Table 4 IoU and fairness scores for knee segmentation across different protected attributes, including race, sex, and age.
From: Fair AI-powered orthopedic image segmentation: addressing bias and promoting equitable healthcare
U-Net Backbone | Model | White/Caucasian | Black/AA* | SER | SD | |
|---|---|---|---|---|---|---|
(a) Racial group IoU & fairness scores | ||||||
ResNet18 | Baseline | 0.924 | 0.924 | 1.005 | 0.000 | |
Balanced | 0.903 | 0.920 | 1.210 | 0.008 | ||
Stratified | 0.916 | 0.917 | 1.008 | 0.000 | ||
Group | 0.918 | 0.917 | 1.008 | 0.000 | ||
EfficientNet-B0 | Baseline | 0.924 | 0.926 | 1.023 | 0.001 | |
Balanced | 0.908 | 0.923 | 1.182 | 0.007 | ||
Stratified | 0.924 | 0.926 | 1.024 | 0.001 | ||
Group | 0.909 | 0.921 | 1.159 | 0.006 | ||
U-Net Backbone | Model | Male | Female | SER | SD | |
|---|---|---|---|---|---|---|
(b) Sex group IoU & fairness scores | ||||||
ResNet18 | Baseline | 0.924 | 0.923 | 1.013 | 0.000 | |
Balanced | 0.902 | 0.909 | 1.073 | 0.003 | ||
Stratified | 0.921 | 0.920 | 1.017 | 0.001 | ||
Group | 0.914 | 0.921 | 1.087 | 0.003 | ||
EfficientNet-B0 | Baseline | 0.926 | 0.924 | 1.019 | 0.001 | |
Balanced | 0.918 | 0.911 | 1.094 | 0.004 | ||
Stratified | 0.922 | 0.920 | 1.033 | 0.001 | ||
Group | 0.919 | 0.918 | 1.010 | 0.000 | ||
U-Net Backbone | Model | Age 50 or Lower | Age 51–64 | Age 65–79 | SER | SD |
|---|---|---|---|---|---|---|
(c) Age group IoU & fairness scores | ||||||
ResNet18 | Baseline | 0.923 | 0.924 | 0.924 | 1.022 | 0.001 |
Balanced | 0.912 | 0.893 | 0.894 | 1.215 | 0.009 | |
Stratified | 0.924 | 0.927 | 0.924 | 1.044 | 0.001 | |
Group | 0.824 | 0.911 | 0.916 | 2.083 | 0.042 | |
EfficientNet-B0 | Baseline | 0.924 | 0.927 | 0.922 | 1.057 | 0.002 |
Balanced | 0.910 | 0.895 | 0.896 | 1.174 | 0.007 | |
Stratified | 0.925 | 0.927 | 0.925 | 1.025 | 0.001 | |
Group | 0.889 | 0.919 | 0.913 | 1.373 | 0.013 | |
