MIT Creates AI to Optimize Brain Cancer Treatment

MIT Creates AI to Optimize Brain Cancer Treatment

Advances in medical science have resulted in higher cancer survival rates, but the treatments consisting mainly of chemotherapy and radiation are still rough on patients. A team of researchers from MIT has devised an AI platform that can help. The software takes into account important metrics based on the clinical records of other patients before recommending a regiment designed to shrink brain tumors while lessening the side effects on patients.

The MIT team started with a simple reinforced learning (RL) system similar to the one Google’s DeepMind used to create an unbeatable Go-playing AI. The program consists of “agents” that recommend actions which get you closer to the desired outcome. If the recommendation is good, the agent is rewarded with greater weighting. If not, the agent is penalized and has to adjust its recommendations. Over time, the AI becomes highly optimized at solving a specific problem, in this case creating treatments for glioblastoma, an aggressive form of brain cancer.

The researchers trained the model on 50 simulated glioblastoma patients who sought treatment in the past. The AI conducted 20,000 trial-and-error tests to optimize itself for the best outcomes. After that, the team ran 50 new patient profiles through the model to see what the machine would recommend. The AI usually thinks patients can get the same or better outcome with much less chemotherapy and radiation.

Doctors refer to a number of standard treatment regimens for glioblastoma, which has a mean survival time of just five years. The goal is basically to poison the tumor cells faster than non-cancerous cells, but the side effects of going after an aggressive disease like this can be devastating. These traditional treatment schedules don’t take into account differences in tumor size, medical histories, genetic profiles, and biomarkers. The system developed by MIT does that, resulting in lowered dosages and sometimes even skipping doses altogether.

Glioblastoma example.
Glioblastoma example.

For many simulated patients, the AI recommends a quarter or half dosage at various intervals. Sometimes, it projects a tumor can still shrink even if the patient only comes in for treatment less often than the standard 30 days. Some simulations recommend administering drugs only every six months to shrink tumors.

This system has only been tested in simulations so far, but the researchers believe the science is sound. The FDA has been seeking ways to enhance medical treatment with technology, and this could provide a more precise way of modifying patient care than trusting doctors to use their own judgment. Hopefully, it’s better than Watson.

Continue reading

Lasers Used to Create Negative Mass Particles

Researchers at the University of Rochester have worked out a way to create negative mass particles using, what else, lasers. Is there anything lasers can't do?

Google’s AutoML Creates Machine Learning Models Without Programming Experience

The gist of Cloud AutoML is that almost anyone can bring a catalog of images, import tags for the images, and create a functional machine learning model based on that.

People Are Using a Neural Network App to Create Fake Celebrity Porn

Machine learning has become so advanced that a handful of developers have created a tool called FakeApp that can create convincing "face swap" videos. And of course, they're using it to make porn.

MIT Researchers Create Color-Shifting Ink for 3D Printers

A new printing technology designed by MIT's Computer Science and Artificial Intelligence Laboratory (CSAIL) has the potential to add significantly more colors to 3D printing without the need for additional print heads.