A PET-based radiomics nomogram for individualized predictions of seizure outcomes after temporal lobe epilepsy surgery.

• Metabolic PET radiomics features were strong predictors of seizure recurrence. • PET-based radiomics score, SGS, and Durmon were significant predictors of seizure-free outcomes. • A nomogram incorporating PET-based radiomics and clinical risk factors demonstrated good performance in predicting surgery outcomes. To establish and validate a novel nomogram based on clinical characteristics and [¹⁸F]FDG PET radiomics for the prediction of postsurgical seizure freedom in patients with temporal lobe epilepsy (TLE). 234 patients with drug-refractory TLE patients were included with a median follow-up time of 24 months after surgery. The correlation coefficient redundancy analysis and LASSO Cox regression were used to characterize risk factors. The Cox model was conducted to develop a Clinic-PET nomogram to predict the relapse status in the training set (n = 171). The nomogram's performance was estimated through discrimination, calibration, and clinical utility. The prognostic prediction model was validated in the test set (n = 63). Eight radiomics features were selected to assess the radiomics score (radscore) of the operation side (Lat_radscore) and the asymmetric index (AI) of the radiomics score (AI_radscore). AI_radscor, Lat_radscor, secondarily generalized seizures (SGS), and duration between seizure onset and surgery (Durmon) were significant predictors of seizure-free outcomes. The final model had a C-index of 0.68 (95 %CI: 0.59–0.77) for complete freedom from seizures and time-dependent AUROC was 0.65 at 12 months, 0.65 at 36 months, and 0.59 at 60 months in the test set. A web application derived from the primary predictive model was displayed for economic and efficient use. A PET-based radiomics nomogram is clinically promising for predicting seizure outcomes after temporal lobe epilepsy surgery. [ABSTRACT FROM AUTHOR]