The research presents a quantitative framework to evaluate the spatial concordance between clinically defined SOZs and statistically anomalous channels detected via time-frequency analysis of chirp events. The pipeline consists of two main steps: first, an unsupervised outlier detection using LOF with adaptive neighborhood selection identifies anomalous channels based on spectro-temporal features such as onset frequency, offset frequency, and temporal duration of chirps. Second, spatial correlation analysis computes exact co-occurrence metrics and weighted index similarity, which accounts for hemispheric congruence and electrode proximity to improve localization accuracy. The study finds that the LOF approach with parameters of 20 neighbors and 0.2 contamination effectively detects outliers. Weighted index matching that incorporates channel proximity outperforms exact matching in localizing SOZs. Performance metrics including precision, recall, and F1 score are highest in seizure-free patients and those with successful surgical outcomes, with mean index precision values of 0.903 and 0.865 respectively. Conversely, patients with surgical failures show significantly lower concordance, with a mean index precision of 0.460. These results suggest that chirp-based outlier detection combined with spatial weighting provides a valuable complementary tool for SOZ localization. This approach could enhance clinical decision-making by improving the identification of epileptogenic zones, particularly in patients likely to benefit from surgery. Future work may focus on refining the methodology and validating it in larger, diverse patient cohorts to further establish its clinical utility.
👉 Pročitaj original: arXiv AI Papers
