Warning: fopen(/home/virtual/enm-kes/journal/upload/ip_log/ip_log_2024-04.txt): failed to open stream: Permission denied in /home/virtual/lib/view_data.php on line 88 Warning: fwrite() expects parameter 1 to be resource, boolean given in /home/virtual/lib/view_data.php on line 89
1Department of Internal Medicine, Seoul National University College of Medicine, Seoul, Korea
2Department of Internal Medicine, Seoul National University Bundang Hospital, Seongnam, Korea
3Department of Internal Medicine, Seoul National University Hospital, Seoul, Korea
Copyright © 2021 Korean Endocrine Society
This is an Open Access article distributed under the terms of the Creative Commons Attribution Non-Commercial License (https://creativecommons.org/licenses/by-nc/4.0/) which permits unrestricted non-commercial use, distribution, and reproduction in any medium, provided the original work is properly cited.
CONFLICTS OF INTEREST
No potential conflict of interest relevant to this article was reported.
Study | Tasks | Data type | Input data amount | Trained algorithm | Train/validation/test set | Main results | Clinical significance |
---|---|---|---|---|---|---|---|
Shim et al. [22] | Screening osteoporosis | DB | 1,792 (34% OP) | ANN, RF, LR, SVM, KNN, DT, GBM | 76%/5-fold CV/24% |
AUROC ANN 0.742, RF 0.727, LR 0.726, SVM 0.724, KNN 0.712, DT 0.684, GBM 0.652 |
Demonstrated performances of 7 ML models to accurately classify osteoporosis, and found ANN as most accurate methods |
Yamamoto et al. [16] | Screening osteoporosis | X-ray | 1,131 (53% OP) | ResNet-18, resNet-34, GoogleNet, EfficientNet b3, EfficientNet b4 | 80%/10%/10% | EfficientNet b3, accuracy 0.885, recall 0.887, NPV 0.865, F1 score 0.894, AUROC 0.937 |
Addition of clinical covariates increased almost all performance metrics in CNN networks over the analysis of hip radiographs alone CNN models can diagnose osteoporosis from hip radiographs with high accuracy |
Yasaka et al. [11] | Screening osteoporosis | CT | 2,045 (% not reported) | CNN (4-layer) | 81%/9%/10% (external validation) | AUROC 0.97 |
By applying a deep learning technique, the BMD of lumbar vertebrae can be estimated from noncontrast abdominal CT Strong correlation was observed between the estimated BMD from CT and the BMD obtained with DXA The study was externally validated in an independent dataset Superior performance of the CNN was more marked in complex types of humerus fractures |
Chung et al. [32] | Fracture detection (humerus) | X-ray | 1,891 (69% fracture) | Resnet-152 | 90%/-/10% | AUROC 1.00, sensitivity 0.99, specificity 0.97 | CNN showed superior performance to that of physicians and orthopedists |
Tomita et al. [29] | Fracture detection (vertebra) | CT | 1,432 (50% fracture) | Resnet-LSTM | 80%/10%/10% | Accuracy 0.892, F1 score 0.908 |
Accuracy and F1 score of CNN were similar to the radiologists’ performance in detecting fracture Visualization by color maps showed that the learning was based on appropriate target lesion |
Mutasa et al. [37] | Fracture detection (hip) | X-ray | 1,063 (69% fracture) | CNN (21- layer) | 72%/18%/10% | AUROC 0.920, accuracy 0.923, sensitivity 0.910, specificity 0.930, PPV 0.960, NPV 0.860 | Data augmentation techniques of generative adversarial networks and digitally reconstructed radiographs showed better performances than those without augmentation |
Su et al. [53] | Fracture prediction (hip) | DB | 5,977 (3% fracture) | CART | 10-fold CV | AUROC 0.73 | Classification of a high-risk group for hip fractures using a classic ML method of CARTs showed a discrimination power similar to that of FRAX ≥3% |
Almog et al. [57] | Fracture prediction (osteoporotic, hip, vertebra) | DB | 630,445 (7% fracture) | Word2Vec, Doc2Vec, LSTM, XGBoost, ensemble | 70%/3-fold CV/30% | AUROC 0.82 |
Development of a short-term incident fracture prediction model based on natural language processing methods Suggested the possibility of using the unique medical history data of the patients over time to predict the risk of fractures |
Muehlematter et al. [56] | Fracture prediction (vertebra) | CT | 120 (50% fracture) | ANN, RF, SVM | 67%/10-fold CV/33% | AUROC 0.97 |
Bone texture analysis combined with ML allows to identify patients at risk for vertebral fractures on CT scans with high accuracy Compared to Hounsfield unit measurements on CT scans, application of bone texture analysis combined with ML may improve fracture risk prediction |
DB, database; OP, osteoporosis; ANN, artificial neural network; RF, random forest; LR, logistic regression; SVM, support vector machine; KNN, k-nearest neighbors; DT, decision tree; GBM, gradient boosting machine; CV, cross validation; AUROC, area under the receiver operating characteristic curve; ML, machine learning; NPV, negative predictive value; CNN, convolutional neural network; CT, computed tomography; BMD, bone mineral density; DXA, dual X-ray absorptiometry; LSTM, long short-term memory; PPV, positive predictive value; CART, classification and regression tree; FRAX, Fracture Risk Assessment Tool.