Optimized seizure detection leveraging band-specific insights from limited EEG channels

Effective seizure detection systems are crucial for health information systems and managing epilepsy, yet traditional multichannel EEG devices can be costly and complex. This study aims to optimize EEG channel selection and focus on specific frequency bands associated with epileptic activity, enhancing the system's usability and accuracy for clinical applications.

This work proposes a novel method by integrating channel selection with band-wise analysis for seizure detection. The channel selection uses an ensemble of mutual information (MI) and Random Forest (RF) techniques to select the most relevant channels. The signals from the selected channels are decomposed into different frequency bands using discrete wavelet transform (DWT). To evaluate the effectiveness of this approach, ten features are extracted from each frequency band and then classified using a support vector machine (SVM) classifier.

This work has obtained a mean accuracy of 97.70%, a mean sensitivity of 86.70%, and a mean specificity of 99.66% for seizure patients from a well-established CHB-MIT dataset and an almost 80% reduction in processing time.

These benefits make seizure detection devices more wearable, less intrusive, and easier to integrate with other health monitoring systems, allowing for discreet and comfortable monitoring that supports an active lifestyle for patients.