Researchers from the American medical company Geisinger have developed an AI capable not only of predicting the probability of death of a patient in the next year from the data of an electrocardiogram, but of doing so with greater reliability than any other similar procedure known.
The patterns of electrical activity in the heart change depending on the condition of the heart and the conditions suffered in the past (such as heart attacks). The key seems to be that this AI perceives indications of this risk in ECG data that doctors consider normal ; but we say “it seems” because the truth is that the way it does it is a mystery.
What does AI see that we don’t?
The Geisinger researchers commissioned an artificial intelligence laboratory to examine the results of 1.77 million EKGs from nearly 400,000 people . Two versions of the AI were created: one trained solely on raw EKG data, while the other used a combination of that data with aspects such as the age and gender of the corresponding patient.
And it was the first version that proved to be better at differentiating the electrocardiograms of patients who, indeed, died at most a year after doing the test, from those who did not. According to Brandon Fornwalt, study principal investigator at Geisinger,
“In all cases, the model based on the electrical activity data was always better than any model that we built from data other than what already appears in the electrocardiogram.”
They measured this using a metric known as AUC, which indicates success in distinguishing between two data sets, where a perfect score is 1 and a score of 0.5 indicates there is no distinction between the two groups. The AI scored above 0.85 , when risk assessment methods currently used by clinicians range from 0.65 to 0.8.
The AI accurately predicted the risk of death even in people who were separately estimated by several cardiologists to have a totally normal EKG. None of them were able to see the risk patterns that the AI did detect.
According to Fornwalt, ” artificial intelligence can teach us things that we have been misinterpreting for decades .”