Glaucoma Diagnosis Using Fundus Image Segmentation and CNN Ensemble
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current glaucoma diagnosis methods, such as Scanning laser Polarimetry, Optical Coherence tomography, and visual field tests, are limited in detecting preperimetric and early glaucoma due to their focus on specific areas, high costs, and inefficiencies, making early detection and classification of glaucoma severity challenging.
Innovation Solution
A glaucoma diagnosis method using a fundus image that involves data amplification by generating multiple transformed images based on a preprocessed image, employing multiple individual learning models, including Convolutional Neural Networks, to create a glaucoma determination model for automatic classification of glaucoma severity and detection of Retinal Nerve Fiber Layer defects.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If OCT measures only RNFL in a limited area from the Optic Disc, then quantification and objectification are possible, but it cannot detect lesions outside the capture area and has limited ability to diagnose early glaucoma
Solution Approach 1:
The patent segments the fundus image into multiple capture areas including the optic disc and multiple peripapillary regions. By dividing the examination area into sectors (temporal, superior, nasal, inferior quadrants) and analyzing RNFL thickness in each segment separately, the system achieves both localized precision and comprehensive coverage, resolving the contradiction between measurement precision and capture area.
2Reliability
If visual field test is used to definitively diagnose glaucoma, then diagnostic accuracy is improved, but it takes a lot of time and effort and requires expensive equipment
Solution Approach 1:
The patent performs preliminary analysis of RNFL thickness and optic disc parameters using fundus images before conducting full visual field tests. By identifying high-risk patients through automated image analysis (measuring RNFL defects, cup-to-disc ratio, and other parameters), the system filters candidates who need comprehensive testing, thereby reducing overall test time and resource utilization while maintaining diagnostic reliability.
3Reliability
If visual field test equipment is used, then diagnostic capability is improved, but equipment cost is higher and medical accessibility is reduced
Solution Approach 1:
The patent uses fundus photography, which is a widely available and relatively inexpensive imaging modality, to create a surrogate diagnostic tool for glaucoma detection. By developing automated analysis algorithms that extract glaucoma risk indicators from routine fundus images (which are already captured in most eye exams), the system replicates the diagnostic capability of expensive visual field testers using affordable, accessible equipment.
4Extent of automation
If only OD area of fundus image is applied to deep learning, then automatic diagnosis is achieved, but preperimetric and early glaucoma detection is limited
Solution Approach 1:
The patent extends the analysis from the two-dimensional optic disc region to include the three-dimensional spatial distribution of RNFL thickness across multiple peripapillary capture areas. By analyzing RNFL defects in the temporal, superior, nasal, and inferior quadrants at different distances from the optic disc, the system detects early structural changes that precede visual field loss, thereby improving early glaucoma detection while maintaining automated diagnosis.
Data Source
AI summary
Disclosed herein are a glaucoma diagnosis method using a fundus image and an apparatus for the same. The glaucoma diagnosis method includes performing data amplification of generating multiple transformed images for an original fundus image based on a preprocessed image of the original fundus image, allowing multiple individual learning models of different types to be learned based on the multiple transformed images and generating a glaucoma determination model based on respective outputs of the learned multiple individual learning models, and diagnosing a class of glaucoma for the original fundus image based on the glaucoma determination model.