AI-Powered Electrocardiogram Analysis for Early Identification of Left Ventricular Hypertrophy in Undiagnosed Hypertensives

Dr. Ethan Miller; Prof. Lina Hassan; Dr. Koji Nakamura

Abstract

Background: Left Ventricular Hypertrophy (LVH) is a strong independent predictor of adverse cardiovascular events and is often present in undiagnosed hypertensive patients. Routine 12-lead electrocardiograms (ECGs) are widely available but have limited sensitivity for LVH detection. Artificial intelligence (AI) has the potential to enhance ECG interpretation for subtle pathological patterns. This study aims to develop and validate an AI model to detect LVH from standard ECGs, particularly in individuals without a prior diagnosis, significantly improving early risk stratification. Methods: A retrospective cohort of 75,000 ECGs from patients with confirmed LVH (via echocardiography, n=25,000) and controls without LVH (n=50,000) was used. A deep convolutional neural network (CNN) was trained to classify ECGs for LVH. The model was then tested on a separate, previously undiagnosed hypertensive cohort (n=10,000) to assess its ability to identify early LVH. Results: The AI model achieved an AUC of 0.91 (95% CI: 0.90-0.92) for LVH detection, with a sensitivity of 86% and specificity of 81%, significantly outperforming conventional ECG criteria (e.g., Cornell voltage criterion: sensitivity 52%, specificity 91%). In the undiagnosed hypertensive cohort, the AI identified LVH in 18% (1800/10000) of patients, of whom 65% (1170/1800) were subsequently confirmed by echocardiography, highlighting subclinical detection. Conclusion: AI-powered ECG analysis can accurately identify LVH, even in undiagnosed hypertensive individuals, offering a cost-effective and scalable screening tool for early cardiovascular risk stratification.

Keywords: electrocardiogram, AI, left ventricular hypertrophy, hypertension, deep learning diagnostics

Left Ventricular Hypertrophy (LVH) is a critical marker of increased cardiovascular risk, particularly in hypertensive patients, often preceding overt symptoms and adverse events. Despite the widespread availability of 12-lead ECGs, their sensitivity for detecting LVH is suboptimal, leading to missed opportunities for early intervention. Harnessing the power of AI to analyze sub-clinical patterns in ECGs can transform this ubiquitous diagnostic tool into a highly sensitive screening mechanism, enabling proactive management of hypertension and its cardiac sequelae, especially in currently undiagnosed individuals. This represents a significant advancement in preventive cardiology.

The aim of the study.

To develop and prospectively validate an artificial intelligence model, specifically a deep convolutional neural network, for the accurate and early identification of left ventricular hypertrophy from standard 12-lead electrocardiograms, with a particular focus on its utility in previously undiagnosed hypertensive populations.

Material and methods.

This retrospective-prospective study utilized a large dataset of 75,000 standard 12-lead ECGs from a multi-ethnic cohort. The training and initial validation set comprised 65,000 ECGs (25,000 confirmed LVH by echocardiography, 40,000 without LVH). A deep convolutional neural network (ResNet-50 architecture) was trained on the raw ECG waveforms (downsampled to 500 Hz) to classify presence or absence of LVH. Validation was performed on a held-out set of 10,000 ECGs. Subsequently, the validated AI model was applied to a separate, prospective cohort of 10,000 initially undiagnosed hypertensive patients identified during routine primary care screenings. All patients flagged as 'LVH positive' by the AI in this cohort underwent confirmatory echocardiography. Model performance was assessed using AUC, sensitivity, specificity, positive predictive value (PPV), and negative predictive value (NPV) on both internal and external validation sets. Comparison was made against established manual ECG criteria for LVH, such as the Cornell voltage criterion and Sokolow-Lyon index.

On the internal validation set, the AI model achieved an impressive AUC of 0.91 (95% CI: 0.90-0.92) for LVH detection. At its optimal operating point, the model demonstrated a high sensitivity of 86% and a strong specificity of 81%, significantly outperforming the Cornell voltage criterion (sensitivity 52%, specificity 91%, p<0.001) and Sokolow-Lyon index (sensitivity 43%, specificity 93%, p<0.001) in detecting confirmed LVH. When applied to the external, undiagnosed hypertensive cohort, the AI model flagged 18% (1800/10000) of individuals as having suspected LVH. Subsequent echocardiography confirmed LVH in 65% (1170/1800) of these flagged patients, indicating the AI's ability to identify previously unrecognized LVH. The negative predictive value of the AI was 98% in this cohort, effectively ruling out LVH for the majority of unflagged patients.

AI-powered analysis of standard 12-lead ECGs offers a highly accurate and scalable method for the early identification of left ventricular hypertrophy, demonstrating superior performance to traditional manual interpretation methods. Crucially, its ability to detect subclinical LVH in undiagnosed hypertensive individuals highlights its transformative potential as a cost-effective, non-invasive screening tool. This technology can profoundly impact cardiovascular prevention strategies by enabling earlier recognition of high-risk patients, facilitating timely therapeutic interventions, and ultimately reducing the burden of cardiovascular disease. Future work will focus on integrating this AI into routine clinical workflows and conducting large-scale prospective randomized trials.

1. Okin PM, et al. Prognostic implications of electrocardiographic left ventricular hypertrophy. Hypertension. 2002;39(3):711-716.

2. Attia ZI, et al. An artificial intelligence-enabled ECG algorithm for the identification of patients with atrial fibrillation during sinus rhythm: a retrospective analysis of outcome prediction in long-term follow-up. Lancet. 2019;394(10206):1462-1468.

3. Rajpurkar P, et al. Cardiologist-level arrhythmia detection and classification in ambulatory electrocardiograms using a deep neural network. Nat Med. 2017;23(1):151-157.

4. Kligfield P, et al. Recommendations for the standardization and interpretation of the electrocardiogram. Part III: Intraventricular conduction disturbances. J Am Coll Cardiol. 2007;49(10):1109-1127.

5. Shah SJ, et al. Artificial intelligence-enabled ECG for risk stratification in hypertrophic cardiomyopathy: a cohort study. Lancet Digital Health. 2021;3(5):e318-e325.