The following is a summary of “AN ECG AI-BASED MULTI-LABEL CLASSIFICATION MODEL ENABLES THE SCREENING FOR INTERVENABLE STRUCTURAL HEART DISEASE,” published in the March 2023 issue of Cardiology by Asfahan, et al.

Echocardiography is an effective tool to evaluate structural heart disease (SHD), but it requires specialized training for its proper utilization. Therefore, for a study, researchers developed an electrocardiography artificial intelligence (ECG-AI) tool that simultaneously identifies multiple SHDs. In addition, the multi-label approach allowed screening of patients for possible treatments, such as valve repair, using SHD-based criteria during ECG-AI analysis.

To develop the model, they utilized 6,182,699 ECGs obtained from 2,161,538 patients. They collected disease labels for seven SHDs, including moderate-to-severe aortic regurgitation, mitral regurgitation, tricuspid regurgitation, aortic stenosis, mitral stenosis, left ventricular ejection fraction (LVEF) <50%, and interventricular septum thickness >15 mm, from echocardiography reports. They defined a “composite” label as positive if any SHD was present. Controls included matched ECGs from patients with no evidence of SHD. They used a train-test-validate split of 50-10-40 to develop the ECG-AI model.

The ECG-AI model demonstrated an area under the composite label’s receiver operating characteristic curve (AUROC) of 0.93. In addition, multi-label classification revealed the highest discrimination for mitral stenosis (MS), while low LVEF exhibited the highest positive predictive value.

The ECG-AI tool developed in the study can effectively screen for major SHDs with significant positive predictive value (PPV). The multi-label classification approach utilized in the model outperformed current approaches, and it can be applied to screen for SHDs that can be treated with intervention. Additionally, the model may be beneficial in optimizing patient selection for clinical trials.

Source: jacc.org/doi/10.1016/S0735-1097%2823%2900460-6