Would you like to know when you die? While you ponder this is not the simplest issue, researchers from Pennsylvania trained artificial intelligence to predict the likelihood of death of a person in the course of the year, after reviewing the patient’s electrocardiogram (ECG). However, the researchers do not understand exactly how the AI does it. The fact that the algorithm indicates the results of the ECG, which seemed to cardiologists is completely normal. But how is this possible and what AI analyses?
Electrocardiography is a method of research and recording of electrical activities of the heart.
According to the publication The New Scientist, the results were impressive and a little frightening. In the course of work, scientists have provided data about the AI ECG 400 thousand patients. All the AI got the record of 1.77 million ECGs taken from patients at different times of day to determine patterns that might indicate future problems with the heart, including the likelihood of heart attack and atrial fibrillation.
According to the results of the study, the model of AI have shown better results than all existing methods that enable the identification of patients whose risk of death increases during the year from those, whom death does not threaten. Moreover, AI has identified heart problems in those patients that were previously treated by cardiologists.
What you don’t see cardiologists?
The study team gave the AI the data in two different ways. First, the algorithm was able to see the raw ECG results, allowing one to track changes in the ECG over time. In another case, the researchers provided ECG data with an indication of the age and sex of patients. Answers AI scientists measured using the AUC — it measures how well the model distinguishes between two groups of people — in this case, patients who died during the year and those who survived. The researchers note that the AUC for the risk model, currently used by doctors, ranging from 0.65 to 0.8. And the AI consistently score above 0.85 points (ideal is 1 point and 0.5 indicates no difference between the two groups).
To always stay up to date with news from the world of science, subscribe to our news channel in the Telegram
AI accurately predicted the risk of death even in people, the results of the EKG which cardiologists believe is normal. Three of the cardiologist, who separately studied the ECG of patients are unable to identify risk, discovered AI. Thus, the model developed by the specialists sees things that are not available to doctors today. Scientists say that some of the things the doctors might have misinterpreted for decades.
This is not the only attempt to use the capabilities of machine learning to predict death. Last year, researchers from Google in mountain view, California, has created a prognostic modelthat uses electronic medical records to predict length of stay in hospital and time of discharge and time of death. Moreover, different models of the AI have also been used to diagnose cardiovascular diseases and lung cancer. In some cases, the diagnosis of AI was more precise diagnosis of doctors.
Despite the accuracy of the predictions of some models, there is one drawback: all these models can not and do not try to explain how AI works. For this reason, many make accurate conclusions about the effectiveness of such models is impossible. Note that according to the world health organization (who) cardiovascular diseases are the most common cause of death in the world, annually taking the lives of 17 to 23 million people. What are the ways of preventing cardiovascular diseases do you know? Tell them about the members of our Telegram chat.