Table 5.
Signed mean ± standard deviation results for knee osteoarthritis patients comparing U-Net and U-Net++ two-dimensional (2D) deep-learning segmentations on transverse ultrasound images to subsequent three-dimensional (3D) reconstructed segmentations. The p-value corresponds to a comparison between 2D and 3D segmentation performance.
| Method | Segmentation | sA/VPD (%) | sMSD (mm) | sHD (mm) |
|---|---|---|---|---|
| U-Net | 2D Transverse | 13.4 ± 9.41 | 0.14 ± 0.05 | 1.14 ± 0.45 |
| 3D Reconstruction | 30.1 ± 15.6 | 0.33 ± 0.09 | 1.91 ± 0.57 | |
| P-value | 0.014 | 0.025 | 0.038 | |
| U-Net++ | 2D Transverse | 10.4 ± 7.58 | 0.11 ± 0.04 | 0.69 ± 1.01 |
| 3D Reconstruction | 31.5 ± 13.9 | 0.34 ± 0.07 | 2.02 ± 0.64 | |
| P-value | 0.025 | 0.007 | 0.049 |