Adaptive heart rate-based epileptic seizure detection using real-time user feedback.

Area of Science:

• Biomedical Engineering
• Neurology
• Signal Processing

Background:

• Automated seizure detection in home environments is increasingly important for managing epilepsy.
• Existing heart rate-based seizure detection methods often lack accuracy due to patient-specific variations.
• Patient-independent classifiers are insufficient, while patient-specific ones require extensive data.

Purpose of the Study:

• To develop a real-time adaptive algorithm for seizure detection that overcomes limitations of current methods.
• To create a system that leverages patient-specific data without requiring large initial datasets.
• To improve the accuracy and reduce false alarms in home-based epilepsy monitoring.

Main Methods:

• An algorithm was developed that begins with a patient-independent classifier and adapts in real-time using user feedback.
• Support vector machine classifiers were updated immediately based on annotated alarms.
• A mechanism was included to handle potentially incorrect feedback by excluding seizure data that generate excessive false alarms.

Main Results:

• The adaptive classifier achieved 77.12% sensitivity with 1.24 false alarms per hour over 2833 hours of data from 19 patients.
• This represents approximately 30% fewer false alarms compared to patient-independent classifiers at similar sensitivity levels.
• The system was evaluated on data including 153 clinical seizures.

Conclusions:

• The proposed low-complexity adaptive algorithm effectively handles incorrect user feedback, making it suitable for seizure warning systems.
• This adaptive approach offers a promising solution for more accurate and reliable home-based seizure detection.
• Future work will involve integrating complementary modalities to further enhance the algorithm's performance.