High-resolution fundus images for ophthalmomics and early cardiovascular disease prediction

Area of Science:

• Ophthalmology and Cardiology
• Artificial Intelligence in Healthcare
• Biomedical Data Science

Background:

• Cardiovascular diseases (CVDs) are a leading global cause of death, necessitating early detection methods.
• Carotid intima-media thickness (CIMT) is a key indicator for atherosclerosis and cardiovascular risk.
• Fundus imaging provides a non-invasive window into systemic vascular health, but linked datasets are scarce.

Purpose of the Study:

• To address the lack of public datasets connecting fundus images and CIMT measurements.
• To introduce the China-Fundus-CIMT dataset for AI-driven CVD prediction.
• To evaluate the impact of integrating clinical data into multimodal predictive models.

Main Methods:

• Compilation of the China-Fundus-CIMT dataset with 2,903 patients, including fundus images, CIMT measurements, age, and gender.
• Development and testing of unimodal and multimodal AI models for CVD risk prediction.
• Comparative analysis of predictive performance using AUC-ROC metrics.

Main Results:

• Multimodal models integrating clinical data showed significant improvements in predictive performance.
• AUC-ROC increased by 3.22% on the validation set and 7.83% on the test set compared to unimodal models.
• The China-Fundus-CIMT dataset proved valuable for AI model development and validation.

Conclusions:

• The China-Fundus-CIMT dataset is a crucial resource for advancing AI-based early CVD screening using fundus images.
• Integrating clinical data alongside imaging significantly enhances the accuracy of CVD predictive models.
• This work facilitates the development of novel, non-invasive methods for cardiovascular risk assessment.