Capsule Endoscopy Image Selection for Celiac Disease Diagnosis
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current capsule endoscopy systems require healthcare professionals to manually review thousands of images for indicators of celiac-like diseases, such as villous atrophy, which is time-consuming and inefficient, potentially leading to missed important information.
Innovation Solution
A method and system that utilize classification scores from deep learning or classical machine learning classifiers to select and display a subset of images from a stream of gastrointestinal tract images, focusing on the proximal small bowel portion, to aid clinicians in diagnosing celiac-like diseases by identifying indicators like villous atrophy, scalloping, and mosaic patterning.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If manual review of all images is performed, then diagnostic accuracy is maintained, but time consumption increases significantly
Solution Approach 1:
The system performs preliminary automated analysis of all images using machine learning classifiers before the physician reviews them. Classification scores are calculated in advance for each image, identifying those most likely to contain indicators of celiac-like disease. This preliminary filtering action reduces the review burden while maintaining diagnostic accuracy by ensuring the physician focuses on pre-identified suspicious images.
Solution Approach 2:
The patent introduces an intermediary automated classification system between the raw image stream and the physician's review process. The machine learning classifier acts as a mediator that processes all images, generates classification scores, and presents a curated subset to the physician. This intermediary layer maintains diagnostic accuracy by systematically identifying relevant images while significantly reducing the time the physician must spend reviewing images.
2Reliability
If all images are reviewed to ensure no important information is missed, then diagnostic reliability improves, but productivity decreases
Solution Approach 1:
The system applies partial action by reviewing only a subset of images that are most likely to contain diagnostic information. Rather than reviewing all images excessively, the machine learning classifier identifies and prioritizes the most relevant images based on classification scores. This partial review approach maintains diagnostic reliability by focusing on high-probability cases while improving productivity by reducing the total number of images requiring physician attention.
Solution Approach 2:
The system changes the parameter of image selection from uniform review of all images to selective review based on classification scores. By introducing a scoring parameter that ranks images by likelihood of containing indicators, the system transforms the review process from exhaustive to targeted. This parameter-based selection maintains reliability by systematically identifying relevant cases while enhancing productivity through efficient prioritization.
3Productivity
If the number of images for review is reduced, then time efficiency improves, but risk of missing important information increases
Solution Approach 1:
The system implements feedback through classification scores that indicate the likelihood of each image containing indicators of celiac-like disease. The machine learning classifier provides quantitative feedback about image content, allowing the physician to prioritize review based on this feedback. This feedback mechanism reduces the risk of missing important information by systematically identifying suspicious images while improving efficiency by directing attention to high-priority cases.
Solution Approach 2:
The patent replaces the mechanical system of manual image review with an automated machine learning classification system. Instead of relying on the physician's manual assessment of all images, the system uses computational algorithms to pre-analyze and rank images. This substitution reduces the risk of missed indicators through systematic automated analysis while improving efficiency by presenting only the most relevant images to the physician for final assessment.
Data Source
AI summary
A method for detecting indicators of a disease characterized by a presence of villous atrophy in images of a gastrointestinal tract (GIT), includes accessing a consecutive set of images of a portion of the GIT comprising a small bowel. Each image is associated with one or more classification scores, and each classification score is indicative of the associated image including a respective indicator of a disease characterized by the presence of villous atrophy. The method further includes selecting a subset of images from the consecutive set of images based on the one or more classification scores of each image of the consecutive set of images, identifying a segment of images which includes all of the images that show a proximal portion of the small bowel, selecting a plurality of images from the identified segment of images that represent the proximal portion of the small bowel, and displaying the selected images.