| | Publication reference | Region of interest | Modality (imaging sequence) | Data set | Network architecture | Performance |
| | Marzorati et al. [64] | Distal FB, proximal TB | CT | 200 images (160 training, 20 validation, 20 testing) | U-Net | DSC: 96% (FB, TB); SN: 96% (FB, TB) | | Raj et al. [3] | FC, TC | MRI (3D-DESS) | SKI10: 100 images (80 training, 20 testing) | μ-Net | DSC: 0.849 (FC), 0.8565 (lateral TC), 0.8066 (medial TC), 0.7847 (PC) | | OAI: 176 images (140 training, 35 testing) |
| | Chaudhari et al. [65] | FC | MRI (3D-DESS) | OAI: 176 images (124 training, 35 validation, 17 testing) | U-Net | DSC± SD: 90.2 ± 1.7% (FC) | | Tack et al. [66] | Menisci | MRI (2D-DESS) | OAI: 1240 subjects (5 datasets) | 2D U-Net (SSM); 3D U-Net | DSC (baseline): 83.8% (medial menisci), 88.9% (lateral menisci) | | Ambellan et al. [67] | FC, TC, FB, TB | MRI: SKI10 (T1, T2, GRE, SPGR FS), OAI (DESS) | SKI10: (60 training, 40 validation, 50 testing) subjects | 2D U-Net (SSM); 3D U-Net (SSM) | Imorphics: DSC ± SD (baseline): 89.4 ± 2.41 (FC), 86.1 ± 5.33 (medial TC), 90.4 ± 2.42 (lateral TC) | | OAI (Imorphics, ZIB): 88 subjects, 507 subjects | ZIB: DSC ± SD: 89.9 ± 3.60 (FC), 85.6 ± 4.54 (TC); ASD ± SD: 98.6 ± 0.30 (FB), 98.5 ± 0.33 (TB) | | SKI10: ASD ± SD: 0.43 ± 0.13 mm (FB), 0.35 ± 0.07 mm (TB) |
| | Tack and Zachow [68] | TC | MRI (DESS) | OAI (Chondrometrics, Imorphics): 1378 subjects, 88 subjects | U-Net | Chondrometrics: DSC ± SD (baseline): 82.85 ± 5.53 (medial TC), 86.11 ± 4.37 (lateral TC) | | Imorphics: DSC ± SD (baseline): 88.02 ± 4.62 (medial TC), 91.27 ± 2.33 (lateral TC) |
| | Iriondo et al. [69] | FC, TC, PC, menisci | MRI (DESS) | OAI (Imorphics): 176 images (1/3 training, 2/3 validation) | CNN | DSC ± SD: 0.890 ± 0.023 (FC), 0.880 ± 0.036 (TC), 0.850 ± 0.068(PC), 0.874 ± 0.024 (menisci) | | Razmjoo et al. [6] | PC, lateral TC, medial TC, medial FC, lateral FC | MRI (MSME spin‐echo sequence) | OAI: 3921 images (training : validation : test set: 65 : 25 : 10%) | 3D V‐Net | DSC ± SD: 0.75 ± 0.11 (lateral TC), 0.69 ± 0.13 (lateral FC), 0.68 ± 0.12(medial TC), 0.69 ± 0.11(medial FC), 0.57 ± 0.17 (PC) | | Tan et al. [70] | FC, TC, PC | MRI (3D-DESS) | OAI: 176 images (120 training, 26 validation, 30 testing) | V-Net with adversarial network | DSC ± SD: 0.900 ± 0.037 (FC), 0.889 ± 0.038 (TC), 0.880 ± 0.043 (PC), 0.893 ± 0.034 (FC, TC, PC) | | Xu and Niethammer [27] | FC, TC, FB, TB | MRI | OAI: 507 images (200 training, 53 validation, 254 testing) | DeepAtlas | DSC ± SD: 97.70 ± 0.65 (FB, TB),81.19 ± 3.47 (FC, TC), 89.45 ± 1.91 (FB, TB, FC, TB) | | Lee et al. [71] | FC, TC | MRI (T1-weighted SPGR) | SKI10: (60 training, 40 validation) images | BCD-Net | DSC ± SD: 97.3 ± 1.9 (FB), 84.4 ± 4.1 (TB), 98.1 ± 1.1(FC), 83.8 ± 5.3(TC) | | Martinez et al. [1] | FB, TB, PB | MRI (3D-DESS) | OAI: 40 images (25 training, 5 validation, 10 testing) | 3D V-Net | DSC: 97.15% (FB), 97.28% (TB), 95.99% (PB) | | Martinez et al. [18] | FB, TB, PB | MRI (3D-DESS) | OAI: 40 images (25 training, 5 validation, and 10 testing) | CNN | DSC: 88.9%–95.2% (FB), 87.0%–95.8% (TB), 85.1%–92.2% (PC) |
|
|
Note. Region of interest: femoral cartilage (FC), tibial cartilage (TC), patellar cartilage (PC), femur bone (FB), tibia bone (TB), and patella bone (PB); modality (imaging sequence): computed tomography (CT) and magnetic resonance imaging (MRI); data set: Osteoarthritis Initiative (OAI); network architecture: statistical shape modelling (SSM) and convolutional neural network (CNN); performance: Dice similarity coefficient (DSC), specificity (SP), sensitivity (SN), average symmetric surface distance (ASD), and standard deviation (SD).
|
