MIT machine learning makes brain cancer treatment less toxic

Register now

Researchers are using machine learning to ease the harsh effects of treating glioblastoma, the form of brain cancer that took the life of Senator John McCain.

The initiative underway at MIT offers hope for those patients suffering from the condition, as the new approach could reduce the toxicity of chemotherapy and radiotherapy.

The machine learning model, which was trained on 50 simulated patients randomly selected from a large database of glioblastoma patients, assesses traditional treatment regimens and iteratively adjusts the doses to the lowest possible potency and frequency while still shrinking brain tumors.

Glioblastoma patients are subjected to a combination of radiation therapy and multiple drugs that produce debilitating side effects. However, using a technique called reinforced learning, MIT’s model calculates dosing regimens that are less toxic but still effective.

In fact, in simulated trials of 50 new patients, the model developed treatment cycles that reduced the potency to a quarter or half of nearly all the doses while maintaining the same tumor-shrinking efficacy.

“We said [to the model], ‘Do you have to administer the same dose for all the patients?’ And it said, ‘No. I can give a quarter dose to this person, half to this person, and maybe we skip a dose for this person,’” says Pratik Shah, a principal investigator at the MIT Media Lab, who presented a paper on the technique earlier this month at Stanford University’s Machine Learning for Healthcare conference.

“That was the most exciting part of this work, where we are able to generate precision medicine-based treatments by conducting one-person trials using unorthodox machine-learning architectures,” adds Shah.

Also See: Differences in equipment and procedures complicates machine learning

The model weighs potential negative consequences of actions (doses) against an outcome (tumor reduction), and in the process learns to favor certain behaviors that lead to that desired outcome.

“If all we want to do is reduce the mean tumor diameter, and let it take whatever actions it wants, it will administer drugs irresponsibly,” Shah observes. “Instead, we said, ‘We need to reduce the harmful actions it takes to get to that outcome.’”

According to Shah, MIT researchers are talking to potential healthcare partners about the possibility of deploying this research into clinical environments for therapy design in order to create optimal treatment plans for patients.

For reprint and licensing requests for this article, click here.