SOFT TISSUE DEGENRATION 10+ YEARS AFTER ACLR AND THEIR ASSOCIATION WITH RADIOGRAPHIC PTOA AND PAIN DEVELOPMENT: RADIOMIC ANALYSIS USING qMRI APPROACH IN MOON COHORT

Abstract

INTRODUCTION

Patients after ACL have a high risk of developing post-traumatic osteoarthritis (PTOA) regardless surgical reconstruction (ACLR). However, long-term soft tissue degeneration after ACLR and their relationship with patient pain development are largely unknown.

OBJECTIVE

To investigate radiomic features of cartilage, menisci and thigh muscle that associates with radiographic PTOA and pain development in patients 10+ years after ACLR using qMRI.

METHODS

169 patients from the Multicenter Orthopedic Outcomes Network (MOON) on-site cohort were studied. Patient-reported outcome measures were collected using knee injury and osteoarthritis outcome scores (KOOS) survey from patients before surgery and at 10 years follow-up (13.1 ± 1.8 years after ACLR). Radiographs and knee and mid-thigh qMRI were also collected at 10 years follow-up. Radiographic PTOA was defined from radiographs based on a KLG ≥ 2. From KOOS, pain score with threshold of ≤ 85.0 were utilized (one SD below the mean KOOS Pain score of healthy subjects). All qMRI data were acquired at three sites using four different 3T MRI scanners and centrally processed, using an established workflow with rigorous quality control. T1rho and T2 maps of knee cartilage and menisci were acquired using MAPSS, along with DESS for registration and segmentation. For the mid-thigh, fat fraction maps and anatomical images were acquired using 6-point Dixon and T1-weighted TSE scans (Figure 1). A total of 17698 radiomic features were extracted from qMRI maps. A Boruta based feature selection was employed to select 20 features associated with radiographic PTOA and KOOS pain. The selected radiomic features with clinical features such as age, graft-type and BMI were trained using gradient boost machine learning model with five-fold cross-validation. The model's performance was evaluated using mean and SD of the area under the receiver-operating curves (AUROC), sensitivity, and specificity on test data.

RESULTS

27% of the patients exhibited radiographic PTOA based on KL grade, while 14% of the patients exhibited KOOS pain ≤ 85 (Table 1). Out of 20 selected features, ten features from cartilage and menisci regions, and muscle fat fraction were relevant to radiographic PTOA, while nine selected features from cartilage and muscle were relevant to KOOS-pain. The selected features alone resulted in high predictability performance for radiographic (0.81 ± 0.09) PTOA and KOOS-pain (0.68 ± 0.13) compared to clinical features (Figure 2).

CONCLUSION

The radiomic analysis show that features from both articular cartilage and thigh muscle were associated with radiographic PTOA and pain development 10 years after ACLR. Radiomic analysis with qMRI may serve as a powerful tool for improving our understanding of PTOA and pain development after ACLR. Future works may involve inclusion of tissue and joint lesions such as fat pat, bone marrow edema-like lesion and effusion.