MRI-based radiomic models of subchondral bone for predicting the occurrence of knee osteoarthritis

Abstract

To extract the MRI-based radiomic features of subchondral bone, and establish multi-regional radiomic models and delta-radiomic models for predicting the 2-year risk of knee osteoarthritis (KOA). The data used in this study were obtained from the osteoarthritis initiative (OAI). Radiomic features of the subchondral bones of the patella, femur and tibia were observed by MRI of the knee joint at baseline and 12 months. A total of 223 knees were selected for the case group, and 223 knees were matched according to clinical data for the control group. They were randomly divided into a training group ( n = 312) and a testing group ( n = 134) at a ratio of 7:3. The least absolute shrinkage and selection operator (LASSO) was used to reduce the feature dimension, and the support vector machine (SVM) was used to build prediction models. A receiver operating characteristic (ROC) curve was used to evaluate the predictive efficiency of these models. The AUCs of the patella, femur, tibia and multiregional radiomic models at baseline in the testing groups were 0.782, 0.807, 0.804 and 0.835, respectively. The AUCs of the 12 M radiomic models for the patella, femur, tibia and multi-region in the testing groups were 0.829, 0.816, 0.806 and 0.840, respectively. The AUCs of the delta-radiomic models for the patella, femur, tibia and multi-region in the testing groups were 0.837, 0.836, 0.823 and 0.844, respectively. MRI-based radiomic models of subchondral bone showed strong predictive efficacy, with the multiregional delta-radiomic model exhibiting the highest predictive performance and outperforming both single-region and single-time point models.