Can MRI-derived imaging biomarkers improve the prediction of knee osteoarthritis progression?

Knee osteoarthritis is a serious and common joint disease that can cause pain, functional limitations, reduced mobility, and lower quality of life. Predicting which patients are likely to progress is important because it could help clinicians identify high-risk individuals earlier and guide more personalized treatment strategies.

Currently, knee osteoarthritis diagnosis and progression assessment rely largely on radiographic imaging, clinical symptoms, and established clinical markers. However, accurately predicting which patients will experience worsening structural damage, worsening pain, or both remains challenging.

‍

Predicting knee osteoarthritis progression using neural network with longitudinal MRI radiomics, and biochemical biomarkers: A modeling study

Wang T, Liu H, Zhao W, Cao P, Li J, Chen T, Ruan G, Zhang Y, Wang X, Dang Q, Zhang M, Tack A, Hunter D, Ding C, Li S. Predicting knee osteoarthritis progression using neural network with longitudinal MRI radiomics, and biochemical biomarkers: A modeling study. PLoS Med. 2025;22(8):e1004665. doi:10.1371/journal.pmed.1004665

‍
‍Wang et al. performed a multicenter longitudinal modeling study using data from 594 participants with Kellgren-Lawrence grades 1–3. Participants were followed over a 2-year period and assessed at baseline, 1 year, and 2 years using knee MRI, biochemical biomarkers, and clinical data.

Based on that, the authors developed an integrated model combining MRI radiomic features from load-bearing knee tissues with biochemical biomarkers and clinical variables. The model was designed to predict four outcomes: both joint space narrowing and pain progression, joint space narrowing progression alone, pain progression alone, and non-progression.

‍

The integrated model showed strong predictive performance in the total test cohort, with AUCs ranging from 0.880 to 0.913 across the different progression outcomes. The model output was also strongly associated with combined joint space narrowing and pain progression, with an adjusted odds ratio of 30.906 (95% CI: 22.470–42.511). In addition, when resident physicians used the model for support, their accuracy in predicting knee osteoarthritis progression improved from 46.9% to 65.4%.

‍

Importantly, the model that integrated MRI radiomics, biochemical biomarkers, and clinical variables outperformed models based on MRI radiomics alone, biochemical biomarkers alone, clinical variables alone, or semiquantitative MRI scoring systems such as the Magnetic Resonance Imaging Osteoarthritis Knee Score (MOAKS). This suggests that combining imaging, biochemical, and clinical information may provide a more complete picture of osteoarthritis progression risk than relying on any single data source.

‍

Although these findings are promising, they should be interpreted with caution. The integrated model demonstrated very good to excellent discrimination, meaning it could rank patients by progression risk fairly well. However, clinical implementation would require further validation in independent populations and real-world settings. The study also used a non-routine MRI sequence, and the model did not incorporate all knee joint structures. In addition, threshold-based performance, such as sensitivity and specificity at clinically meaningful cutoffs, will be important to define before such a tool can be used to classify individual patients as progressors or non-progressors in practice.

‍

Overall, this study highlights the potential value of MRI-derived radiomic biomarkers for predicting knee osteoarthritis progression. More broadly, it reinforces an important theme in imaging AI: multimodal models that combine imaging with clinical and biological context may be better positioned to support personalized risk prediction than imaging-only approaches. As imaging AI moves beyond detection and toward prognosis, studies like this show how radiomics and multimodal biomarkers may help identify patients at risk before irreversible disease progression occurs.

‍