PREDICTIVE VALUE ANALYSIS OF MR IMAGING FEATURES ON THE RISK OF KNEE REPLACEMENT IN PATIENTS WITH KNEE ARTHRITIS

Abstract

This study used knee MR imaging features to quantify the severity of knee injury in patients and analyzed the predictive value of knee MR imaging features in their risk of knee replacement. A total of 120 patients with knee arthritis were included from a public knee arthritis database FNIH OAI. First, univariate logistic regression was used to screen clinical features to obtain the clinical risk factors. Then the minimum redundancy maximum correlation (mRMR) method was used to reduce the image features, and the LASSO method was used to further screen the retained image features to construct a Rad model. Next, the multivariate logistic regression method was used to combine the Rad model and the screened clinical risk factors to construct a combined model and its corresponding nomogram. Finally, ROC curve and its related metrics, Hosmer–Lemeshow test and Delong test were used to evaluate and compare the accuracy and consistency of the model performance. Age and body mass index (BMI) were found as significant clinical risk factors for knee replacement. After using mRMR and LASSO methods, 147 image features were reduced to 30 and 7 features, respectively, then these 7 features were linearly combined to construct the Rad model. Age, BMI, and the Rad model were combined to establish the combined model and its corresponding nomogram. The resulted two models showed high accuracy ([Formula: see text]) and consistency (H-L test: [Formula: see text]) on both the training set and test set. Finally, the comparison results showed that the prediction performance of the combined model was better than that of the Rad model, but their difference was not significant (Delong test: [Formula: see text]). This paper studied the predictive performance of MR imaging features on the risk of knee replacement and built a model that can be used to predict the potential risk of knee replacement in patients with knee arthritis. The resulting models showed good predictive accuracy and consistency. The drawn nomogram could be used as a useful tool to personalize the prognosis of patients with knee arthritis and guide clinical decisions on knee replacement.