Article type
Year
Abstract
Background: Knee osteoarthritis (OA) is a chronic and progressive joint disease with a higher contributors to global disability, mainly in the elderly and particularly in women.The available diagnostic approaches such as X-ray, computed tomography and magnetic resonance imaging have large precision errors and low sensitivity. Machine learning (ML) is the application of probabilistic algorithms to train a computational model to make predictions, it has great potential to become a valuable clinical diagnostic tool.
Objectives:The aim of this study was to determine the diagnostic and prediction accuracy of different machine learning methods for Knee Osteoarthritis
Methods: Four electronic databases were searched from their inception until July 2019. The Quality Assessment of Diagnostic Accuracy Studies (QUADAS-2) tool was used for risk of bias and applicability assessment. Outcomes assessed were test characteristics such as accuracy, sensitivity, specificity, and area under the receiver operating characteristic curve (AUC). RevMan 5.3 software used to pool data and to carry out the meta-analysis if it is possible.
Results: Based on objective selection, we included six studies that validated the performance of their machine learning method using either a new group of patients or retrospective datasets. Six methods for machine learning able to diagnose and predict knee OA were identified. Five studies reported sensitivity and specificity of 73.2%–94.4% and 73.9%–100.0%, respectively. Two studies report accuracies of 66.71% and 75%. Three study provides an area under the receiver operating curve (AUC) of 0.81, 0.93 and 0.972. In addition to diagnostic performance, two study also reported sensitivity of 77.97% and 88.9%, specificity of 78% and 82% for prediction knee Osteoarthritis.
Conclusions: Of the currently included studies, machine-learning algorithms have demonstrated promising results and certainly have the potential to aid radiologists with detection and screening knee OA. We should interpret the findings of these reviews with caution, considering the problem of over-fitting in machine learning method, and large datasets need to be builted to verify it in the future.
Patient or healthcare consumer involvement: Not application
Objectives:The aim of this study was to determine the diagnostic and prediction accuracy of different machine learning methods for Knee Osteoarthritis
Methods: Four electronic databases were searched from their inception until July 2019. The Quality Assessment of Diagnostic Accuracy Studies (QUADAS-2) tool was used for risk of bias and applicability assessment. Outcomes assessed were test characteristics such as accuracy, sensitivity, specificity, and area under the receiver operating characteristic curve (AUC). RevMan 5.3 software used to pool data and to carry out the meta-analysis if it is possible.
Results: Based on objective selection, we included six studies that validated the performance of their machine learning method using either a new group of patients or retrospective datasets. Six methods for machine learning able to diagnose and predict knee OA were identified. Five studies reported sensitivity and specificity of 73.2%–94.4% and 73.9%–100.0%, respectively. Two studies report accuracies of 66.71% and 75%. Three study provides an area under the receiver operating curve (AUC) of 0.81, 0.93 and 0.972. In addition to diagnostic performance, two study also reported sensitivity of 77.97% and 88.9%, specificity of 78% and 82% for prediction knee Osteoarthritis.
Conclusions: Of the currently included studies, machine-learning algorithms have demonstrated promising results and certainly have the potential to aid radiologists with detection and screening knee OA. We should interpret the findings of these reviews with caution, considering the problem of over-fitting in machine learning method, and large datasets need to be builted to verify it in the future.
Patient or healthcare consumer involvement: Not application