Comprehensive Summary
This study applies machine learning (ML) to enhance the traditional Kellgren–Lawrence (KL) radiographic grading system for predicting clinical outcomes after Genicular Artery Embolization (GAE) in patients with knee osteoarthritis. Sixty-six patients (72 knees) underwent radiography and MRI before GAE and were reassessed six months post-procedure. Using subject-specific variables and pre- and post-treatment Visual Analog Scale (VAS) pain scores, the ML model identified bone marrow lesions, cartilage loss, bone attrition, and higher Whole-Organ Magnetic Resonance Imaging Scores (WORMS) as key imaging features associated with a ≥50% reduction in pain. However, because WORMS were derived from non-contrast MRI—less sensitive than contrast-enhanced imaging—the precision of WORMS in evaluating GAE outcomes may be limited.
Outcomes and Implications
Patient outcomes following Genicular Artery Embolization (GAE) vary widely, underscoring the need for more precise candidate selection. The conventional Kellgren–Lawrence (KL) grading system fails to capture the full range of imaging and clinical factors that influence treatment response. By integrating machine learning (ML) algorithms, the authors identified additional predictive variables beyond the KL scale, refining the ability to forecast patient outcomes. These advancements may enhance physicians’ decision-making in developing individualized OA treatment plans. In order for clinical implementation, further studies incorporating contrast-enhanced MRI are warranted to validate its accuracy.