Deep learning-based fundus image analysis for cardiovascular disease: a review

Area of Science:

• Ophthalmology
• Cardiology
• Artificial Intelligence

Background:

• The retina offers insights into cardiovascular health, with microvascular changes linked to cardiovascular disease (CVD).
• Current CVD risk prediction methods rely on invasive and costly assessments of preclinical features, risk factors, or biomarkers.
• Cardiovascular diseases remain a global health challenge despite extensive prevention efforts.

Purpose of the Study:

• To review current evidence on using deep learning (DL) applications with retinal fundus images for cardiovascular disease prediction.
• To explore the potential of DL as an adjunct tool to enhance CVD prevention strategies.

Main Methods:

• Analysis of existing studies employing deep learning algorithms on retinal fundus images.
• Evaluation of DL model performance in predicting CVD risk factors like age, gender, smoking status, hypertension, and diabetes.

Main Results:

• Deep learning models demonstrate comparable performance to human experts in predicting cardiovascular risk factors from fundus images.
• AI-based analysis of retinal images presents a potential non-invasive and cost-effective alternative to traditional CVD risk assessment.

Conclusions:

• Deep learning applied to fundus images is a promising tool for predicting cardiovascular disease risk.
• Further prospective clinical trials are needed to address ethical and medicolegal implications before widespread clinical adoption.