Can a Computer Diagnose Cancer?

MIT says yes.

In order to make an accurate cancer diagnosis, a physician may reference numerous books, study dozens of similar cases, and follow a variety “best practice” guidelines. But researchers at MIT wondered if it could be easier and quicker. Why isn’t there a way to simply put test results and medical history into a computer and then have it do the work?

Now, MIT researchers have created an algorithm can distinguish between different lymphomas in real time. PhD student Yuan Luo and MIT professor Peter Szolovits, both of whom work in MIT’s Computer Science and Artificial Intelligence Lab (CSAIL), have teamed up with researchers from Mass General to develop what MIT says is a “computational model that aims to automatically suggest cancer diagnoses by learning from thousands of data points from past pathology reports.”

The work was recently published in the Journal of the American Medical Informatics Association (JAMIA).

Lymphoma is a common cancer with more than 50 distinct subtypes. The researchers focused on the three most prevalent that can sometimes be difficult to distinguish. Dr. Ephraim Hochberg, director of the Center for Lymphoma at Mass General and a co-author of the paper, says that between 5 to 15 percent of lymphoma cases are initially “misdiagnosed or misclassified, which can be a significant problem when different lymphomas require dramatically different treatment plans.”

According to a report from MIT, lymphoma classification has long been a source of debate:

There were at least five different sets of guidelines until 2001, when the World Health Organization (WHO) published a consensus classification. In 2008 the WHO revised its guidelines in a labor-intensive process that involved an eight-member steering committee and over 130 pathologists and hematologists around the world. In addition, only around 1400 cases from Europe and North America were reviewed to cover 50 subtypes, meaning that on average a subtype’s diagnosis criteria was based on what happened to only a limited number of people.

Szolovits says that the new model can help doctors make more accurate lymphoma diagnoses and could one day be incorporated into new WHO guidelines.

“Our ultimate goal is to be able to focus these techniques on extremely large amounts of lymphoma data, on the order of millions of cases,” Szolovits said in a statement. “If we can do that, and identify the features that are specific to different subtypes, then we’d go a long way towards making doctors’ jobs easier – and, maybe, patients’ lives longer.”