Comprehensive Summary
This study develops an Automatic Deep Learning (DL)-based MRI analysis of Inflammatory signs in Rheumatoid Arthritis (RA) (ADMIRA) system for inflammation assessment. The model was developed with MRI scans of the wrists, metacarpophalangeal (MCP) and metatarsophalangeal (MTP) joints from patients with varying arthritis progress. The ADMIRA system showed great overall performance, with a mean Intraclass Correlation Coefficient (ICC) of 0.9 and 0.8 for the synovitis and tenosynovitis test and validation. The results for bone marrow edema were slightly lower, being 0.8 and 0.7. This level of performance is close to human experts on the same data. Therefore, the model is able to produce fast and accurate inflammation estimation on par with human experts.
Outcomes and Implications
Rheumatoid arthritis is a painful autoimmune disease causing inflammation in joints, severely affecting the quality of life of patients. Early detection and treatment increases the chances of improved quality of life. Nevertheless, current method of detecting joint inflammation is time consuming since it involves visual evaluation of signs such as bone marrow edema (BME), tenosynovitis, and synovitis. Therefore, a fully automatic system would be able to reduce labor cost and increase efficiency of diagnosis. Moreover, with enough validation, AI models would be able to produce more consistent results over humans, saving more precious resources.