AI-reconstructed ECG facilitates heart monitoring with fewer leads

Researchers have developed an artificial intelligence (AI) algorithm that can reconstruct an ECG from the standard 12 leads to just three, enabling faster and more accessible heart assessments.

In a clinical assessment, a reconstructed ECG that uses three electrodes—two limbs and one precordial leads—was able to detect features consistent with acute myocardial infarction (MI), with an area under the operating characteristic curve of 0.95. This performance, according to the researchers, was similar to that of the original 12-lead ECG. [NPJ Digit Med 2024;7:201]

When interpreted by cardiologists, the reconstructed ECG had an accuracy of 81.4 percent for identifying ECG features of ST-segment elevation MI (STEMI). This was comparable to the accuracy of 84.6 percent achieved when using the original 12-lead ECG.

The work, which was described in a paper published on NPJ Digital Medicine, is a leap towards transforming healthcare delivery—from traditional brick-and-mortar clinical facilities to remote, direct-to-participant care.

“This opens up the door to patients being able to get really high-quality, time-sensitive clinical data without traveling to somewhere that has a 12-lead ECG,” said one of the study authors Dr Evan Muse from Scripps Research Translational Institute, California, US.

“It likely means not only increased access to ECG technology, but decreased costs and improved patient safety,” Muse added.

The algorithm was built based on data from 627,842 12-lead ECGs that had been collected from 277,174 unique individuals. Of these ECGs, 33.66 percent were recorded from individuals 18–60 years old, 16.75 percent from non-White individuals, and 47.73 percent from women.

A little more than half of the ECGs (56.88 percent) had normal sinus rhythm, while the rest had some form of arrythmia, including sinus in 26.06 percent, atrial in 17.12 percent, and ventricular in 7.31 percent.  Cardiac conduction disorders were present in 30.71 percent of the ECGs, while a repolarization abnormality in the ST segment or T wave was seen in 25.91 percent. Ventricular hypertrophy, deviations of the cardiac axis, and ischaemia were present in 7.51 percent, 14.72 percent, and 10.27 percent of the ECGs, respectively.

In the subset of ECGs associated with MI, 47.40 percent had signs of past MI while 2.94 had evidence of acute MI. The anatomical location of the acute MI was anterior in 8.21 percent, septal in 0.37 percent, lateral in 5.83 percent, anteroseptal in 5.26 percent, anterolateral in 6.31 percent, inferolateral in 4.08 percent, inferior/posterior in 32.22 percent, and unspecified in 40.90 percent.

From 12 to 3 leads

“Our results suggest that it is possible to reconstruct a 12-lead ECG from the measurement of a limited set of one precordial and two limb leads, and that the synthetized signal can be used by a cardiologist for the detection of STEMI,” Muse and colleagues wrote in their paper.

“While most of previous investigations assumed that septal lead V2 is the most important precordial lead, we have shown that measuring anterior lead V3 provides the best accuracy for the reconstruction. This finding may be attributed to the central position of V3, resulting in stronger correlations not only with anterior lead V4 but also with septal and lateral leads,” they explained. [J Electrocardiol 2000;33:163-166]

Additionally, the reconstructed ECG requires only commercial sensors, a solution that is particularly valuable in low-resource settings where acquiring a full 12-lead ECG is impractical, the researchers said. “Solutions like [this] … may enable medical examinations, such as cardiac stress tests, to be performed in a home setting, making the health system more agile, especially in combination with the other possibilities offered by telemedicine.”

In a hospital setting, the reconstructed ECG may also eliminate the need for a technician, which can bring down the time required to record the standard ECG leads, and provide a preliminary diagnosis procedure during emergency room admissions or ambulance transports, they added.

“This is an optimal case for AI—taking a few leads of the 12-lead ECG—to make it remarkably informative, which has big practical implications for patients in the future,” stated Dr Eric Topol, one of the study authors and founder of the Scripps Research Translational Institute, in a news release.