A Researcher from Geisinger Health care in Pennsylvania, the US has developed a new AI by training 1.77 million ECG results from nearly 400,000 people to predict if a person is going to die within a year or not.
It is quite astonishing that the researchers don’t know how this AI is working. The AI can perfectly predict the results even if the ECG report shows normal results. The accuracy is a whopping 85% checked by AUC.
Report from New Scientist suggests that The researchers trained two AI’s, In first the training data is the only raw electrocardiogram (ECG) data while in the second they include gender and sex of the patient too along with ECG data.
An ECG report records the electrical activity of the heart of patients and during a heart attack or any heart disease, its pattern reflects the condition by changing itself.
No matter what, the voltage-based model was always better than any model you could build out of things that we already measure from an ECG.
Brandon Fornwalt, Geisinger Health Care
Doctors who reviewed a patient’s ECG report it as normal while the AI predicted a totally opposite result which is quite perfect.
That finding suggests that the model is seeing things that humans probably can’t see, or at least that we just ignore and think are normal.
Brandon Fornwalt, Geisinger Health Care
Mr. Fornwalt presented this impressive research at the American Heart Association’s Scientific Sessions in Dallas.
