Structural damage assessment in the spine of patients with axial spondyloarthritis - results from an international OMERACT multi-reader exercise using MRI-based synthetic CT.

OBJECTIVE: To develop and perform preliminary cross-sectional validation of a magnetic resonance imaging (MRI)-based outcome measure for assessing spinal structural damage in spondyloarthritis clinical trials, with a specific focus on MRI-based synthetic computed tomography (sCT) and related MRI-based techniques. METHODS: Consensus meetings within the Outcome Measures in Rheumatology (OMERACT) MRI in arthritis working group established consensus definitions for spinal new bone formation and scoring rules. Then, low-dose CT and sCT images of twenty patients with axial spondyloarthritis and five healthy controls were independently assessed by seven experienced readers using the agreed-upon definitions for: marginal syndesmophytes, non-marginal syndesmophytes, and osteophytes. sCT images were reconstructed using a deep learning algorithm (BoneMRI v1.8, MRIGuidance B.V., Utrecht, the Netherlands). Sensitivity and specificity of synthetic CT were calculated using low-dose CT as the reference standard, and inter-reader reliability was assessed. RESULTS: A total mean of 35.5 lesions was scored on both sCT and low-dose CT in all participants. sCT demonstrated an overall good sensitivity (0.77) and excellent specificity (≥ 0.94) with low-dose CT as reference standard. Inter-reader agreement was substantial for both low-dose CT and sCT, with an overall intra-class coefficient of 0.80 (low-dose CT) and 0.86 (sCT), and corresponding kappa values of 0.68 and 0.70, respectively. CONCLUSION: This OMERACT international multi-reader exercise established consensus definitions for spinal new bone formation and provided preliminary evidence that sCT meets key OMERACT criteria of domain match and feasibility. This MRI technique for generating CT-like images shows promise as a novel method for assessing spinal structural damage in spondyloarthritis clinical trials.