Retina Layer Detection in OCT Images Using Learned Models
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Conventional medical image processing methods for OCT images struggle with accurate boundary detection of retina layers, especially in diseased eyes with irregular shapes due to conditions like glaucoma, bleeding, and vitiligo, leading to erroneous detections.
Innovation Solution
A medical image processing apparatus and method using a learned model trained on tomographic images to detect retina layers, employing machine learning algorithms for image segmentation, which can identify layer boundaries regardless of disease or site irregularities.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If conventional image processing methods are used for boundary detection, then the processing is simple and fast, but the detection accuracy deteriorates in diseased eyes with irregular retinal shapes
Solution Approach 1:
The patent transforms the boundary detection problem from geometric shape analysis to pixel intensity value analysis. By changing the detection parameter from spatial coordinates to intensity values in the tomographic image, the system can accurately detect boundaries even when retinal shapes are irregular due to disease. This parameter transformation enables the use of machine learning models that process intensity data rather than geometric data.
Solution Approach 2:
The patent replaces conventional geometric-based boundary detection algorithms with machine learning-based detection. The learned model processes pixel intensity values and automatically identifies boundaries, substituting the mechanical/geometric approach with an intelligent system that can adapt to various retinal conditions including diseases like glaucoma, diabetic retinopathy, and age-related macular degeneration.
2Reliability
If machine learning-based detection is used, then the detection accuracy improves for irregular shapes, but the processing time and computational complexity increase
Solution Approach 1:
The patent performs preliminary preparation by collecting and organizing training data that includes various retinal conditions and their corresponding ground truth boundaries. The machine learning model is trained in advance on this diverse dataset, enabling it to quickly and accurately detect boundaries in clinical practice without requiring complex real-time processing. This preliminary training phase transfers the computational burden from the detection stage to the model development stage.
3Ease of manufacture
If conventional boundary detection is used, then the system is simple to implement, but it produces erroneous detections in diseased eyes
Solution Approach 1:
The patent fundamentally changes the detection parameter from geometric shape characteristics to pixel intensity values. This transformation allows the system to detect boundaries based on optical properties rather than assuming regular geometric shapes, thereby maintaining implementation feasibility while dramatically improving detection precision in diseased eyes with irregular retinal morphology.
Data Source
AI summary
A medical image processing apparatus including an obtaining unit configured to obtain a tomographic image of an eye to be examined, and a first processing unit configured to perform first detection processing for detecting at least one layer of a plurality of layers in the obtained tomographic image, by using the obtained tomographic image as an input data of a learned model, wherein the learned model has been obtained by using training data including data indicating at least one layer of a plurality of layers in a tomographic image of an eye to be examined.