Preoperative positron emission tomography for predicting postoperative epilepsy risk: A radiomics and clinical...

Epilepsy is a common neurological disorder characterized by recurrent seizures, and predicting the risk of postoperative epilepsy is critical for patient management. The integration of radiomics and clinical factors offers a promising approach for accurate risk assessment.

In this study, preoperative positron emission tomography (PET) imaging and clinical data of 197 epilepsy patients were analyzed. The imaging features were extracted from three-dimensional regions of interest (ROIs). Key radiomics features were selected using the SelectKBest and the Least Absolute Shrinkage and Selection Operator (LASSO), while key clinical features were identified through univariate and multivariate analyses. Three models-radiomics, clinical, and fusion (fusing radiomics and clinical features)-were constructed using the Support Vector Machine (SVM) and Multilayer Perceptron (MLP) algorithms. The models were evaluated based on sensitivity, specificity, and area under the curve (AUC). A nomogram based on the best-performing model was developed and assessed for predictive accuracy using calibration and decision curve analysis (DCA) to estimate the risk of postoperative epilepsy following neurosurgery.

A total of 11 key radiomics features were identified, with prior brain damage (PBD) as the key clinical feature. The MLP fusion model demonstrated the highest predictive performance, with an AUC of 0.85 in the validation set. The nomogram showed good predictive accuracy and superior clinical utility.

The integration of radiomics and clinical features provides a more accurate and clinically useful tool for predicting postoperative epilepsy risk. The fusion model demonstrates significant potential in guiding clinical decision-making and personalized treatment for epilepsy patients.