Retinal Deep Learning Pipeline for Diabetic Kidney Disease Staging
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current methods fail to accurately predict the stage of diabetic kidney disease (DKD) due to lack of routine screening for microalbuminuria, especially in low-resource settings, and do not utilize deep learning techniques to analyze retinal images for vascular abnormalities that correlate with DKD progression.
Innovation Solution
A system and method using a two-step deep learning approach, involving an AI retina model to analyze retinal images for pathological data and a clinical model to predict DKD stage based on retinal findings and clinical parameters, enabling accurate staging of DKD.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Device complexity
If routine screening for microalbuminuria is not performed, then healthcare costs and complexity are reduced, but DKD stage prediction accuracy deteriorates
Solution Approach 1:
The patent uses retinal images as an intermediary marker to predict DKD stage. Instead of directly measuring microalbuminuria or kidney function, the system analyzes retinal vascular abnormalities (such as arteriolar narrowing, venular dilation, and microaneurysms) which serve as surrogate indicators of kidney disease progression. This intermediary approach enables DKD staging without requiring complex microalbuminuria screening while maintaining prediction accuracy.
2Measurement precision
If deep learning analysis of retinal images is implemented, then DKD stage prediction accuracy is improved, but device complexity and computational requirements increase
Solution Approach 1:
The patent extracts and focuses on specific pathological features from retinal images that are most predictive of DKD stage. The deep learning model is trained to identify and quantify key vascular abnormalities (arteriolar narrowing, venular dilation, microaneurysms) rather than analyzing the entire image spectrum. This extraction approach improves prediction accuracy while reducing computational complexity by concentrating on the most relevant visual features.
3Measurement precision
If comprehensive pathological feature extraction from retinal images is performed, then prediction accuracy is improved, but processing time and computational resources increase
Solution Approach 1:
The patent performs preliminary analysis by pre-processing retinal images to enhance vascular structures and pre-identifying potential pathological regions before the main deep learning analysis. The system also pre-trains models on large datasets of retinal images with known DKD stages, allowing for faster inference on new images. This preliminary action reduces processing time while maintaining comprehensive feature extraction capabilities.
Data Source
AI summary
Embodiments herein disclose a method and system for staging diabetic kidney disease using deep learning techniques. An image capturing unit captures a set of ophthalmic images of a user. The ophthalmic images set undergoes pre-processing before being fed to a first deep learning module. The first deep learning module extracts pathological data indicative of vascular abnormalities from the pre-processed set of ophthalmic images. The first deep learning module quantifies the extracted pathological data, and maps them to a stage of diabetic retinopathy and urine protein levels. A second deep learning module receives as input the quantified pathological data, the mapped diabetic retinopathy stage and urine protein levels, and clinical and demographic parameters. Based on this input, the second deep learning module predicts a stage of diabetic kidney disease.