Cartilage morphometry and magnetic susceptibility measurement for knee osteoarthritis with automatic cartilage segmentation

Abstract

Background: Automatic segmentation of knee cartilage and quantification of cartilage parameters are crucial for the early detection and treatment of knee osteoarthritis (OA). The aim of this study was to develop an automatic cartilage segmentation method for three-dimensional water-selective (3D_WATS) cartilage magnetic resonance imaging (MRI) and conduct cartilage morphometry and magnetic susceptibility measurements such as cartilage thickness, volume, and susceptibility values for knee OA assessment. Methods: Sixty-five consecutively sampled subjects, who had undergone health checks at our hospital, were enrolled in this cross-sectional study and were divided into three groups: 20 normal, 20 mild OA, 25 severe OA. Sagittal 3D_WATS sequence was used to image cartilage at 3T. The raw magnitude images were used for cartilage segmentation and the phase images were used for quantitative susceptibility mapping (QSM)based assessment. Manual cartilage segmentation was performed by two experienced radiologists, and the automatic segmentation model was constructed using nnU-Net. Quantitative cartilage parameters were extracted from the magnitude and phase images based on the cartilage segmentation. Pearson correlation coefficient and intra-class correlation coefficient (ICC) were then used to assess the consistency of obtained cartilage parameters between automatic and manual segmentation. Cartilage thickness, volume, and susceptibility values among different groups were compared using one-way analysis of variance (ANOVA). Support vector machine (SVM) was used to further verify the classification validity of automatically extracted cartilage parameters. Results: The constructed cartilage segmentation model based on nnU-Net achieved an average Dice score of 0.93. The consistency of cartilage thickness, volume, and susceptibility values calculated using automatic and manual segmentations ranged from 0.98 to 0.99 (95% CI: 0.89–1.00) for the Pearson correlation coefficient, and from 0.91–0.99 (95% CI: 0.86–0.99) for ICC, respectively. Significant differences were found in OA patients; including decreases in cartilage thickness, volume, and mean susceptibility values (P<0.05), and increases in standard deviation (SD) of susceptibility values (P<0.01). Moreover, the automatically extracted cartilage parameters can achieve an AUC value of 0.94 (95% CI: 0.89–0.96) for OA classification using the SVM classifier. Conclusions: The 3D_WATS cartilage MR imaging allows simultaneously automated assessment of cartilage morphometry and magnetic susceptibility for evaluating the severity of OA using the proposed cartilage segmentation method.