Double-reading mammograms using artificial intelligence technologies: A new model of mass preventive examination organization

• DOI: 10.17816/dd321423
• Publication Date: 2023-05-25T10:37:34Z
• Authors: Yuriy A. Vasilev; Ilya A. Tyrov; Anton V. Vladzymy...; Kirill M. Arzamasov; Igor M. Shulkin; Daria D. Kozhikhina; Lev D. Pestrenin
• Keywords: Digital Mammography

BACKGROUND: In recent years, the availability of medical datasets and technologies for software development based on artificial intelligence technology has resulted in a growth number solutions diagnostics, particularly mammography. Registered as device, this program can interpret digital mammography, significantly saving time, material, human resources healthcare while ensuring quality mammary gland preventive studies.
 AIM: This study aims to justify possibility effectiveness intelligence-based first interpretation mammograms maintaining practice radiologists second description X-ray images.
 MATERIALS AND METHODS: A dataset 100 mammography studies (50 absence target pathology 50 ― presence pathology, with signs malignant neoplasms) was processed by that registered device Russian Federation. Receiver operating characteristic analysis performed. Limitations include values diagnostic accuracy metrics obtained versions, relevant at end 2022.
 RESULTS: When set 80.0% sensitivity, specificity 90.0% (95% CI, 81.798.3), 85.0% 78.092.0). 100% specificity, demonstrated 56.0% sensitivity 42.269.8) 78.0% 69.986.1). 100%, 54.0% 40.267.8), 77.0% 68.885.2). Two approaches have been proposed, providing an autonomous using intelligence. The approach is evaluate image higher than double-reading mammogram radiologists, comparable level specificity. implies will determine category (absence or pathology), indicating degree confidence result, depending corridor into which predicted value falls.
 CONCLUSIONS: Both proposed provide to, if not superior double-image reading radiologists. economic benefit from practical implementation nationwide range 0.6 5.5 billion rubles annually.