Rapid AI-assisted Poincaré plot analysis for early detection of subclinical left ventricular dysfunction in hypertensive patients admitted to the emergency department: A prospective observational study.

Abstract

Introduction: Early detection of subclinical left ventricular (LV) dysfunction in hypertensive patients presenting to the emergency department (ED) is of critical importance. We aimed to evaluate the performance of artificial intelligence (AI)-assisted Poincaré plot analysis of electrocardiogram (ECGs) to identify subclinical LV dysfunction rapidly.

Methods: 60 hypertensive patients and 55 normotensive controls were prospectively enrolled in the ED. After stabilisation, all participants underwent 5-minute ECG recordings. Heart rate variability (HRV) measurements were calculated, and Poincaré plots were generated. A convolutional neural network (CNN) model was trained to classify the Poincaré plot images. Transthoracic echocardiography was performed within 24 hours to measure left ventricular ejection fraction (LVEF) and global longitudinal strain (GLS). Subclinical dysfunction was defined as GLS > -18% with preserved LVEF.

Results: The CNN model detected subclinical LV dysfunction by achieving an area under the curve (AUC) of 0.91 (95% CI: 0.84-0.96). Conventional HRV measurements showed a moderate correlation with GLS (r = 0.48), while AI-based Poincaré plot analysis had a stronger correlation (r = 0.70). The time from ECG acquisition to automatic analysis was less than 5 minutes.

Conclusions: AI-assisted Poincaré plot analysis enables rapid and non-invasive detection of subclinical LV dysfunction in hypertensive patients in the ED, supporting early risk stratification.