Capsule Endoscopy Image Clustering for Boundary Detection
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
In capsule endoscopic examinations, the large number of images captured by a capsule endoscope poses a significant burden for image interpreters to find and distinguish region boundaries, as existing automatic distinction methods include errors and require time-consuming verification.
Innovation Solution
An endoscopic image observation system that uses a processor to output an accuracy score for each image, groups images into clusters based on this score, and identifies candidate images for region boundaries, facilitating efficient image interpretation by highlighting likely boundary images.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Productivity
If automatic region distinction is performed based on image color, then the efficiency of image observation is improved, but the accuracy of boundary identification deteriorates due to errors in automatic distinction
Solution Approach 1:
The patent segments the images into multiple clusters based on color characteristics, with each cluster representing a specific region. By dividing the images into clusters and identifying boundary images between clusters, the system achieves both automated processing efficiency and accurate boundary identification, resolving the contradiction between productivity improvement and measurement precision deterioration.
2Loss of time
If the number of images to be reviewed is reduced by automatic distinction, then the time burden on interpreters is reduced, but the reliability of boundary identification deteriorates due to errors in automatic results
Solution Approach 1:
The patent introduces boundary images as intermediary elements between clusters. These boundary images serve as a bridge between automated clustering results and reliable boundary identification. By using boundary images with specific color characteristics that represent transitions between regions, the system maintains both time efficiency and reliability in boundary identification.
3Reliability
If all images are manually reviewed to ensure accurate boundary identification, then the reliability of boundary identification is maintained, but the productivity of image interpretation deteriorates due to the enormous number of images
Solution Approach 1:
The patent extracts and highlights only the boundary images from the large set of captured images. By taking out the specific boundary images that contain region transition information and presenting them separately, the system enables interpreters to quickly identify boundaries without reviewing all images, thus maintaining reliability while improving productivity.
4Reliability
If the scope of images to be searched is expanded to ensure complete boundary detection, then the reliability of boundary identification is improved, but the loss of time increases due to extensive searching requirements
Solution Approach 1:
The patent performs preliminary clustering of images by color characteristics before boundary identification. By pre-grouping images into clusters and pre-identifying potential boundary images between clusters, the system reduces the search scope for interpreters. This preliminary action ensures complete boundary detection within a reduced time frame, resolving the contradiction between reliability and time loss.
Data Source
AI summary
An endoscopic image observation system supports the observation of a plurality of images captured by a capsule endoscope. The endoscopic image observation system includes a distinguishing unit that outputs an accuracy score indicating the likelihood that each of the plurality of images represents an image of a region sought to be distinguished; a grouping unit that groups the plurality of images into a plurality of clusters in accordance with the accuracy score; and an identification unit that identifies a candidate image for a boundary of the region from among the plurality of images in accordance with the grouping into the plurality of clusters.