Fig. 3

Confidence score of correct and incorrect predictions. A histogram of confidence scores for correct and incorrect protocol predictions. Confidence score obtained from model output on scale of 0–1. a For spine MRI prediction, confidence score distribution of incorrect predictions differed significantly from correct predictions (Kolmogorov–Smirnov test, p < 0.001). Predictions were skewed towards high confidence, while incorrect predictions were more uniformly distributed. b For brain MRI predictions, confidence score distribution of incorrect predictions differed significantly from correct predictions (Kolmogorov–Smirnov test, p < 0.001). Correct predictions were skewed towards high confidence