Can Machine Learning Be Used to Generate a Model to Improve Management of High-Risk Breast Lesions?
Radiology. 2018 Mar;286(3):819-821. doi: 10.1148/radiol.2017172648. Shaffer K1.
Management of high-risk breast lesions such as papilloma, radial scar, atypical lobular hyperplasia, and lobular carcinoma in situ, is a complex and controversial topic. Review of the literature for individual diagnoses often yields articles with divergent recommendations, from watch and wait to universal surgical excision (1–3). Strongly competing forces are at work in this clinical area. In the era of cost containment in medicine, every effort should be made to avoid unnecessary biopsies. However, missed cancers generate high costs as well and increase exposure of practitioners to other potential risks such as lawsuits.
Such a complex topic, with a diversity of disorders and a large set of potential risk factors, from patient demographics or breast density to complex genetics, could benefit from analysis with the use of newer methods of computational analysis such as machine learning. Machine learning has undergone impressive growth in recent years, beginning in nonmedical areas but now expanding into clinical use. In a recent PubMed search of the words “machine learning in medicine,” more than 300 references were obtained from 2017 alone. Ideal problems for machine learning approaches often involve “big data,” large volumes of data that are rapidly accrued, with diverse types of information (4), which might be available for high-risk breast lesions.