Fig. 5: Predicting response to DBT using a machine learning framework.

A Schematic of the machine learning pipeline for the predictive model of the patient response to the DBT. The pipeline includes seven parts as feature selection, feature combination, model benchmarking, model selection, parameter tuning, model refitting, and model evaluation. B The different evaluation metrics including the area under the receiver operating characteristic curve (AUROC), accuracy, precision, recall, F1, macro and weighted scores for models with different feature combinations. The metrics were evaluated by the different models that were refitted on the 100 different cohort splits. The data was represented by the mean value and the color band indicates the 95% confidence interval. C The feature importance for the model with the integrated all features. The data was represented by the mean value and the error bar indicates the 95% confidence interval (repeats n = 100). D The confusion matrix of the models with different feature combination on the independent validation cohort. E The bar plot depicting the model performance on the independent validation cohort. The different evaluation metrics were used for the evaluation including AUROC, accuracy, precision, recall, F1, macro and weighted scores. Source data are provided as a Source Data file.