Robust Dynamical Component Analysis and Initial Results of Its Application to EEG Data of Epileptic Seizures

Understanding and analyzing complex time series data is a fundamental challenge in various scientific disciplines, especially in neuroscience. Identifiying the underlying deterministic structures in electroencephalographic (EEG) recordings is crucial for accurate diagnosis and effective treatment. However, real EEG data is often distorted by noise, missing values and artifacts, making accurate seizure detection difficult.We address these challenges by leveraging Robust Dynamical Component Analysis (DyCA) - a robust extension of DyCA - to reconstruct EEG signals and extract eigenvalue based seizure indicators. In contrast to traditional and more widely used methods such as principal component analysis (PCA), Robust DyCA effectively reduces noise while preserving essential seizure-related features. Our approach was validated on EEG data from the Temple University Hospital Seizure Corpus (TUSZ) dataset, where a sliding window analysis of DyCA eigenvalues showed an increase during seizure events. While this suggests the potential of DyCA-based features for AI-assisted seizure detection, further studies on additional datasets are necessary to confirm their reliability and generalizability in clinical applications.