Fig. 4. Distribution of acute graft-versus-host disease (aGVHD) predictive scores in the test cohort.
Distribution of aGVHD prediction scores calculated by the convolutional neural network (CNN) model are displayed among patients in the test cohort (N = 3753). A For grade II–IV aGVHD. Low (scores, 0.136–0.209), n = 375 (10.0%); Int (scores, 0.209–0.770), n = 3003 (80.0%); and High (scores, 0.770–0.894), n = 375 (10.0%). B For grade III–IV aGVHD. Low (scores, 0.080–0.138), n = 375 (10.0%); Int (scores, 0.138–0.840), n = 3003 (80.0%); and High (scores, 0840–0.951), n = 375 (10.0%). Higher scores indicate a higher risk of developing aGVHD. The data used to plot the graphs is in Supplemental Data 7.