Capsule Endoscopy Image Deduplication for Event Indicator Review
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
The existing capsule endoscopy systems require physicians to review thousands of images, which is time-consuming and tiresome, and often include false detections of event indicators like ileocecal valves being misclassified as colon polyps, leading to inefficient and prolonged report generation.
Innovation Solution
A system utilizing a consecutive-image tracker and a false detection filter to deduplicate images showing the same event indicator, such as colon polyps, by retaining only one track of the occurrence and removing false detections, thereby reducing the number of images for review.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If all captured images are reviewed by physicians, then no event indicators are missed, but the reading time becomes excessively long and the process becomes tiresome
Solution Approach 1:
The system extracts and selects only the most relevant images from the thousands of captured images using an iterative selection process. A selection module identifies seed images with high event indicator presence scores, then selectively reviews only those images and their neighboring frames, extracting the essential diagnostic information while discarding redundant images.
Solution Approach 2:
The system changes the parameter of image selection from reviewing all images to reviewing only images above a certain score threshold. The iterative selection process dynamically adjusts selection criteria, focusing on images with highest event indicator presence scores and their temporal neighbors, thereby changing the review parameter from quantity-based to quality-based selection.
2Reliability
If all images including duplicates are reviewed, then all potential event indicators are captured, but the number of images increases unnecessarily prolonging the review process
Solution Approach 1:
The system performs preliminary deduplication by tracking objects across consecutive images before the physician review stage. An object tracking module identifies and merges duplicate detections of the same event indicator, retaining only representative images from each tracked sequence. This preliminary action removes redundant images before they reach the review process, improving efficiency without sacrificing detection reliability.
3Reliability
If all detected event indicators are reviewed, then false detections are not missed, but the review process becomes more complex and time-consuming
Solution Approach 1:
The system performs preliminary filtering using multiple detection models with different specializations. Some models are trained to detect specific event indicator types while others detect general anomalies. The results are merged and deduplicated before presentation to the physician, performing the complex filtering work in advance to simplify the final review process.
Solution Approach 2:
The detection system is segmented into multiple specialized detection models, each trained for specific event indicator types or detection tasks. This segmentation allows parallel processing of different image sets and enables the system to handle false detections from different models separately before merging results, reducing the complexity of the overall review process.
Data Source
AI summary
A system for selecting images of an event indicator includes a processor and a memory storing instructions which, when executed, cause the system to: access images of a portion of a gastrointestinal tract captured by a capsule endoscopy device; for each of the images, access one or more scores indicating a presence of an event indicator; select seed images from among the images based on the one or more scores; deduplicate the seed images for images showing the same occurrence of the event indicator, where the deduplicating utilizes a consecutive-image tracker; and present the deduplicated seed images in a graphical user interface to display potential occurrences of the event indicator.