Capsule Endoscopy Image Stream Segmentation for GIT Transition Detection
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
The existing capsule endoscopy (CE) systems require manual review of thousands of images, which is time-consuming and tiresome for physicians, and there is a need for efficient methods to automate the detection of transitions between gastrointestinal tract segments to reduce the number of images that need to be analyzed.
Innovation Solution
A system utilizing deep learning neural networks to analyze a stream of images from a capsule endoscopy device, refining transitions between gastrointestinal tract segments by comparing classification scores and identifying optimal transition points based on cumulative differences or fluctuations, allowing for automated detection and reduction of irrelevant images.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If manual review of all captured images is performed, then diagnostic accuracy is maintained, but reading time increases to half an hour to an hour and the task becomes tiresome
Solution Approach 1:
The patent segments the gastrointestinal tract into distinct anatomical regions (stomach, small bowel, colon) and uses deep learning models to automatically identify transition points between these regions. This segmentation allows the system to divide the large set of thousands of images into manageable regional groups, enabling physicians to focus review efforts on specific areas of interest rather than manually examining every image from the entire procedure.
Solution Approach 2:
The patent introduces an intermediary deep learning analysis system that processes images and identifies transition points between GIT segments. This intermediary system acts as a bridge between the raw image data and the physician's review process, automatically filtering and organizing images by anatomical region to reduce the reading burden while preserving diagnostic accuracy.
2Reliability
If all thousands of captured images are processed and reviewed, then complete diagnostic coverage is achieved, but the procedure becomes time-consuming and inefficient
Solution Approach 1:
The patent performs preliminary automated analysis of all captured images using deep learning models to identify transition points between gastrointestinal tract segments before the physician begins review. This preliminary action pre-organizes the image stream into anatomical regions, so when the physician accesses the study, the images are already segmented and ready for efficient review, maintaining complete diagnostic coverage while significantly improving procedure efficiency.
Solution Approach 2:
The patent replaces the mechanical process of manual image-by-image review with an automated deep learning system that uses neural networks to analyze images and identify anatomical transitions. This substitution of automated intelligence for manual mechanical review maintains reliable diagnostic coverage while dramatically increasing procedure efficiency by reducing reading time from half an hour to an hour down to minutes.
3Loss of time
If the number of images to be reviewed is reduced through automated transition detection, then reading time decreases, but the complexity of the system increases due to deep learning neural networks
Solution Approach 1:
The patent implements a self-service automated system where the deep learning models independently perform image analysis, transition point detection, and anatomical region segmentation without requiring manual configuration or intervention. The system serves itself by automatically processing the image stream and providing pre-organized results to the physician, reducing reading time while managing complexity through automation rather than manual processes.
Data Source
Figure 1
Figure 2
Figure 3
AI summary
In accordance with aspects of the present disclosure, a system includes at least one processor and at least one memory storing instructions which, when executed by the processor(s), cause the system to access images of a portion of a gastrointestinal tract (GIT) captured by a capsule endoscopy device; for each of the images, provide, by a deep learning neural network, scores for classifying the image to each of consecutive segments of the GIT; classify each image of a subset of the images, whose scores satisfy a confidence criterion, to one of the consecutive segments of the GIT; refine the classifications of the images in the subset by processing a signal over time corresponding to the classifications of the images in the subset; and estimate, among the images in the subset, a transition (1010) between two adjacent segments of the GIT based on the refined classifications of the images in the subset.