Abstract
OBJECTIVES: This study aimed to evaluate the clinical utility of positron emission tomography (PET)-based Z-score distribution mapping (Z-map) in the noninvasive presurgical localization of seizure onset zones (SOZs), with a particular focus on regional differences in performance and methodological robustness. METHODS: We analyzed a cohort of 120 patients with drug-resistant epilepsy who underwent stereoelectroencephalography (SEEG) implantation between 2021 and 2024. Multimodal imaging data, including PET and structural MRI, were processed using FreeSurfer and 3D Slicer to reconstruct electrodes, segment gray matter, and generate Z-maps. Hypometabolic regions were defined as the bottom 0.5% of Z-scores. The classification performance of the Z-map was validated against SEEG-defined SOZs. RESULTS: The Z-map showed high classification performance in the frontal and parietal lobes (sensitivity: 0.74/0.72; κ: 0.68/0.65), but reduced effectiveness in the temporal and insular lobes (sensitivity: 0.41/0.65; κ: 0.44/0.47). The overall specificity (0.94) and negative predictive value (0.91) indicated strong exclusionary capability. Regional disparities were primarily attributed to anatomical complexity and technical limitations. CONCLUSION: Z-map offers clinically valuable support for localizing SOZs in the frontal and parietal cortex and optimizing SEEG implantation, particularly in patients with inconclusive noninvasive findings. However, its limited sensitivity in temporal and insular regions and high dependency on preprocessing quality underscore the need for standardized pipelines and multimodal integration. Future research should focus on improving robustness and reproducibility to facilitate clinical translation.