Artemisia: validation of a deep learning model for automatic breast density categorization

Abstract

Breast density is the terminology used in mammography to describe the proportion between fibroglandular tissue and adipose tissue. It is estimated that 50% of women who undergo mammography examinations have dense breast patterns (1). There is evidence that mammographic density is as strong a predictor of risk for breast cancer in African-American and Asian-American women as for white women (2). High breast density is an independent risk factor for breast cancer (3-6). Furthermore, it may link to higher percentages of interval cancers (7). Dense breast tissue can mask lesions and has a negative impact on the sensitivity of the mammography with rates ranging from 85.7% for the adipose patterns to 61% for the extremely dense patterns. It can also generate an increase in false positives from 11.2% for the non-dense patterns to 23% for dense breasts (8). Breast density can be measured through qualitative or quantitative methods. The American College of Radiology (ACR) has established a structured system for the visual Original Article