| Year | Insti-tution | Number of patients | Number of images for machine learning | Fracture type (femoral neck/ trochanteric fracture) | Images including implants on hip or spine | Accuracy (%) | Sensitivity (%) | Specificity (%) | AUC | Grad-CAM | Clinician test (AI-aided test) |
---|---|---|---|---|---|---|---|---|---|---|---|---|
Adams et al. [17] | 2018 | 1 | 805 | 805 | femoral neck fracture | excluded | 90.6 | N/A | N/A | 0.98 | no | no |
Urakawa et al. [19] | 2018 | 1 | 1773 | 3346 | femoral trochanteric fracture | excluded | 95.5 | 93.9 | 97.4 | 0.97 | no | no |
Cheng et al. [18] | 2019 | 1 | 3605 | 3605 | both | included | 91 | 98 | 84 | 0.98 | yes | no |
Yamada et al. [21] | 2019 | 1 | 1047 | 2923 | both | excluded | 98.0 | 98.0 | 98.0 | N/A | no | no |
Krogue et al. [22] | 2020 | 1 | 1118 | 3026 | both | included | 93.7 | 93.2 | 94.2 | 0.98 | yes | yes |
Cheng et al. [20] | 2020 | 1 | 3605 | 3605 | both | excluded | 91.0 | 98.0 | 84.0 | N/A | yes | yes |
Current study | 2021 | 3 | 4851 | 10,484 | both | included | 96.1 | 95.2 | 96.9 | 0.99 | yes | yes |