Table 5.
General characteristics, methodological details, main results, and overall quality score for bone segmentation studies.
| Reference | Country (population) | Task | Modality | Data amount | Data amount external validation | Inputs (no.) | Model | CV train / validation / test | K-fold cross validation | Evaluation metrics | Best result | Quality score (max 12) |
|---|---|---|---|---|---|---|---|---|---|---|---|---|
| Cheng et al.96 | China | SS | CT | 15 | 15 | IMG10 | Custom U-Net | 66.6% train, 33.3% test | Dice coefficient, location error, detection rate, IoU, Hausdorff distance, pixel accuracy | Dice coefficient: 0.95 | 10 | |
| Deng et al.97 | China | HS | CT | 100 | IMG | U-Net | 85% train, 15% test | 10-fold | Dice coefficient, average surface distance, sensitivity, specificity | Dice coefficient: 0.98 | 10 | |
| Kim et al.98 | South Korea | SS | X-ray | 797 | IMG | U-Net, hybrid | 80% train, 20% test | Dice coefficient, precision, sensitivity, specificity, area error, Hausdorff distance | Dice coefficient: 0.92 | 10 | ||
| Kim et al.99 | South Korea | SS | X-ray | 339 | IMG | U-Net, R2U-Net, SegNet, E-Net, dilated recurrent residual U-Net | 80% train, 20% test | 5-fold | Sensitivity, specificity, accuracy, dice coefficient | Dice coefficient: 0.93 | 7 | |
| Park et al.100 | South Korea | SS | CT | 467 | 102 | IMG | U-Net | 80% train, 20% test | Dice coefficient | Dice coefficient: 0.93 | 11 | |
| Suri et al.101 | USA | SS | X-ray, CT, MRI | 6975 | IMG | Custom CNN | 5-fold | 5-fold | Accuracy, IoU, dice coefficient | Dice coefficient: 0.95 | 10 | |
| Wang et al.102 | China | HS | CT | 50 | IMG | U-Net | 66% train, 20% validation, 14% test | 5-fold | Dice coefficient, precision, sensitivity | Dice coefficient: 0.92 | 8 | |
| Wei et al.103 | China | FS | X-ray | 1274 | IMG | hybrid ResNet+FPN and DeepLabv3 | 60% train, 20% validation, 20% test | MAP, AUC | AUC: 0.98 | 10 | ||
| Yang et al.104 | China | FS | DXA | 720 | IMG2 | U-Net Resblock | 83% train-validation, 17% test | 5-fold | Dice coefficient, Jaccard index | Dice coefficient: 0.99 | 10 | |
| Yang et al.105 | China | HS | CT | 160 | IMG | DenseUnet and Mask R-CNN | 75% train, 25% test | Accuracy, dice coefficient | Accuracy: 0.89, dice: 0.90 | 6 | ||
| Zhao et al.106 | China | SS | MRI | 222 | 25 | IMG | U-Net | 70% train, 30% test | Dice coefficient | Dice coefficient: 0.912 | 7 |
Abbreviations: FS, forearm segmentation; SS, spine segmentation; HS, hip segmentation; IMG, image; CNN, convolutional neural network; IoU, intersection over union; MAP, mean average precision; AUC, area under the curve of the receiver operating characteristic. If several models were evaluated for a given task, the best performing is highlighted in bold.