detection strategies in a closed-loop setting, both at a hospital setting and integrated into a medical device to automatically trigger warnings or interventions in people with epilepsy.