Figure 7. .
Prediction of surgical outcome using a machine learning algorithm (RUSBoost) with leave-one-out cross validation. As input variables we used the normalized decrease in seizure propagation after virtual resection of the RA, δIR(RA), the size of optimal resections S(Rop), and the overlap of optimal and clinical resections Ov(Rop, RA). Panels A, B show the confusion matrix and predictor importance for the validation cohort (N = 34, 8 NSF), and panels C, D are for the combined cohort (N = 49, 12 NSF).