Fig. 4

RMSE functions for SVM linear kernel (a, b) and SVM radial basis function (RBF) kernel (c, d) for both left and right hemispheres, depending on the global thresholds. Numbers of unique configurations for left and right damaged patients, depending on the global-threshold (e, f). The RMSE functions for the binary training are represented in black, and for the original training in dashed grey. The values of the RMSEs computed with the median-threshold binarization are reported here for comparison (Left SVM linear kernel: original-graded training RMSE = 5.5561, thresholded-binary training RMSE = 5.6454; SVM RBF kernel: original-graded training RMSE = 5.344, thresholded-binary training RMSE = 5.9467. Right SVM linear kernel: original-graded training RMSE = 5.3307, thresholded-binary training RMSE = 5.9663; SVM RBF kernel: original-graded training RMSE = 5.1841, thresholded-binary training RMSE = 5.8949)