Diagnostic performance and reliability of robotic ultrasonography and artificial intelligence-driven synovitis assessment in rheumatoid arthritis: results from the Controlled ARTHUR Trial.
Authors
Affiliations (4)
Affiliations (4)
- Copenhagen Center for Arthritis Research, Rigshospitalet Center for Rheumatology and Spine Diseases, Copenhagen, Denmark [email protected].
- Copenhagen Center for Arthritis Research, Rigshospitalet Center for Rheumatology and Spine Diseases, Copenhagen, Denmark.
- Department of Clinical Medicine, University of Copenhagen Faculty of Health and Medical Sciences, Copenhagen, Denmark.
- Department of Medicine and Therapeutics, The Chinese University of Hong Kong Faculty of Medicine, Hong Kong, Hong Kong.
Abstract
To evaluate the reliability and diagnostic performance of artificial intelligence (AI) driven robotic ultrasonography (RUS) compared with human ultrasonography (HUS) performed by a trained rheumatologist for assessing synovitis in the hands of healthy controls and patients with rheumatoid arthritis (RA). 20 healthy controls and 29 patients with RA eligible for initiation or intensification of disease-modifying anti-rheumatic drugs, with at least one clinically swollen joint in the hands, were included. The ultrasound robot scanned controls and patients twice without interval, whereas HUS was performed once. Synovitis was scored from 0 to 3 for Greyscale (GS) and Doppler (CD) by RUS (using AI software) and HUS according to the European League Against Rheumatism-OMERACT scoring system (range 0-66).Intrarobot and human-robot reliability and agreement were assessed. Furthermore, the diagnostic performance of RUS for detecting arthritis was evaluated using the clinical arthritis diagnosis as reference. Intrarobot reliability was moderate to good, with intraclass correlation coefficients (ICCs) values of 0.65 (GS) and 0.86 (CD) in patients with RA. Human-robot agreement was moderate, with ICCs of 0.59 (GS) and 0.64 (CD). At joint level, RUS obtained higher scores for MCP joints than HUS, while scores were comparable for other joint groups. The diagnostic accuracy of RUS for detecting clinically-detected arthritis was 59%. A moderate to good overall agreement was seen between RUS and HUS in assessing synovitis in RA hands. However, at the joint level, low agreement was observed, particularly for MCP joints in both intrarobot and human-robot comparisons, which affected diagnostic performance.