Место издания:https://eular.conference2web.com/#!contentsessions/46093 Eular 2020
Номер статьи:#OP0327
Аннотация:EVALUATION OF THE ARTIFICIAL INTELLIGENCE SYSTEM ACCURACY IN DETERMINING THERADIOGRAPHIC STAGE OF KNEE OSTEOARTHRITISOlga Georginova 1, Margarita Kobzar 21. Medical Research and Educational Center of Lomonosov Moscow State University. Moscow,Russia2. GlaxoSmithKline Consumer Healthcare, Moscow, RussiaBackground: Within the last decade, quick development of artificial neural networks and machinereading programs and their introduction into medical practice is observed [1,2,3]. Recently, aninnovative program, based on the artificial intelligence (AI) technologies (a neural network and machinereading) that analyses knee X-ray images for determining the radiographic stage of OA was created. Itwas launched on the Osteoscan.ru website and is available for use by patients and doctors.Objectives: To evaluate the accuracy of the knee X-ray analyses performed by the system in determiningthe radiographic stage of OA.Methods: Initially, 1300 x-rays of both knee joints where used for training the neural network. Of these,350 were presented in the form of film scans, 950 in the DICOM format.The accuracy of the system in recognition of OA stage by the X-ray was evaluated on a control sample of130 cases. Independently, the X-rays were assessed by certified radiologists and the System.Results: In 124 out of 130 cases the conclusion of a specialist and the System was the same, which is95.4%. Coincidence or discrepancy is a qualitative attribute, so, the accuracy of the estimation wascalculated. Assuming a discrepancy of 0, and coincidence - of 1, = 0,954, the standard error s p = 1.8%.It can be concluded that in 95% of cases the accuracy of the system assessment will be in the range from91.8% to 99%.Conclusion: Osteosan is a program developed on the base of AI technologies, analyzes x-ray images ofthe knee joints for determining OA stage. It provides high accuracy in OA stage determining by assessingknee x-ray, in 95% of cases, the accuracy of the system varies from 91.8% to 99%.