CLINICAL VALIDATION OF ARTIFICIAL INTELLIGENCE (AI)-BASED CARTILAGE ANALYSIS PREDICTING KNEE REPLACEMENT (KR)

Robust biomarkers are essential for developing OA therapies. Ideally, these biomarkers will function as surrogate endpoints that reliably reflect clinically meaningful outcomes in the relevant phenotype. Cartilage loss is central to the pathogenesis of OA. As such, cartilage morphology represents such a scalable endpoint for use in clinical trials, and to this end, morphometry should best be performed automatically, without human interaction.

To clinically validate AI- and MRI-based morphometric cartilage analysis as a potential surrogate, by studying cartilage loss prior to knee replacement (KR).

KRs at 36- to 60-months' follow-up in the OAI were matched pairwise (1:1) with controls by age, sex and radiographic OA status. Cartilage segmentation was performed using automated, AI-based (sag DESS / cor FLASH), and manual methods (sag DESS). 2-year cartilage thickness change prior to KR was compared between KR cases and controls, using Cohen's D as an effect size measure. The medial compartment (MFTC) was the primary analytic endpoint.

132 case and matched control knees (113 participants; 55% women; age 64.4±8.6 [mean ± SD] years) had DESS, and 35 of those had FLASH in both the case and control knee. In case knees, the MFTC cartilage thickness loss by automated segmentation of the DESS was -263µm (95% CI: -337, -189; SRM -0.63), whereas by manual analysis it was -259µm (95% CI -331, -188; SRM -0.64). In controls, the rate of cartilage thickness change was substantially lower by either analysis method, with the effect size being similar for automated (Cohen's D -0.40) vs. manual segmentation (-0.35). For automated FLASH the Cohen's D was -0.59 vs. -0.30 for manual DESS segmentation in the same knees.

AI-based cartilage analysis performed at least as well as manual expert analysis in differentiating rates of cartilage loss in KR vs. control knees. This was consistent across DESS and FLASH MRI. The current findings support automated MRI-based cartilage analysis as a scalable imaging biomarker linked to a clinically meaningful outcome.