Signal Processing Techniques for the Localization of Seizure-Onset Zones in the Brain

This paper presents a streamlined yet comprehensive approach to localizing seizure onset zones (SOZs) in drug-resistant epilepsy patients, using electroencephalographic (EEG) data. Bandpass filtering, targeting specific frequency bands including delta (1-4 Hz), theta (4-8 Hz), alpha (8-13 Hz), beta (13-30 Hz), gamma (30-100 Hz), and the high-frequency oscillations band (100-250 Hz), was applied before meticulous feature extraction. The extracted features encompass spectral features such as power, total energy, mean energy, percentile 75 energy, Hjorth mobility, Hjorth complexity, and Teager-Kaiser energy operator and entropy-related features such as spectral entropy, approximate entropy, sample entropy, singular value decomposition entropy and power spectral entropy. Through this method, statistically significant biomarkers indicative of epileptic activity are identified, aiming to improve precision in SOZ localization for non-invasive EEG and potentially reduce the need for invasive procedures, thus enhancing patient surgical experiences. This approach underscores the importance of early diagnosis and intervention in epilepsy management, highlighting the potential of computational methodologies to enhance diagnostic accuracy and treatment efficacy. Our findings demonstrate consistent disparities between SOZ and non-SOZ channels across multiple recording sessions of the same patient, indicating the utility of these spectral and entropy-related features. Our analytical framework, adaptable to each patient in the dataset, offers valuable insights into the spatiotemporal patterns of epileptogenic activity, paving the way for more personalized treatment strategies. Moreover, modern artificial intelligence (AI) techniques offer opportunities for learning complex decision functions, promising improved SOZ classification. By integrating various relevant biomarkers, these AI models can enhance the accuracy and reliability of SOZ localization. Future research will focus on expanding the analysis to include a larger cohort of patients to enhance generalizability and explore the integration of machine learning algorithms to further improve localization accuracy. It is imperative to acknowledge the limited spatial resolution inherent in EEG recordings compared to intracranial electroencephalographic (iEEG) recordings. This disparity underscores the necessity of identifying relevant biomarkers within EEG data to facilitate precise localization of SOZs, particularly considering the discomfort and risk associated with iEEG procedures for patients.