Smart software is getting better at predicting if someone will have a heart attack

Scientists from the American Johns Hopkins University have developed a series of algorithms developed that can detect the risk of a patient having a heart attack. The technology can even predict when it would happen, allowing doctors to quickly decide what type of treatment to use.

First, a neural network is trained to construct a personalized assessment for each cardiac patient examined. The AI ​​does this by analyzing heart images of people with heart scarring, learning to recognize patterns that can’t be seen with the naked eye.

“The images contain critical information that doctors have not had access to until now,” said Dan Popescu, one of the scientists. “These scars say something about the survival rate of patients. There is information hidden in them.”

More accurate predictions than doctors

Once the visual data was reviewed, a second neural network was trained on patient data from the last ten years. This took into account 22 factors, such as the age, weight and race of the subjects studied.

By combining the results of the two algorithms, the scientists were able to predict the risk for each patient with greater accuracy than the doctors. The trial has been successfully replicated in 60 different health centers across the United States, suggesting the technology can be applied anywhere.

Personalized health care

According to Natalia Trayanova, one of the study’s authors, heart attacks are responsible for around 20% of all deaths worldwide. “But we still know little about the causes and who is most at risk.” As a result, high-risk patients often did not receive the right treatment in time, causing them to die prematurely.

Using this technology, scientists hope to give patients personalized predictions, allowing them to intervene early.

The team now also wants to develop algorithms that can detect other heart diseases. According to Trayanova, the same technology could also be used in other areas of medicine that use visual diagnostics.