Retinal Layer Boundary Extraction Using Luminance Gradients
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing methods for extracting boundary lines from tomographic images, particularly in low-luminance regions, suffer from inaccurate edge detection and unreliable boundary line extraction due to insufficient edge information and one-dimensional weighting, leading to difficulties in extracting ambiguous or disappearing lines.
Innovation Solution
An image processing apparatus and method that combines boundary position information from an existence probability image with luminance value information, using weighted luminance gradients to enhance the accuracy of boundary line extraction, and dynamically adjusts the search range based on previously extracted lines to improve extraction precision.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Extent of automation
If edge detection is used to extract boundary lines from tomographic images, then the extraction process can be automated, but the accuracy deteriorates in low-luminance regions where edge information is insufficient
Solution Approach 1:
The patent combines multiple information sources (edge information, luminance gradient information, and existence probability from machine learning) into a unified evaluation function. This merging allows the system to leverage complementary strengths of each method, particularly using existence probability to guide edge detection in low-luminance regions where traditional edge detection fails.
Solution Approach 2:
The patent introduces an existence probability map as an intermediary that guides the boundary line extraction process. This probability map, generated by machine learning, acts as a mediator that highlights likely boundary locations, thereby improving edge detection accuracy in regions where direct edge information is weak or absent.
2Productivity
If simple edge detection methods are used, then the processing speed is high, but the reliability of boundary line extraction deteriorates in ambiguous or disappearing line regions
Solution Approach 1:
The patent performs preliminary processing by generating an existence probability map using machine learning before conducting the actual boundary line extraction. This preliminary action prepares guidance information that significantly improves the reliability of subsequent extraction, especially in ambiguous regions where simple edge detection would fail.
Solution Approach 2:
The patent changes the parameter representation from simple edge magnitude to a composite evaluation function that incorporates existence probability, luminance gradients, and edge information. This parameter transformation allows the system to maintain processing efficiency while dramatically improving reliability in difficult extraction regions.
3Measurement precision
If multiple information sources are combined for boundary extraction, then the accuracy improves, but the device complexity increases due to multiple processing components
Solution Approach 1:
The patent creates a unified evaluation function that serves multiple purposes: it guides boundary detection, handles low-luminance regions, and manages ambiguous boundaries all through a single integrated mechanism. This multi-functionality reduces the need for separate specialized components for each challenging scenario.
Solution Approach 2:
The machine learning model learns existence probabilities from training data and uses this learned information to guide its own extraction process. The system essentially serves itself by using the probability map it generates to improve its own boundary detection performance, reducing the need for external complex processing components.
Data Source
Figure 1
Figure 2
Figure 3
AI summary
Edges of layers are detected from an input image to create a boundary line candidate image that represents the detected edges. A luminance value of the input image is differentiated to create a luminance value-differentiated image that represents luminance gradient of the layers. An evaluation score image is created which is obtained by weighting calculation at an optimum ratio between a boundary line position probability image and the luminance value-differentiated image. The boundary line position probability image is obtained from the boundary line candidate image and an existence probability image that represents existence of a boundary line to be extracted. A route having the highest total evaluation score is extracted as the boundary line. According to such an image processing apparatus and image processing method, boundary lines of layers can be extracted with a high degree of accuracy from a captured image of a target object composed of a plurality of layers.