Capsule Endoscopy Image Selection Using Grouped Pathology Scores
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
The challenge of efficiently reviewing large volumes of images captured during capsule endoscopy, such as those of the GI tract, where important pathologies like polyps and bleeding sites may be missed due to the overwhelming number of images, necessitating a method to highlight a smaller, clinically relevant set while ensuring all pathologies are represented.
Innovation Solution
A method for selecting images by grouping similar images and identifying a greatest general score (GS) within each subgroup, iteratively selecting images based on this score, and modifying scores based on distance and relevance to reduce redundancy, ensuring a small set of clinically important images is displayed.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If all images captured during capsule endoscopy are reviewed, then complete pathology detection is achieved, but review time and effort increase significantly
Solution Approach 1:
The patent segments the large set of captured images into multiple sub-sets based on similarity criteria. Images are grouped together if they share common characteristics (e.g., similar anatomy, pathology type, or visual features), allowing physicians to review representative images from each segment rather than every individual image, thus reducing review time while maintaining detection completeness.
Solution Approach 2:
The patent extracts and highlights only the most relevant images that likely contain pathologies or are representative of the GI tract anatomy. By using algorithms to identify and extract critical frames based on predefined criteria (such as changes in anatomy, presence of abnormalities, or visual distinctiveness), the system presents a reduced set of high-value images for physician review, eliminating redundant images while preserving diagnostic information.
2Loss of time
If a small number of images are selected for display, then review time is reduced, but risk of missing pathologies increases
Solution Approach 1:
The patent applies local quality by differentiating the treatment of different image sets. Instead of uniformly reviewing all images, the system identifies and prioritizes specific local regions or image characteristics that are most likely to contain pathologies. Images are weighted and selected based on their local significance, ensuring that a small sample represents the most critical areas while reducing overall review time.
3Quantity of substance
If images are grouped by similarity, then redundant images are reduced, but distinction between different pathology types may be lost
Solution Approach 1:
The patent implements dynamic grouping where the similarity criteria and image selection process can adapt based on the content being analyzed. The system dynamically adjusts grouping parameters to maintain sensitivity to different pathology types while still reducing the overall image count. This allows the same mechanism to effectively handle diverse pathology presentations without losing discriminatory capability.
Data Source
Figure 1
Figure 2
Figure 3~4B
AI summary
A method executed by a system for selecting images from a plurality of image groups originating from a plurality of imagers of an in-vivo device includes calculating, or otherwise associating, a general score (GS) for images of each image group, to indicate the probability that each image includes at least one pathology, dividing each image group into image subgroups, identifying a set Set(i) of maximum general scores (MGSs), a MGS for each image subgroup of each image group; and selecting images for processing by identifying a MGS|max in each set S(i) of MGSs; identifying the greatest MGS|max and selecting the image related to the greatest MGSlmax. The method further includes modifying the set Set(i) of MGSs related to the selected image, and repeating the steps described above until a predetermined criterion selected from a group consisting of a number N of images and a score threshold is met.