Machines can learn how to heal you

It’s 2045, and American troops are engaged in a firefight in a remote desert. A soldier is down, unconscious, struggling to breathe —- and there’s not a medic in sight. A fellow infantryman reaches for a special backpack. Inside is TRACIR, which stands for Trauma Care In A Rucksack.

It’s an inflatable vest that uses artificial intelligence algorithms to measure heart rate, blood pressure and blood oxygen. The wounded soldier is placed on the TRACIR vest, and it quickly finds a collapsed lung. The vest detects where the collapse is located, and then automatically inserts a needle between his ribs to reinflate his lung. He can breathe again. It may be just enough to allow him to survive until he can be evacuated.

TRACIR is a real project at the University of Pittsburgh and Carnegie Mellon University, funded with millions of dollars in research money from the Department of Defense. It’s just one of dozens of examples cited at a recent conference on the emerging use of artificial intelligence and machine learning in healthcare. Hosted by Pitt’s Center for Military Medicine Research, the event was funded by the Jewish Healthcare Foundation and DSF Charitable Foundation.

Using machine learning algorithms, AI can discern patterns in huge arrays of data — and nothing produces data like the nation’s healthcare system — and learn new ways to detect, diagnose and treat disease. Besides TRACIR, here are just a few of the other projects the conference highlighted:

When carbon monoxide gets into the bloodstream, it latches onto the hemoglobin in red blood cells, and prevents them from carrying oxygen. Dr. Jason Rose, a researcher in Pitt’s Division of Pulmonary, Allergy and Critical Care Medicine, and his team have developed compounds that serve as CO magnets, so the poisonous gas latches onto them rather than hemoglobin. They use an AI program that predicts what shape proteins will assume, based on their amino acids, to craft antidotes that are stable and highly attractive to CO molecules.

One of the biggest risks to injured soldiers is hemorrhaging, where sudden bleeding can cause blood pressure to crash beyond a patient’s ability to recover. Victor Convertino, a Ph.D. researcher at the U.S. Army Institute of Surgical Research, has done carefully controlled lab studies using a negative pressure chamber. While lying flat, people’s lower bodies are encased in an airtight chamber which draws blood away from their heads and torsos, mimicking what happens during a blood loss injury.

An algorithm that learned more than 130 million data points based on real-time measurements allows a new device to assess how well each person is able to compensate for their plummeting blood pressure. This can help identify who is most likely to go into circulatory shock.

Several studies have shown that AI algorithms can do a better job than human doctors at detecting signs of disease in X-rays and MRI scans. Shandong Wu heads up the Center for Artificial Intelligence Innovation in Medical Imaging, which encompasses more than 100 researchers from Pitt, CMU and UPMC. Among their research initiatives: breast cancer imaging, liver disease, heart disease, lower back pain and mental health.

The use of AI/ML algorithms is still in its formative stages and may be several years away from widespread use. But research leaders believe many of these computer techniques are on the verge of a breakthrough.

Michael Pinsky, a pioneering critical care medicine researcher at Pitt, helped develop a course in AI/ML for Pitt medical students. It focuses on the three major uses of such algorithms in medicine: detecting whether you’re sick, identifying what’s making you sick and tracking whether you’re getting better. Students who show a strong interest in the topic can go on to get specialized computer science training after taking the elective.

“Machine learning is not some special thing,” he said, “but will be a big part of medicine going forward.”