Fig. 2: Modified X-ray images generated by AttrNzrs.
From: Enhancing fairness in AI-enabled medical systems with the attribute neutral framework
a–d Average modified X-ray images of the attribute modification. The modified attributes are (male) to (female), (female) to (male), (male, ≥60 y) to (female, <60 y), and (female, <60 y) to (male, ≥60 y), respectively. Other average modified X-ray images with other modified attributes are shown in Supplementary Fig. S4. e–h Examples of modified X-ray images. The modified attributes are the same as a–d. Examples of X-ray images with other modified attributes are shown in Supplementary Fig. S5. Each subfigure consists of four rows. The first row displays a bar chart depicting the Structural Similarity Index Measure (SSIM) between the modified X-ray image and the original X-ray image. The second row shows the original X-ray image and the modified X-ray images. The third row displays the difference image between the modified X-ray images and the original X-ray image. In comparison to the original X-ray image, regions of high intensity in the modified X-ray image are marked in red, while regions of low intensity are marked in green. The fourth row exhibits the modification intensities α with values ranging from 0.0 to 1.0 sequentially (0.0, 0.1, 0.2, 0.3, 0.4, 0.5, 0.6, 0.7, 0.8, 0.9, 1.0). a, b, e, and f are generated by the single-attribute AttrNzrs, while c, d, g, and h are generated by the multi-attribute AttrNzrs.
