We propose a Koopman-inspired dynamic graph (KDG) framework for SOZ localization from iEEG, which models the temporal evolution of FC graphs using an operator-based dynamical representation.
hup_koopman_baselines/contains implementations of the baseline methods used for comparison in the manuscript. These methods were reproduced according to the methodological descriptions provided in the corresponding publications.compute_operator.ipynbimplements the core KDG pipeline. It constructs dynamic functional connectivity (FC) graphs from iEEG recordings, models the temporal evolution of these graphs using a Koopman-inspired operator, and applies singular value decomposition (SVD) to the learned operators to extract their singular vectors.graph_features_calculation.pyderives four graph-theoretic measures from the singular-vector representations: node strength, degree standard deviation, eigenvector centrality, and global efficiency.metric_loso_soz_classification.pyevaluates the ability of these graph-theoretic features to localize the seizure onset zone (SOZ) using leave-one-subject-out (LOSO) cross-validation and records the corresponding localization performance.