General, Research, Technology

Artificial Intelligence Predicts ECG Death Probability

Would you like to know when you die? While you are pondering this not the simplest question, researchers from Pennsylvania have trained artificial intelligence to predict the likelihood of a person dying within a year after reading the patient's electrocardiogram (ECG). True, the researchers themselves do not understand exactly how AI does it. The fact is that the algorithm indicates those ECG results that seemed completely normal to cardiologists. But how is this possible and what exactly does AI analyze?

Scientists don't know why AI makes accurate predictions about patients dying from heart disease

Electrocardiography is a research method and registration of the electrical activity of the heart.

According to The New Scientist,the results were impressive and a little scary. In the course of the work, the scientists provided AI data on the ECG of 400 thousand patients. In total, AI received records of 1.77 million ECGs that were taken from patients at different times of the day to find out patterns that could indicate future heart problems, including the likelihood of a heart attack and atrial fibrillation.

According to the results of the study,the AI ​​model showed better results than all the methods that exist today that distinguish patients whose risk of death increases during the year from those who are not at risk of death. Moreover, AI revealed heart problems in those patients who had previously been treated by cardiologists.

What do cardiologists not see?

During the study, a team of specialistsprovided AI data in two different ways. At first, the algorithm could familiarize itself with the unprocessed ECG results, by which it was possible to track changes in the cardiogram over time. In another case, the researchers provided ECG data indicating the age and gender of the patients. Scientists measured AI responses using the AUC indicator - it measures how well the model distinguishes between two groups of people - in this case, patients who died within a year and those who survived. Researchers note that the AUC for risk assessment models currently in use by doctors varies from 0.65 to 0.8. And AI invariably scored above 0.85 points (1 point is considered ideal, and 0.5 indicates the absence of differences between the two groups).

To stay up to date with news from the world of science, subscribe to our news channel in Telegram

AI accurately predicted the risk of death even in humans,ECG results that cardiologists consider normal. Three cardiologists who separately studied the ECG of patients failed to identify the risk detected by AI. Thus, the model developed by specialists sees things that are not available to doctors today. Scientists note that doctors could misinterpret some of the things for decades.

Cardiovascular disease is the most common cause of death in the world.

This is not the only attempt to usemachine learning capabilities for predicting death. Last year, researchers from Google in Mountain View, California, created a predictive model that uses electronic medical records to predict the length of a patient’s stay in the hospital and the time of discharge and the time of death. Moreover, different AI models have also been used to diagnose cardiovascular disease and lung cancer. In some cases, the diagnosis of AI was more accurate than the diagnosis of doctors.

Despite the accuracy of some predictionsmodels, there is one significant drawback: all these models cannot and do not try to explain how AI works. For this reason, many can not yet make accurate conclusions about the effectiveness of such models. Note that according to the World Health Organization (WHO), cardiovascular disease is the most common cause of death in the world, annually killing 17 to 23 million people. And what methods of preventing cardiovascular diseases do you know? Tell the participants about our Telegram chat about them.