Patients with glioblastoma, a persistent and difficult to treat brain cancer, often end up suffering through multiple rounds of chemo and radiation therapy. Scientists at MIT have been working on harnessing the power of artificial intelligence to better optimize the therapy dosages, sparing the patients the brunt of the treatments while maintaining their clinical effectiveness.
Their software, which uses a technique called reinforced learning, assesses different data points about a given patient, and uses information obtained from thousand of previous similar cases to produce a treatment plan that is better optimized than existing regimens. So far the technology has been tested in silico on virtual patients, pointing to significant reductions in dosage levels while maintaining the ability to reduce the size of tumors. In the 50 “patients” that were “treated” using the technology, the dosage levels were frequently reduced by half or more, and therapy sessions were often significantly reduced in their frequency. “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 leader of the research.
The reinforced learning technique relies on virtual agents that attempt to complete different tasks in order to meet a goal. The closer to the goal the agent gets to, the greater the reward it receives. The agent learns from these rewards and adjusts its actions to maximize future rewards. Because this is done within a computer and can be performed thousands of times, the agents eventually produce better and better predictions of what actions should be taken.
The research is to be presented this week at the 2018 Machine Learning for Healthcare conference at Stanford University.