The first 10,000 mammography exams performed as part of the “Description and interpretation of mammography data using artificial intelligence” service.

Abstract

Artificial intelligence technologies have great potential in improving the effectiveness of screening programs in the detection of malignant neoplasms of the breast. Given the high social, demographic and economic importance of mass preventive research, there is no doubt that the diagnostic accuracy of artificial intelligence must match or even exceed the accuracy of radiologists. In this regard, studies are needed to compare the accuracy of software based on artificial intelligence technology and radiologists during the mammography examinations in a clinical environment. P u r p o s e : to assess the quality of the medical service “Description and interpretation of mammography data using artificial intelligence” as part of screening. M a t e r i a l s a n d m e t h o d s . The sample for analysis consisted of 9684 digital mammograms. For each study, the BI-RADS category was determined by a radiologist and using software based on artificial intelligence technologies (AI based software) registered in the Russian Federation as a medical device. Forty-five studies from this sample with significant discrepancies in physician and software assessments were subject to peer review, which resulted in a BI-RADS category according to the physician expert. R e s u l t s . When evaluating weighted averages, there were no statistically significant differences between physician results and AI based software for 9684 digital mammography exams. Evaluation of physician and software consistency showed that matches are observed in 43,89% of cases for the BI-RADS scale and in 80,69% – 84,10% for binary scales. The presence of a case in which the pathology identified with the help of software and confirmed during the review of the results by the expert was missed by the doctor indicates the promise of using AI based software for evaluating mammography studies and requires further research. F i n d i n g s . When evaluating mammography studies, the agreement between the AI based decision and the radiologist reaches 84,10%, with the software assigning a higher BI-RADS category more often. Expert review of part of these discrepancies showed a potential reduction in the number of missed breast malignancies with the help of software.

Ähnliche Arbeiten