MIT researchers are developing novel machine-learning techniques to reduce toxic chemotherapy and radiotherapy dosing for patients diagnosed with glioblastoma, the most aggressive form of brain cancer. The goal is to improve the quality of life for these patients.
A malignant tumor, glioblastoma stems in the brain or spinal cord and the prognosis for adults is no more than five years. Additionally, patients must go through a combination of drugs with radiation Health care provider’s usually administer a minimum safe drug dose that will possibly shrink the tumor. Unfortunately, many of these drugs cause debilitating side effects in patients.
Now, in a study presented at the 2018 Machine Learning for Healthcare, researchers described an artificial intelligence model that could make dosing regimens less toxic but still hold high efficacy. The "self-learning" machine-learning technique seeks to look at treatment regimens currently in practice and then iteratively adjusts the doses. Eventually, the model conforms a treatment plan that has the lowest possible potency and frequency of doses but still could reduce the size of tumors to a degree that is comparable to that of traditional regimens. "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," explains Pratik Shah, the principal investigator of the study at the MTI Media Lab.
The AI model utilizes a behavioral psychology method known as reinforced learning (RL) where a model learns to favor certain behavior that eventually produces a desired outcome. The technique combines artificially intelligent "agents" with what is referred to as complete "actions" in an unpredictable manner in order to reach a desired "outcome." Furthermore, researchers adapted the RL model for glioblastoma treatments that combine four drugs: temozolomide (TMZ) and procarbazine, lomustine, and vincristine (PVC). Additionally, researchers had to make sure the model considers the maximum number and potency of doses. "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," explains Shah says. "Instead, we said, 'We need to reduce the harmful actions it takes to get to that outcome.'"
The model was also designed to treat patients on a personalized level. "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," explains Shah.
Source: Massachusetts Institute of Technology