SeizFt: Interpretable Machine Learning for Seizure Detection Using Wearables

• DOI: 10.3390/bioengineering10080918
• Publication Date: 2023-08-02T15:24:26Z
• Authors: Irfan Al-Hussaini; Cassie S. Mitchell
• Keywords: Technology; xai; seizure; EEG; augmentation; xai; interpretability; imbalanced classes; electroencephalogram; artificial intelligence; machine learning; QH301-705.5; seizure; T; augmentation; EEG; Biology (General); interpretability; imbalanced classes; Article

This work presents SeizFt-a novel seizure detection framework that utilizes machine learning to automatically detect seizures using wearable SensorDot EEG data. Inspired by interpretable sleep staging, our approach employs a unique combination of data augmentation, meaningful feature extraction, and an ensemble decision trees improve resilience variations in increase the capacity generalize unseen Fourier Transform (FT) Surrogates were utilized sample size class balance between labeled non-seizure epochs. To enhance model stability accuracy, SeizFt through CatBoost classifier classify each second recording as or non-seizure. The SeizIt1 dataset was used for training, SeizIt2 validation testing. Model performance evaluated two primary metrics: sensitivity any-overlap method (OVLP) False Alarm (FA) rate epoch-based scoring (EPOCH). Notably, placed first among array state-of-the-art algorithms part Seizure Detection Grand Challenge at 2023 International Conference on Acoustics, Speech, Signal Processing (ICASSP). outperformed black-box models accurate minimized false alarms, obtaining total score 40.15, combining OVLP EPOCH across tasks representing improvement ~30% from next best approach. interpretability is key advantage, it fosters trust accountability healthcare professionals. most predictive features extracted were: delta wave, interquartile range, standard deviation, absolute power, theta ratio theta, binned entropy, Hjorth complexity, + Higuchi fractal dimension. In conclusion, successful application suggests its potential real-time, continuous monitoring personalized medicine epilepsy.