Table 3 Dice score and fairness scores for hip 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 dice scores & fairness scores | ||||||
ResNet18 | Baseline | 0.927 | 0.920 | 1.094 | 0.003 | |
Balanced | 0.914 | 0.910 | 1.039 | 0.002 | ||
Stratified | 0.919 | 0.913 | 1.073 | 0.003 | ||
Group | 0.911 | 0.908 | 1.030 | 0.001 | ||
EfficientNet-B0 | Baseline | 0.922 | 0.918 | 1.061 | 0.002 | |
Balanced | 0.903 | 0.903 | 1.009 | 0.000 | ||
Stratified | 0.927 | 0.922 | 1.077 | 0.003 | ||
Group | 0.917 | 0.909 | 1.096 | 0.004 | ||
U-Net Backbone | Model | Male | Female | SER | SD | |
|---|---|---|---|---|---|---|
(b) Sex group dice scores & fairness scores | ||||||
ResNet18 | Baseline | 0.925 | 0.923 | 1.028 | 0.001 | |
Balanced | 0.912 | 0.910 | 1.030 | 0.001 | ||
Stratified | 0.924 | 0.920 | 1.055 | 0.002 | ||
Group | 0.908 | 0.907 | 1.015 | 0.001 | ||
EfficientNet-B0 | Baseline | 0.922 | 0.919 | 1.036 | 0.001 | |
Balanced | 0.910 | 0.892 | 1.206 | 0.009 | ||
Stratified | 0.923 | 0.918 | 1.058 | 0.002 | ||
Group | 0.903 | 0.913 | 1.118 | 0.005 | ||
U-Net Backbone | Model | Age 50 or Lower | Age 51–64 | Age 65–79 | SER | SD |
|---|---|---|---|---|---|---|
(c) Age group dice score & fairness scores | ||||||
ResNet18 | Baseline | 0.921 | 0.925 | 0.923 | 1.058 | 0.002 |
Balanced | 0.912 | 0.896 | 0.897 | 1.180 | 0.007 | |
Stratified | 0.920 | 0.921 | 0.917 | 1.059 | 0.002 | |
Group | 0.817 | 0.911 | 0.910 | 2.049 | 0.044 | |
EfficientNet-B0 | Baseline | 0.919 | 0.921 | 0.919 | 1.028 | 0.001 |
Balanced | 0.894 | 0.882 | 0.880 | 1.136 | 0.006 | |
Stratified | 0.922 | 0.925 | 0.922 | 1.039 | 0.001 | |
Group | 0.785 | 0.917 | 0.903 | 2.581 | 0.059 | |
