LV diastolic dysfunction (LVDD) is a common precursor and important feature of heart failure with preserved ejection fraction (HFpEF) but requires numerous Doppler echocardiographic parameters for detection, limiting widespread screening. Today, Doctor Moon-Seung Soh (Ajou University School of Medicine - Suwon, South Korea) presents a deep-learning AI model developed to identify LVDD from 12-lead ECGs.

This retrospective analysis included paired ECG and echocardiographic examinations from 21,385 patients. Patient-level data were split 7:1:2 into training, validation and test sets. A vision transformer model, pretrained on unlabelled ECGs, was fine tuned to classify LVDD. The definition of LVDD was based on 2016 ASE/EACVI recommendations using four parameters: mean E/e′ ratio >14, septal e′ velocity <0.07 m/s or lateral e′ velocity <0.10 m/s, tricuspid regurgitation velocity >2.8 m/s and left atrial volume index >34 mL/m². Patients meeting more than half of the available criteria were classified as having LVDD.

The researchers found an LVDD prevalence of 15.3% in the test cohort. The model achieved areas under the receiver operating characteristic curve (AUROC) of 0.908 in the validation set and 0.913 in the test set. Using a threshold defined in the validation cohort where sensitivity equalled specificity, the model demonstrated sensitivity of 0.838, specificity of 0.825 and overall accuracy of 0.827 in the test set. The positive predictive value was 0.463, whereas the negative predictive value reached 0.966, supporting its potential role as an initial rule-out screening tool.

The authors conclude that ECG-based AI screening may provide a scalable and non-invasive strategy for the early identification of diastolic dysfunction and may facilitate earlier identification of patients who could benefit from further echocardiographic assessment before the development of symptomatic HFpEF. If prospectively validated, this approach could help prioritise patients for echocardiography and improve the efficiency of HFpEF screening in routine clinical practice. As this was a retrospective single-centre study, prospective external validation will be required before implementation in routine clinical practice.