AI Program Could Help Cancer Treatments Become Less Toxic For Patients

Over 14 million new cases of cancer pop up each year around the world. While the disease itself is deadly, popular forms of cancer treatment can leave the body in devastating shape in order to kill the cancer. A team from Massachusetts Institute of Technology created an Artificial Intelligence system that could help to boost cancer survival while reducing the chances of cancer treatment being too much for a person.

The team's machine learning system doesn't just track any form of cancer. The MIT researchers trained it to reduce the chemo and radiotherapy treatments for glioblastoma -- the most common and aggressive form of brain cancer. 

This machine learning model sures that doses are effective while being less toxic. Over time, the system uses a "self-learning" technique to adjust current doses of treatment given the data its fed. Eventually, the AI will discover a particular patient's optimal treatment plan -- one that still aggressively combats the cancer while not over-treating a person.

The MIT team tested the AI system on simulated trials of 50 patients. The AI successfully created treatment cycles that often skipped doses of treatment in order to let the body fight the cancer and rest.

“We kept the goal, where we have to help patients by reducing tumor sizes but, at the same time, we want to make sure the quality of life — the dosing toxicity — doesn’t lead to overwhelming sickness and harmful side effects,” said Pratik Shah, a principal investigator at the Media Lab who supervised this research.

[see-also]

The team used a system called Reinforced Learning (RL). It's a similar strategy to training a dog by rewarding favorable behaviors that get people to a desired outcome. Here's how the reward system worked with the algorithm: 

"The technique comprises artificially intelligent “agents” that complete “actions” in an unpredictable, complex environment to reach a desired “outcome.” Whenever it completes an action, the agent receives a “reward” or “penalty,” depending on whether the action works toward the outcome. Then, the agent adjusts its actions accordingly to achieve that outcome."

It was a similar strategy used by Alphabet on its DeepMind program in 2016 that helped produce the company's popular Go champion-system. The model combs trhough the regimens created by doctors as generalized guidelines for being treated with cancer. The AI then determines and predicts a change in tumor size in response to earliet treatments to determine whether a new dose of medicine would be too much for the body. 

“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.’ 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,” Shah said.

Conventional wisdom among oncologists is to estimate how much medicine to give a patient. This new AI -- while not perfect -- provides significantly more consistency in observing patterns in treatment. 

Nicholas Schork serves as a professor and director of human biology at the J. Craig Venter Institute, and an expert in clinical trial design.

“[Humans don’t] have the in-depth perception that a machine looking at tons of data has, so the human process is slow, tedious, and inexact,” he said of the AI. “Here, you’re just letting a computer look for patterns in the data, which would take forever for a human to sift through, and use those patterns to find optimal doses.”