Table 1 Summary of findings
ID | Expert radiologists involved as a control | Mean age, years | Gender, N (%) | AI model | Reference standard | Outcomes | |
|---|---|---|---|---|---|---|---|
Male | Female | ||||||
Ou Yang et al. (2021)/Taiwan [33] | Yes | 81.4 ± 6.95 | 3053 | 2929 | ANN, SVM, RF, KNN, LR | DXA | Machine learning algorithms improve the performance of screening for osteoporosis |
de Vries et al. (2021)/The Netherlands [26] | Yes |  > 50 | 2564 | 5014 | ANN, RSF | DXA | Major Osteoporotic Fracture can be done with adequate discriminative performance |
Shtar et al. (2021)/Israel [20] | Yes | 83.1 ± 7.4 | 514 | 1382 | AdaBoost, CatBoost, ExtraTrees, KNN, RF, SVM, XGBoost | DXA | hip fracture patients are superior to linear and logistic regression models |
Kuo et al., (2020)/China [32] | Yes | 66.1 ± 1.7 | 18 | 151 | Deep-TEN, ResNet-18 | DXA | The bone texture model can detect osteoporosis and predict the FRAX score |
Engels et al., (2020)/Germany [27] | Yes | 75.67 ± 6.20 | 147,377 | 140,709 | LR, SVM, RF, RUSBoost, Superlearner, XGBoost | DXA | Super learners showed poorer discrimination and calibration in the validation set |
Villamor et al., (2020)/Spain [34] | Yes | 81.4 ± 6.95 | NA | 137 | SVM, LR, NN, RF | DXA | Prediction of the hip fracture without interrupting the actual clinical workflow |
Galassi et al., (2020)/Spain [28] | Yes | 81.4 ± 6.95 | NA | 137 | LR, SVM, DT, RF | DXA | Clinical, geometric, and biomechanical variables from the finite element simulation of a side fall are used as independent variables to train the models |
amamoto et al. (2020)/Japan [15] | Yes | 82.7 ± 8.3 | 346 | 877 | CNN; ResNet18, ResNet34, GoogleNet, E cientNet b3, E cientNet b4 | DXA | High accuracy for the CNN models diagnosed osteoporosis from hip radiographs |
Erjiang et al., (2020)/China [24] | Yes | 60.24 ± 10.56 | 107 | 1162 | XGB, BFDA, NN, CB, LR, RF, SVM | DXA | MLTs could improve DXA detection of osteoporosis classification in older men and women |
Kong et al., (2020)/Republic of Korea [31] | Yes | 61.2 ± 8.7 | 970 | 1257 | CB, SVM, LR | DXA | CatBoost model, the top predicting factors |
Hussain et al., (2019)/Republic of Korea [30] | Yes | NA | 150 | RF | DXA | RF will reduce workload and improve the use of X-ray devices | |
Ho-Le et al., (2017) [29]/Australia | Yes | 69.1 ± 6.4 | NA | 1167 | ANN, LR, KNN, SVM | DXA | ANNs can predict hip fractures |
Kruse et al., (2017) [19]/Denmark | Yes | 74.5 ± 65.5, 69.3 ± 59.9 | 717 | 4722 | bagFDA, xgbTree | DXA | Machine learning can improve hip fracture prediction beyond logistic regression |