Artificial intelligence (AI) — specifically in the form of machine learning — plays an increasingly important role in medicine and healthcare. Machine learning is a type of AI in which the “learning” is derived from statistical analytics applied to data, rather than from explicitly programmed rules. We recently wrote about AI outperforming dermatologists in detecting melanoma. We’ve also covered machine learning identifying suicidal adolescents, predicting pain, and more.

A research team from New York’s Rory Myers College of Nursing published a study showing the relative accuracy of five machine learning algorithms in detecting lymphedema in breast cancer surgery survivors. Lymphedema is a condition in which impaired lymph fluid flow causes swelling, usually in the arms or legs. Breast cancer treatment is the most common cause of lymphedema. Lymphedema occurs in 6% to 70% of breast cancer surgery survivors, the variation depending on the type of cancer and type of treatment. Lymphedema may occur anytime from shortly after treatment to as long as 20 years later. Lymphedema is one of the most feared side effects of breast cancer treatment, with more than 20 symptoms that range from a mild feeling of heaviness to swelling that is both disfiguring and disabling.

The researchers tested five different machine learning algorithms to analyze data from 355 patients from 45 states in the U.S. Data included demographic and clinical information, lymphedema status, and symptoms. Of the five systems, the artificial neural network (ANN) produced the best results; it accurately detected lymphedema in 93.75% of the cases.

The implications of the Rory Myers College of Nursing test for detecting lymphedema in breast cancer survivors are clear, though with some reservations. More studies are needed to test the validity and reliability of the artificial neural networks. Also, the ANN system that works well at detecting lymphedema may not be as accurate with other breast cancer side effects or other types of cancer. The prospect of testing a multitude of conditions and treatments against a panel of machine learning algorithms may reveal a single strongest approach for many or most diseases.