Photo Credit: Cinefootage Visuals
A deep-learning algorithm showed acceptable performance for detecting inflammation on magnetic resonance imaging (MRI) of sacroiliac joints in patients with axial spondyloarthritis (axSpA), researchers reported in Annals of the Rheumatic Diseases. The study tested the previously trained algorithm in an external validation set of MRI scans for 731 patients with non-radiographic and radiographic axSpA. A pair of expert human readers centrally evaluated the scans (in cases of disagreement, an adjudicator weighed in) for inflammation per the 2009 Assessment of SpondyloArthritis International Society definition. The algorithm, blinded to clinical information and central expert readings, then processed the scans. Compared with the human readers, the algorithm showed 70% sensitivity, 81% specificity, 84% positive predictive value, and 64% negative predictive value. The Cohen’s kappa was 0.49, and absolute agreement was 74%. The algorithm may offer “potential to be a useful tool to aid the early and accurate diagnosis of axSpA, particularly for non-expert radiologists and rheumatologists,” researchers wrote.