Fig. 4

Performance of the ensemble neural net for APL image-level recognition. Performance metrics for the binary classification of APL vs. healthy bone marrow donor samples (top row) and APL vs. non-APL AML samples (bottom row) were calculated as areas under the curve for precision-recall curves (A, C) and the receiver operating characteristic (B, D) using threefold cross-validation (cv 0, 1, 2 illustrated in light blue, orange, and green) and averaging results (Macro avg, dark blue). Calculations were performed in Python. std. dev. – standard deviation of the mean; TPR – true positive rate; FPR – false positive rate