Extended Data Fig. 10: Receiver operating characteristics for all models using repeated subsampling.
ROC curves showing the results from the first 10 subsamples as thin colored lines and the averaged and ±1 standard deviation ROC curve calculated from the distribution of the true positive rate in bins of the false positive rate across all subsamples as a blue curve surrounded by a gray band. For each model and fold result, the subset of N = 247 patients included depending on the available modalities and train–test split is indicated.
