Fig. 4

Confusion matrices of six radiologists performing hierarchical diagnosis and binary classification with and without PTs-HDM assistance. a) Binary Classification: Each matrix compares the performance of the same six radiologists for differentiating between fibroadenomas and phyllodes tumors. b) Hierarchical Diagnosis: Each matrix represents the distribution of predictions for FAs, benign PTs, and borderline/malignant PTs across six radiologists. Rows indicate the actual labels, and columns indicate the predicted labels. Across both binary classification and hierarchical diagnosis tasks, PTs-HDM assistance improved diagnostic accuracy, reducing misclassification rates and increasing consistency, especially for borderline/malignant cases (PTs-M) This effect was more pronounced for residents compared to seniors and attendings, reflecting the potential of PTs-HDM to augment less experienced radiologists. FAs, fibroadenomas; PTs, phyllodes tumors; -B, Benign; -M, Borderline/Malignant; +, with PTs-HDM assistance