Screening chronic kidney disease through deep learning utilizing ultra-wide-field fundus images

Area of Science:

• Ophthalmology
• Nephrology
• Artificial Intelligence

Background:

• Chronic kidney disease (CKD) poses a significant global health challenge.
• Current CKD screening methods face limitations in scalability and accessibility.
• Retinal imaging offers a potential non-invasive window into systemic health, including renal function.

Purpose of the Study:

• To develop and validate a deep learning model for CKD screening using ultra-wide-field (UWF) fundus images.
• To enhance model interpretability by analyzing retinal microvascular parameters (RMPs).
• To compare the performance of the UWF-based model against a model using only central retinal images.

Main Methods:

• A deep learning model, UWF-CKDS, was developed to predict CKD from UWF fundus images.
• Retinal vessels and RMPs were extracted for correlation analysis with renal function.
• The model was validated using data from 23 tertiary hospitals in China.
• Performance was compared to CTR-CKDS, a model using central retinal images.

Main Results:

• UWF-CKDS accurately determined CKD status by integrating UWF images, RMPs, and medical history.
• A significant correlation was found between renal function and RMPs, enhancing model interpretability.
• UWF-CKDS demonstrated superior performance compared to CTR-CKDS.
• The peripheral retina's contribution to predicting renal function was highlighted.

Conclusions:

• UWF-CKDS is a highly implementable deep learning tool for large-scale CKD screening.
• The model leverages UWF fundus imaging and RMPs for accurate and efficient CKD detection.
• This approach offers a promising strategy for population-level screening and early identification of CKD.