Artificial intelligence in the diagnostics of deforming osteoarthritis of large joints of the lower limbs — diagnostic accuracy assessment in the real clinical settings

BACKGROUND: A development of mathematical methods, digitalization of medical diagnostic equipment, and growth of computing capabilities have created conditions for the emergence of new tools for automated analysis of biomedical data — artificial intelligence (AI) technologies. In clinical practice, computer vision has become the most widespread among promising AI technologies. Since 2023, the Moscow Experiment has been using AI services to diagnose injuries and conditions of the musculoskeletal system, which has allowed studying a quality of the relevant tools for the first time on a large-scale.

AIM: To study a diagnostic significance of software based on artificial intelligence technologies for the diagnosis of deforming osteoarthritis of large joints of the lower limbs.

MATERIALS AND METHODS: The scientific work performed in the design of the diagnostic study according to the STARD 2015 methodology included two stages — retrospective and prospective. The retrospective stage was a calculation of diagnostic accuracy indicators (AUROC, sensitivity, specificity, and accuracy). The prospective phase consisted of regular monitoring of the diagnostic quality of the AI service while analysing the actual flow of radiography results (n=198,821). A match of radiologist and AI service results was calculated, as well as an integral clinical evaluation was performed. The duration of the study was 1 year and 8 months.

RESULTS: Five Russian AI-based software for detecting signs of deforming osteoarthritis were studied. Only two of them successfully passed the stage of retrospective diagnostic accuracy assessment and were accepted to participate in the prospective stage. Both AI services demonstrated the sufficient technical reliability in clinical conditions. One of the AI services had a medium-high level of diagnostic value with the median clinical score of more than 88.0%, while the other had a high level of diagnostic value with the median clinical score of more than 93.0%.

CONCLUSIONS: The achieved level of AI-based software development allows applying them to improve the accuracy and productivity of radiologists when providing radiology reports of large joints of the lower limbs (in the context of diagnosing deforming osteoarthritis).