Using technology to predict heart attacks before they happen
For now, it’s not much more than a kick – but researchers want to refine the technology to predict heart attacks before they happen. For the first time, a group of scientists has been able to show that a computer is able to learn to recognize people with a few years to live. Trained with machine-learning methods, the computers observed tomographies taken from the chest of a number of people. Then they had to say which of them would die in the next 5 years. Of course the test was done with old CT scans – and some of these people had, in fact, died a half decade after the examination. Thus, it was possible to assess whether Dr. Computer’s opinion was correct…
They’ve created an algorithm that says if you’re about to die
The percentage of times he scored was 69% – not much more than a mere kick. However, it is the same level that a trained medical specialist for this would be able to do, with years of experience observing CT scans. What was impressive indeed was not the precision of the machine guessing. And, yes, the speed with which the computer managed to find similar patterns between who would live longer – and the group of those who were about to die. Analyzing only 48 images, the algorithm reached 15,957 biomarkers that helped to estimate the life span of the patient portrayed on the CT.
The authors of the study admit that they still have much to improve on the project, but believe they are on the right track to developing a new tool for the field of precision medicine. The next step is to train the algorithm with thousands of scans, both healthy people and sick people. The goal is for biomarkers to become so accurate that they help diagnose a heart attack even before symptoms that are detectable by conventional doctors and tests appear.