Endoscope Image Region Screening for Reliable Lesion Detection
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current medical image processing systems face challenges in accurately detecting regions of interest due to the presence of structures or artifacts like dark portions, blurriness, residues, or specular reflections, which reduce detection accuracy and reliability of evaluation results.
Innovation Solution
An endoscope system that illuminates a subject with different light patterns, captures images, and uses classifiers to evaluate and enhance regions of interest, removing artifacts and enhancing lesion detection by dividing images into multiple regions and applying specific light spectra to improve accuracy.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If an image analysis is performed on a medical image including structures or artifacts (dark portions, blurriness, residues, specular reflections), then the analysis can be completed, but the detection accuracy of regions of interest is reduced
Solution Approach 1:
The image is divided into multiple regions, and each region is evaluated separately by the first classifier to generate region evaluation values. This segmentation allows the system to identify and handle different types of artifacts in different regions independently, improving overall detection accuracy by processing homogeneous regions separately.
Solution Approach 2:
The patent extracts and removes regions with low evaluation values (containing artifacts like dark portions, blurriness, residues, or specular reflections) before performing the final lesion evaluation. By taking out these problematic regions, the second classifier can focus on evaluating only the suitable regions, thereby improving detection accuracy.
2Measurement precision
If a single classifier evaluates the entire image, then the processing is simple, but the lesion evaluation accuracy is reduced due to artifacts
Solution Approach 1:
The image evaluation process is segmented into two distinct stages: first, region-level evaluation by the first classifier to identify suitable regions; second, lesion-level evaluation by the second classifier on extracted suitable regions. This segmentation improves measurement precision by separating artifact detection from lesion evaluation.
Solution Approach 2:
The first classifier acts as an intermediary between the raw image and the second classifier. It processes the entire image to generate region evaluation values, and the second classifier uses these values as a mediator to select and evaluate only the suitable regions, thereby improving lesion evaluation accuracy while managing complexity through functional decomposition.
3Reliability
If the entire image is used for lesion evaluation, then all potential lesions are covered, but artifact regions reduce the overall evaluation reliability
Solution Approach 1:
The system extracts only the suitable regions (those with high evaluation values from the first classifier) and excludes artifact-contaminated regions from the final lesion evaluation. This extraction process maintains evaluation reliability by ensuring that only high-quality regions are used, while the excluded regions are identified and set aside rather than completely discarded.
Solution Approach 2:
Different regions of the image are treated with different quality standards. Regions with high evaluation values (clear, artifact-free) are used for reliable lesion evaluation, while regions with low evaluation values (containing artifacts) are excluded. This local quality approach ensures that the evaluation reliability is maintained by using only high-quality local regions.
Data Source
AI summary
An endoscope system that illuminates a subject and captures light from the subject acquires an examination image based on an image signal captured by an endoscope, divides the examination image into a plurality of regions as an input image, inputs the input image divided into the plurality of regions to a first classifier to output region evaluation values for the plurality of regions, and inputs an input image in which the region evaluation values are added to the plurality of regions to a second classifier to output a lesion evaluation value.