Table 3.
MAPE between manual measurements and grading and automatic measurements and grading, as well as between manual measurements and grading and measurements and grading from different networks segmentation, p < 0.05.
| MAPE of thickness measurement | |||||
|---|---|---|---|---|---|
| Automatic | UNet | UNet++ | ResUNet | TransUNet | |
| Medial Femur | 6.06% | 15.18% | 18.96% | 20.15% | 15.46% |
| Lateral Femur | 7.70% | 12.18% | 15.97% | 18.14% | 15.02% |
| Medial Tibia | 13.38% | 25.53% | 24.54% | 22.34% | 20.68% |
| Lateral Tibia | 13.26% | 23.45% | 21.89% | 18.73% | 18.73% |
| Total | 10.10% | 19.09% | 20.34% | 19.84% | 17.47% |
| Accuracy of grading | |||||
|---|---|---|---|---|---|
| Automatic | UNet | UNet++ | ResUNet | TransUNet | |
| Medial Femur | 93.08% | 86.17% | 76.79% | 80.74% | 78.02% |
| Lateral Femur | 90.12% | 90.86% | 76.29% | 81.97% | 82.96% |
| Medial Tibia | 87.40% | 60.49% | 66.66% | 76.04% | 84.19% |
| Lateral Tibia | 89.38% | 71.35% | 82.22% | 89.13% | 87.16% |
| Total | 89.99% | 77.21% | 75.49% | 81.97% | 83.08% |
For a unified statistical analysis, we aggregated the data from the 12 regions into four areas.