Capsule Endoscope Image Labeling via Automated Feature Extraction
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
The existing methods for managing in-vivo images captured by capsule endoscopes are labor-intensive, requiring manual text input for labeling abnormal findings, which is inefficient due to the large number of images generated during an examination, typically around 60,000 images per session.
Innovation Solution
An image management apparatus and method that calculates feature quantities of images, extracts relevant additional information, generates icons for display, and allows users to associate images with labels through intuitive operations like dragging and dropping, reducing the burden of manual labeling.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If manual text input is used for labeling each in-vivo image, then labeling accuracy and completeness are improved, but work load and time consumption increase significantly
Solution Approach 1:
The system automatically generates labels by extracting text from medical reports and automatically associates them with corresponding in-vivo images based on temporal synchronization, eliminating the need for manual labeling while maintaining accuracy through automated image-report matching
Solution Approach 2:
The system performs preliminary extraction of label information from medical reports before the actual labeling process, and pre-establishes the correspondence between reports and images through temporal synchronization, so that labeling can be completed automatically without manual intervention
2Measurement precision
If manual text input is used for labeling, then labeling precision is maintained, but productivity decreases due to the large number of images (around 60,000 images per session)
Solution Approach 1:
The system automatically processes the large volume of images by extracting labels from medical reports and automatically matching them with images based on capture timing, achieving both high precision and high productivity without manual intervention
Solution Approach 2:
The patent replaces the mechanical manual text-input process with an automated information extraction and matching system that uses computational methods to associate labels with images, dramatically increasing productivity while maintaining precision through automated verification
3Productivity
If automated label extraction is implemented, then productivity is improved, but device complexity increases
Solution Approach 1:
The system uses a single integrated approach that simultaneously extracts text from medical reports, synchronizes with image capture timing, and automatically assigns labels to images, reducing overall system complexity by combining multiple functions into one unified process rather than requiring separate complex subsystems
Data Source
AI summary
An image management apparatus includes: a storage unit that stores a plurality of types of additional information assigned to a plurality of images; a calculation unit that calculates a feature quantity of each of the images; an extracting unit that extracts, based on the feature quantity, additional information of the plurality of types of additional information; a control unit that generates one or more icons corresponding to the one or more types of additional information and displays the icons on a screen; an input unit that receives input according to a user's operation; a selecting unit that selects an image according to the signal; and an assigning unit that assigns to the selected image, when input of an operation signal associating the image selected by the image selecting unit with an icon selected by the user is received, additional information corresponding to the icon associated with the selected image.