Deep Learning Auto-Quantifies Knee Effusion-Synovitis Volume on MRI, Outperforms Semiquantitative Scores

DL model auto-segmented knee effusion-synovitis volume (ESV) on MRI (Dice 0.79, 95% CI 0.71–0.86) in 4,698 OA knees. ESV showed stronger associations with structural features and symptoms than semiquantitative scores, but needs external validation.

• Retrospective study using Osteoarthritis Initiative data (n=4,698); model trained/tested on manual segmentations from 101 knees.
• ESV correlated moderately with WORMS semiquantitative scores (ρ=0.50) and MOAKS scores (ρ=0.65), but produced larger standardized effect sizes for structural MRI features and WOMAC symptoms (minimum Δβstd=0.03, 95% CI 0.00–0.07).
• Key limitation: single-cohort validation only (OAI); authors explicitly call for independent external cohort validation before clinical or research deployment.

RadPigeon summaries are original and for information only. They are not clinical advice.