The researchers took CT scans of 48 patients’ chests, which included tissue images of the aorta, heart, lungs, fat located around the heart, body fat, muscle, and bone. Then, they “taught” an artificial intelligence program to analyze the data.
The results? The computer was able to predict which patients would die within 5 years with 69 percent accuracy—a rate that’s comparable to “manual” predictions made by doctors, the researchers say in a press release.
The researchers could not ID what exactly the computer was using to come up with its predictions, they do know that it use large volumes of data to discern subtle patterns that may be particularly telling for health risks.
The predictions were most accurate for patients with severe chronic diseases, like congestive heart failure or emphysema.
While the study is still preliminary—and only included a small number of patients—the researchers are hopeful that the technology can also one day be used to predict other health problems, like heart attacks before they occur. Their next step is to test their research on a larger scale, with a study analyzing images from tens of thousands of patients.