Presurgical structural connectivity predicts postsurgical cognitive impairment in glioma patients.

Glioma patients frequently suffer from cognitive impairments after surgery, but predicting these impairments preoperatively at an individual level remains challenging. Cognitive functions are increasingly studied from a network perspective, where an important role is played by the Default Mode Network (DMN) and Frontoparietal Network (FPN). Hypothesizing that postsurgical cognitive impairments arise from structural network vulnerabilities, we trained models using presurgical structural connectivity of DMN and FPN regions to predict postsurgical cognitive impairment. We obtained individualized structural connectomes in 63 glioma patients (grades II-IV) who underwent diffusion-weighted MRI before surgery (T0) and neuropsychological screening 3 months after surgery (T3) and, for a small majority, adjuvant treatment. Random forest classifiers were trained on a combination of baseline (sociodemographic and clinical), tumour location and structural network variables available before surgery to predict postsurgical cognitive impairment in individual patients. Classifier performance was measured as area under curve of the receiver operating characteristic (AUC-ROC), testing statistical significance via permutation testing. Predictor importance was calculated post-hoc using Shapley additive explanations for trees. Postsurgical impairment was predicted by baseline variables available at T0 (AUC = 0.69,