Abstract
OBJECTIVE: Epilepsy is one of the most serious brain disorders which can be treated with antiepileptic drugs. However, for those with refractory focal epilepsy, surgical resection of the epileptogenic zones (EZ) is the gold standard for achieving long-term seizure-free status. The aim of this study was to construct a predictive model to localize EZs from SPECT images using radiomics. METHODS: Twenty sets of ictal and interictal SPECT images were collected retrospectively. Image preprocessing, including normalization and registration of the SPECT data and the calculation of z-score images, was performed using statistical parametric mapping software (SPM12). In this study, we extracted radiomic features from ictal images and z-score images using two methods: the voxel-based and the map-based methods. Subsequently, six radiomic models (two extraction methods for each of the ictal, the z-score, and the combined ictal/z-score models) were constructed, and their performances were compared to nuclear medicine physician readings. RESULTS: The voxel-based combined model achieved the highest sensitivity (0.954 ± 0.044) and AUC value (0.918 ± 0.044), followed by the map-based combined model (AUC = 0.895 ± 0.065). The voxel-based ictal model achieved the third-highest AUC value (0.848 ± 0.052) and the highest specificity (0.848 ± 0.052). In comparison, physician readings had a sensitivity of 0.679 ± 0.277 and a specificity of 0.980 ± 0.007. CONCLUSION: Radiomic analysis of SPECT images showed a promising ability to improve the prediction of EZ locations in epilepsy patients. These findings suggested that radiomic models can effectively support physicians in clinical decision making. SUPPLEMENTARY INFORMATION: The online version contains supplementary material available at 10.1007/s13139-025-00921-5.