Artificial Intelligence and Mammography: Conceptual Perspectives and Clinical Applications

Steven Marks

^*President, MedCom Consultants, Inc, Potomac MD

Address correspondence to: Steven Marks, MedCom Consultants Inc, 1311 Fallsmead Way, Potomac, Maryland 20854. E-mail: steven.marks52@gmail.com

Disclosure statement: Steven Marks reports having no financial or advisory relationship with any corporate, medical, or political organization doing work related to this paper or other business activity at MedCom Consultants, Inc.

 

ABSTRACT

Mammography is the cornerstone of breast cancer screening and has made a major contribution to reducing disease-related mortality. Many interval cancers are missed, however, and a large portion of these are found retrospectively to have been detectable. Although offering superior accuracy compared with mammography, three-dimensional digital breast tomosynthesis (DBT), also has a number of shortcomings, including longer reading times and differences in cancer conspicuity by mammographic view. Traditional computer-aided detection, now a mainstay of radiology practices, sought to overcome these problems, but its clinical value remains a subject of debate. Artificial intelligence (AI), the use of computer-based technologies to develop and exploit algorithms that capture cognitive human behavior and processes, has been proposed as a means to overcome all of these deficiencies. Recent advances in AI, which involve more sophisticated algorithms and more powerful computers, have the potential to identify essential and visually undetectable anatomic features and then, via machine-learning models and deep neural networks, assist radiology professionals in assessing breast cancer risk, identifying true-positive and true-negative findings on mammography and DBT, determining treatment response, and predicting the risk of recurrence or metastasis. Emerging evidence confirms the utility of AI algorithms in the evaluation of mammographic acquisitions and in improving workflow, reducing costs, and optimizing the quality of patient care.

This course will investigate the emerging place of AI in mammography. It will begin with a discussion of the basic terms and techniques found in the most common AI models and methods of validation, evaluation, and interpretation. It will focus on the key fields and subfields, the various types of learning processes, and the technical challenges in mammography. The primary clinical applications will be examined along with AI's impact on cancer detection rates, risk stratification, and its intersection with another emerging computer-based approach, radiomics. The intersection of AI and DBT also will appraised. Finally, the challenges and limitations of AI in mammography will be described as will its future prospects including the adoption of AI-based tools in community clinical practice.