An Overview of Deep-Learning-Based Methods for Cardiovascular Risk Assessment with Retinal Images.

Area of Science:

• Oculomics and Medical Imaging
• Artificial Intelligence in Healthcare
• Cardiovascular Disease Diagnostics

Background:

• Cardiovascular diseases (CVDs) are a leading cause of premature mortality, necessitating early detection strategies.
• Retina fundus imaging (RFI) offers a non-invasive method to identify systemic disease indicators, including CVDs.
• Existing RFI data, primarily for ocular conditions, presents an opportunity for broader health screening.

Purpose of the Study:

• To review recent advancements in deep learning (DL) approaches for automated CVD diagnosis using RFI.
• To provide a comprehensive overview of datasets, preprocessing techniques, and DL models applied in this field.
• To propose a taxonomy for classifying DL-based CVD prediction targets and identify future research challenges.

Main Methods:

• Systematic literature review of 30 studies on deep learning for automated CVD diagnosis from RFI.
• Analysis of commonly used datasets, preprocessing methods, and evaluation metrics in the reviewed studies.
• Categorization of studies based on prediction targets and summary of identified research gaps.

Main Results:

• Deep learning models demonstrate significant potential for automated CVD risk assessment from RFI.
• A variety of datasets, preprocessing techniques, and DL architectures are employed in current research.
• The review establishes a classification taxonomy and highlights key challenges for future development.

Conclusions:

• Automated CVD diagnosis using RFI and DL is a promising, scalable approach for public health.
• Further research is needed to address challenges in data standardization, model generalizability, and clinical validation.
• This review provides a roadmap for advancing AI-driven oculomics in cardiovascular health.