Fig. 4

Comparison of performance between pathologists’ consensus calls and model produced calls. a Survival analyses for pathologist’s consensus calls of Ki67 status. b Survival analyses for model produced calls of Ki67 status. c Survival analyses for pathologist’s consensus calls of NHG. d Survival analyses for model produced calls of NHG. For Ki67, the calls from the model had better performance in survival analyses than that of pathologist’s consensus calls. For NHG, the performance of model produced calls only showed a trend, this was likely due to the much smaller sample size (N = 59 as compared to N = 2917 from the pathologist’s calls, see Table 1)