EEG Derivation Optimization
Year: 2025
Type: software
Abstract: Electroencephalography (EEG) plays a central role in automated epileptic seizure detection and is increasingly being incorporated into wearable neurological monitoring systems. However, conventional EEG-based approaches typically rely on dense electrode montages and high-dimensional feature representations, limiting their feasibility for portable and real-time healthcare applications. This study proposes a two-stage optimization pipeline that simultaneously reduces the EEG feature space and acquisition configuration while preserving seizure detection performance. The methodology was evaluated using the publicly available CHB-MIT Scalp EEG Database. Initially, 270 wavelet-based features were extracted from 18 EEG derivations. Redundant descriptors were removed through Pearson correlation analysis, Recursive Feature Elimination, and Sequential Feature Selection, resulting in a compact set of 10 representative features. These features were subsequently mapped to their corresponding EEG derivations, enabling the identification of an optimized acquisition configuration comprising four derivations and eight scalp electrodes. Eight supervised machine learning algorithms were evaluated for classifying ictal and non-ictal EEG segments. The optimized configuration achieved a maximum classification accuracy of 94.74%, with the Random Forest classifier providing the most balanced performance in terms of accuracy, precision, sensitivity, specificity, F1-score, and area under the receiver operating characteristic curve. The proposed methodology substantially reduces EEG acquisition complexity without compromising classification performance, providing a practical strategy for the development of compact, computationally efficient, and wearable seizure monitoring systems suitable for future healthcare technologies.
Language: English
Subfield: Cognitive Neuroscience
Field: Neuroscience
Domain: Life Sciences
Open Access status: green
OpenAlex ID: W7171941484