Πλοήγηση ανά Συγγραφέας "Mouzakitis, Nikolaos"

Τώρα δείχνει 1 - 1 of 1
  • Μικρογραφία εικόνας
    Τεκμήριο
    Epilepsy detection from multi-channel EEG using cross-recurrence quantification analysis and machine learning
    (ΕΛΜΕΠΑ, Σχολή Μηχανικών (ΣΜΗΧ), ΠΜΣ Μηχανικών Πληροφορικής, 2026-06-04) Mouzakitis, Nikolaos; Μουζακίτης, Νικόλαος; Tsiknakis, Emmanouil; Τσικνάκης, Εμμανουήλ
    Epileptic seizures originate from abnormal, large scale synchronization of neuronal populations in the human brain, making the analysis of inter regional brain dynamics essential for reliable detection. This thesis presents a nonlinear dynamical framework for automated seizure classification utilizing multichannel scalp electroencephalography (EEG). Cross-Recurrence Quantification Analysis (CRQA) is employed to characterize the interactions between EEG channels in their reconstructed phase space, enabling the quantification of the coupling patterns related to the seizure activity. EEG recordings are preprocessed by a denoising and normalization pipeline before being segmented into windows of fixed-length. For each window, CRQA metrics are calculated across all channel pairs and aggregated to form a compact representation of brain’s dynamical interaction. These features are used to train and evaluate traditional machine learning classifiers, under multiple validation schemes, emphasizing on subjectindependent protocols for ensuring a realistic performance assessment. The methodology is evaluated on a publicly available scalp EEG dataset which contains annotations of the seizure events. Results show that the recurrence-based interaction hold the ability to capture informative seizure related dynamics and support the discrimination among epileptic and non-epileptic EEG windows. The findings, also point out the value of nonlinear inter-channel analysis as a complementary approach to traditional single-channel or spectral methods in this domain for analysis.