$n$-Cylindrical Symbolic Response, a Standalone and Synergistic Biomarker for Epilepsy Diagnosis on EEG Modality

Misdiagnosis of Epilepsy in an interictal state is a serious global concern. Erroneously prescribing anti-epileptic drugs (AEDs) to healthy individuals can eventually induce epilepsy. Additionally, challenges such as non-stationarity, homogeneity, and heterogeneity are prevalent within and across the subjects while collecting electroencephalogram (EEG) data for Epilepsy diagnosis and ictal-interictal classification. Another limitation is the variability of the steps in the preprocessing pipeline, such as filtering and normalization, manifesting differently across subjects, within-subjects, and across different datasets and classes. This is attributed to artifact inclusion. To address these research gaps, we propose a novel data acquisition pipeline and an EEG dataset of 140 subjects (70 each of healthy control and an interictal epileptic class), as per International League Against Epilepsy (ILAE) guidelines. We propose a computational biomarker namely, $n$-cylindrical-based symbolic response ($n$CSR) vector that contributes features similar to Interictal Epileptiform Discharges (IEDs), and are robust to preprocessing. This biomarker uses eigenvalues, eigenvectors, and eigen directions along with the fusion of features of FFT and DWT to address the issue of homogeneity between epileptic patients (interictal state) and healthy controls, serving both as a standalone and synergistic feature. The proposed approach demonstrated an improvement of 38.54% and 40.29% test precision over the Siena and BIOMED datasets, respectively, compared to the combination of baseline feature fusion. Furthermore, owing to the simplicity and synergy, the proposed method can be seamlessly integrated into the clinical setting to bridge the gap between AI and cognitive neuroscience for translation towards clinical practice applications.