Fig. 7: Violin plots of ROC-AUCs and sensitivity SDs of DDMs.

The violin plot shows the distribution of ROC-AUCs or sensitivity SDs across all findings (15 findings in ChestX-ray14, 14 findings in MIMIC-CXR, and 14 findings in CheXpert). The distribution of ROC-AUCs in ChestX-ray14 (a), MIMIC-CXR (b), and CheXpert (c). The distribution of sensitivity SDs in ChestX-ray14 (d, e), MIMIC-CXR (h, i, j, k), and CheXpert (f, g). The attributes assessed for unfairness are as follows: sex (d, f, h), age (e, g, i), race (j), and insurance (k). There are multiple DDMs trained using either original X-ray images or neutralized X-ray images. The neutralized attributes include (sex), (sex, and age), (sex, age, and race), and (sex, age, race, and insurance). In the violin plot, the central white dot represents the median, while the thick line inside the violin indicates the interquartile range. The whiskers represent the range of the data, excluding outliers. Note: The neutralized attribute of the model training data and the attributes where unfairness is assessed may differ. For instance, in subfigure k, the assessed attribute is insurance, but only one model is trained on data with insurance neutralized. The ROC-AUCs and sensitivity SDs of each finding are shown in Supplementary Figs. S14 and S15. Source data are provided as a Source Data file.