Oppurtunistic Deep Learning–Derived Popliteal Artery Wall Biomarkers from Knee MRI and Medial Compartment Knee Osteoarthritis

Purpose: Automated image-analysis pipelines may enable opportunistic vascular phenotyping from routine musculoskeletal MRI without additional imaging or radiation, supporting scalable analytics and standardized biomarker extraction.

Aim: To test whether automated popliteal artery wall biomarkers derived from baseline knee MRI are associated with (1) baseline medial tibiofemoral cartilage defects and bone marrow lesions (BMLs) and (2) 24-month worsening on MRI Osteoarthritis Knee Score (MOAKS) outcomes.

Methods/Materials: This retrospective analysis used the FNIH OA Biomarkers Consortium (600 participants). Baseline knee MRI (3D DESS) was processed with a validated deep learning vessel-wall pipeline to quantify popliteal artery eccentricity ratio and normalized wall index (NWI). Two experienced readers scored baseline and 24-month MRI using MOAKS cartilage morphology and BML features. Regression models evaluated associations with baseline severity (cartilage defect and BML burden) and with 24-month worsening, adjusting for age, sex, and body mass index.

Results: Higher eccentricity ratio was associated with worse baseline medial tibial cartilage defect severity, including greater number of involved subregions and higher full-thickness and surface-area defect scores (e.g., β≈0.5 for involved subregions and full-thickness; β≈0.9 for surface-area score). Higher eccentricity ratio was also associated with worse baseline medial tibial BML severity (e.g., β≈0.4 for involved subregions; β≈0.7 for BML size score). Mean NWI showed positive associations with baseline cartilage defect severity primarily in males. Popliteal artery wall biomarkers were not associated with 24-month worsening of cartilage defects or BMLs.

Conclusions: Popliteal artery wall biomarkers derived from routine knee MRI, particularly eccentricity ratio, were associated with baseline medial tibial structural damage (cartilage defects and BMLs) but did not predict 24-month worsening, supporting their role in vascular–OA phenotyping rather than short-term prognostication and providing a scalable informatics approach for opportunistic vascular biomarkers that may help standardize risk phenotyping in large imaging cohorts.