[THE ROLE OF AI TECHNOLOGY IN MAMMOGRAPHIC SCREENING OF BREAST CANCER WITH EVALUATION OF EFFICIENCY AND POTENTIAL FOR PRACTICAL HEALTH CARE].

Abstract

In conditions of increasing workload on organized mammographic screening programs and deficiency of qualified personnel, implementation of auxiliary technologies on the basis of artificial intelligence (AI) acquires special actuality. The results of retrospective cohort study included 100 histologically verified cases of breast cancer (BI-RADS category 6) permitted to evaluate diagnostic efficiency of AI algorithm as compared to independent conclusions of three certified radiologists with length of service more than 10 years. The sensitivity of technology of AI amounted 86%-96%, depending on dichotomization model. The percentage of total coincidence with appraisal of experts varied within the limits 87-91%. The level of concordance, assessed using weighted Cohen's kappa coefficient, came up to κq = 0.76 (95% CI: 0.68-0.83) that corresponds to "good" concordance. In 1%-3% of cases, the AI algorithm identified suspicious findings missed by experts, at that without increasing volume of groundless examinations due to conservative classification of most disputable cases as BI-RADS 3 with ordering dynamic observation. The obtained results confirm clinical applicability of AI technology as standardizing and relieving tool in practical health care, capable to increase sensitivity of screening and to decrease risk of false-negative conclusions without violation of established clinical protocols.