Artificial Intelligence–Derived Retinal, ECG and Echocardiographic Biomarkers for Early Detection and Risk Prediction of Heart Failure in Hypertension and Diabetes Clinics

About the Project

This project develops and validates AI-derived retinal, ECG, echocardiographic, and serum biomarkers to enable early detection and risk prediction of heart failure in patients with hypertension and diabetes. Using retrospective and prospective data, it integrates multimodal deep learning and clinical tools to create scalable, non-invasive screening strategies for earlier diagnosis, improved risk stratification, and better clinical outcomes in high-risk populations.

Heart failure (HF) frequently develops as a late-stage complication of hypertension and diabetes mellitus, conditions that are routinely managed in outpatient clinics long before HF is diagnosed. People with hypertension have about a 71 % higher relative risk of developing HF compared to people without hypertension [1]. In the Framingham Heart Study, 91 % of new HF cases were preceded by hypertension [2]. People with diabetes have a 1.7 to 4.9 fold higher risk of developing HF [3,4]. Despite regular follow-up, early or preclinical HF often remains undetected due to non-specific symptoms and limited access to advanced cardiac imaging. Novel, scalable biomarkers are needed to enable earlier identification of individuals at risk, facilitating timely intervention.

Retinal vessels share anatomical and physiological characteristics with coronary microcirculation, and retinal microvascular abnormalities have been associated with hypertension, diabetes, and adverse cardiovascular outcomes [5,6]. AI-derived retinal biomarkers may indicate systemic microvascular and cardiometabolic dysfunction and can help early HF identification or risk of HF in people with hypertension (HTN) or diabetes mellitus (DM) [7]. Integration of retinal AI with AI derived parameters from ECG [8], NTproBNP and AI assisted echocardiography [9] can aid early detection and risk stratification of HF. Integrating multimodal AI biomarkers across retinal imaging, ECG, and echocardiography may provide a powerful, low-burden strategy for early HF detection in high-risk clinic populations.

Aim

To develop and evaluate AI-based retinal, serum (NTproBNP), ECG, echocardiographic biomarkers for early HF detection in individuals with hypertension or diabetes.

Objectives

1. To identify AI-derived retinal biomarkers associated with prevalent and incident HF using retrospective data.
2. To prospectively validate retinal AI biomarkers for early HF detection in hypertension and diabetes clinics.
3. To assess the incremental diagnostic and predictive value of integrating AI ECG, AI echocardiography with AI retinal biomarkers.
4. To evaluate feasibility and clinical applicability of multimodal AI screening pathways in non-cardiology settings.

Experimental Approach

• Work Package 1
  • Retrospective analysis of existing retinal fundus photographs (± OCT) linked to electronic health records.
  • Deep learning (foundation models, multimodal models) based feature extraction capturing vascular geometry, microvascular lesions, and retinal structure.
  • Deep learning (foundation models, multimodal models) Time-to-event modelling for incident HF
• Work Package 2
  • Recruitment of patients without known HF from hypertension and diabetes clinics.
  • Baseline retinal imaging, serum biomarkers, AI echo integrated into clinic workflows.
  • Longitudinal follow-up for HF diagnosis, biomarker elevation, or echocardiographic abnormalities.
• Work Package 3
  • AI ECG analysis of standard 12-lead ECGs to detect latent ventricular dysfunction.
  • AI echocardiography for automated assessment of systolic/diastolic function and myocardial strain.
  • Multimodal data fusion and explainable AI modelling.
  • Comparison with standard diagnostic pathways.

Outcomes

This project is expected to deliver validated AI-derived retinal biomarkers for HF risk, demonstrate their prospective utility in hypertension and diabetes clinics, and establish the added value of integrating point of care NTproBNP, AI ECG and AI echo. The findings may inform scalable, non-invasive screening strategies and support earlier intervention to reduce HF burden. By leveraging AI across retinal imaging, ECG, and echocardiography, this project aims to transform early heart failure detection in high-risk outpatient populations, bridging data-driven innovation with real-world clinical impact.