Language-based AI model spots early heart disease in ECGs, reaching 94.2% accuracy

A machine-learning model based on Transformer architecture, a form of artificial intelligence originally developed for language processing, can be used to detect heart disease from electrocardiograms (ECGs), according to research published in the International Journal of Medical Engineering and Informatics. Tests show it works well with data from several well-known medical datasets.

Heart disease is a major health care problem—almost 18 million people die prematurely each year because of it. The challenge is finding ways to detect cardiovascular disease early enough to make treatment effective. An ECG is the standard way to record the heart’s electrical activity and is thus a common diagnostic tool. Interpretation of the trace requires significant expertise, is time-consuming and is not without the risk of misinterpretation.

The researchers discuss a one-dimensional (1D) Transformer model that can analyze ECG signals in parallel with other clinical data. In tests, it was up to 94.2% accurate in spotting the early stages of heart disease. Such precision, coupled with expert clinical assessment, could give health care teams more reliable options in taking a patient to the next step in diagnosis and potential treatment.

The researchers suggest that their approach needs further development and validation with independent clinical datasets before it can be tested in a live clinical setting.

Who’s behind this story?

Lisa Lock

BA art history, MA material culture. Former museum editor, paramedic, and transplant coordinator. Editing for Science X since 2021.

Full profile →

Advertisements

Andrew Zinin

Master’s in physics with research experience. Long-time science news enthusiast. Plays key role in Science X’s editorial success.

Full profile →