Fig. 4: Performance comparison of various unfairness mitigation algorithms in disease diagnosis.
From: Enhancing fairness in AI-enabled medical systems with the attribute neutral framework
Critical Difference (CD) diagrams for Macro-ROC-AUC (a), Macro-Accuracy (b), Macro-Sensitivity (c), and Macro-Specificity (d). In each diagram, the Friedman test and the Nemenyi post-hoc test are performed across 6 {dataset, attribute} combinations: {ChestX-ray14, age}, {ChestX-ray14, sex}, {MIMIC-CXR, age}, {MIMIC-CXR, sex}, {CheXpert, age}, and {CheXpert, sex}. The CD value is 3.68. Violin plots of ROC-AUC for various DDMs in ChestX-ray14 (e, f), MIMIC-CXR (g–j), and CheXpert (k, l). The violin plot shows the distribution of ROC-AUCs across all findings (15 findings in ChestX-ray14, 14 findings in MIMIC-CXR, and 14 findings in CheXpert). The attributes corresponding to unfairness mitigation include age (e, g, k), sex (f, h, l), insurance (i), and race (j). 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. Source data are provided as a Source Data file.
