Table 2.
Agreement between manual and U-Net-based automated segmentations determined from n = 21 knees in the test set in the primary run (top) and the repeated run (bottom)
| DSC | VOE (%) | HD (mm) | ASSD (mm) | |||||||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| corFLASH | sagDESS | corFLASH | sagDESS | corFLASH | sagDESS | corFLASH | sagDESS | |||||||||
| Mean | SD | Mean | SD | Mean | SD | Mean | SD | Mean | SD | Mean | SD | Mean | SD | Mean | SD | |
| Main run: before post-processing | ||||||||||||||||
| MT | 0.92 | 0.02 | 0.91 | 0.02 | 15.5 | 3.0 | 17.1 | 2.7 | 8.27 | 13.32 | 2.79 | 1.08 | 0.15 | 0.04 | 0.13 | 0.03 |
| cMF | 0.88 | 0.03 | 0.89 | 0.03 | 21.9 | 4.8 | 20.0 | 4.3 | 5.72 | 11.02 | 6.14 | 11.24 | 0.28 | 0.13 | 0.18 | 0.07 |
| LT | 0.92 | 0.02 | 0.92 | 0.02 | 14.9 | 3.3 | 15.4 | 2.9 | 8.06 | 8.80 | 5.31 | 8.46 | 0.17 | 0.05 | 0.17 | 0.04 |
| cLF | 0.88 | 0.02 | 0.90 | 0.02 | 20.8 | 3.8 | 17.8 | 2.8 | 3.66 | 2.33 | 3.86 | 5.54 | 0.26 | 0.09 | 0.14 | 0.03 |
| Main run: after post-processing | ||||||||||||||||
| MT | 0.92 | 0.02 | 0.91 | 0.02 | 15.5 | 3.0 | 17.0 | 2.7 | 2.28 | 0.67 | 2.39 | 0.47 | 0.14 | 0.03 | 0.13 | 0.03 |
| cMF | 0.91 | 0.03 | 0.89 | 0.03 | 16.4 | 4.6 | 20.1 | 4.4 | 2.60 | 1.21 | 3.36 | 0.72 | 0.13 | 0.08 | 0.17 | 0.06 |
| LT | 0.92 | 0.02 | 0.92 | 0.02 | 14.9 | 3.3 | 15.4 | 2.9 | 3.00 | 1.09 | 3.21 | 0.92 | 0.16 | 0.04 | 0.17 | 0.04 |
| cLF | 0.91 | 0.02 | 0.90 | 0.02 | 15.8 | 4.0 | 17.9 | 2.8 | 2.08 | 0.56 | 2.65 | 0.55 | 0.13 | 0.06 | 0.14 | 0.03 |
| Repeated run: before post-processing | ||||||||||||||||
| MT | 0.92 | 0.02 | 0.91 | 0.02 | 15.3 | 2.7 | 16.4 | 2.8 | 6.38 | 8.39 | 2.52 | 0.78 | 0.14 | 0.03 | 0.13 | 0.03 |
| cMF | 0.84 | 0.03 | 0.89 | 0.03 | 26.8 | 5.2 | 20.0 | 4.5 | 8.70 | 10.12 | 8.21 | 11.86 | 0.44 | 0.21 | 0.18 | 0.09 |
| LT | 0.92 | 0.02 | 0.92 | 0.02 | 14.8 | 3.4 | 15.0 | 2.8 | 5.64 | 7.24 | 3.00 | 1.00 | 0.15 | 0.04 | 0.16 | 0.03 |
| cLF | 0.88 | 0.03 | 0.91 | 0.02 | 21.2 | 4.7 | 17.0 | 2.6 | 5.63 | 8.64 | 7.15 | 12.77 | 0.25 | 0.12 | 0.14 | 0.05 |
| Repeated run: after post-processing | ||||||||||||||||
| MT | 0.92 | 0.02 | 0.91 | 0.02 | 15.2 | 2.7 | 16.4 | 2.8 | 2.21 | 0.68 | 2.30 | 0.62 | 0.14 | 0.03 | 0.13 | 0.03 |
| cMF | 0.91 | 0.02 | 0.89 | 0.03 | 17.0 | 3.9 | 20.2 | 4.6 | 2.50 | 1.00 | 3.40 | 0.75 | 0.14 | 0.06 | 0.18 | 0.07 |
| LT | 0.92 | 0.02 | 0.92 | 0.02 | 14.7 | 3.4 | 15.0 | 2.8 | 3.02 | 0.93 | 3.00 | 1.00 | 0.15 | 0.04 | 0.17 | 0.03 |
| cLF | 0.90 | 0.03 | 0.91 | 0.02 | 17.5 | 4.8 | 17.1 | 2.5 | 2.18 | 0.48 | 2.42 | 0.49 | 0.16 | 0.08 | 0.13 | 0.02 |
Agreement was assessed from coronal FLASH (corFLASH) and sagittal DESS (sagDESS) MRI acquired at the OAI baseline visit
DSC dice similarity coefficient, VOE volume overlap error, HD Hausdorff distance, ASSD average symmetric surface distance, MT/LT medial/lateral tibia, cMF/cLF central medial/lateral femoral condyle