Medical Image Search Using Added Finding Feature Matching
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing medical image search methods require manual labor to find similar images when a query image including a desired finding is not available, especially for comparing different diseases in the same anatomical part, which is impractical for users like doctors.
Innovation Solution
An image search device that uses a processor to derive added finding and normal feature amounts from a query image, comparing them with registered feature amounts in a database to automatically extract similar reference images based on similarity, utilizing encoding learning models to quantify and encode image features.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Ease of operation
If a user checks images stored in the database one by one to search for a query image, then the search can be performed, but the work requires a lot of labor and is not practical
Solution Approach 1:
The patent replaces the manual mechanical process of checking images one by one with an automated computer-based image search system. The system uses feature amount extraction, similarity calculation, and automated image retrieval to substitute human labor in the image search process, making the search operation efficient and practical.
Solution Approach 2:
The image search system performs self-service by automatically extracting features from query images, comparing them with database images, and retrieving similar images without requiring manual intervention. The system serves itself by autonomously completing the entire search process from query input to result delivery.
2Adaptability or versatility
If a query image including a desired finding is not at hand, then it is not possible to perform the search, but the user still needs to search for images with specific findings
Solution Approach 1:
The system allows users to perform preliminary actions by inputting finding information and selecting anatomical regions before the actual search begins. The query base image is prepared in advance with annotated findings, enabling the search to proceed even when a complete query image with the desired finding is not available.
Solution Approach 2:
The patent segments the image search process into independent components: finding information input, query base image preparation, feature amount derivation, and similarity comparison. This segmentation allows the system to handle partial information and reconstruct the search query step by step, improving adaptability when complete query images are unavailable.
3Measurement precision
If the system searches for images with specific findings, then the search becomes more targeted, but the complexity of processing and comparing images increases
Solution Approach 1:
The patent extracts and isolates specific features from images, such as finding regions, anatomical structures, and visual characteristics. By taking out these key features and representing them as feature amounts, the system simplifies the comparison process while maintaining precision in searching for images with specific findings.
Solution Approach 2:
The system transforms complex image data into simplified parameter representations (feature amounts) that can be easily compared. By changing the representation from raw image pixels to extracted feature parameters, the system reduces processing complexity while maintaining the ability to perform precise searches for specific findings.
Data Source
AI summary
A processor receives an input of finding information indicating at least one finding, which is desired to be searched for, on a query base image, which includes a part desired to be searched for and is a source of a query image, to derive the query image to which the finding has been added, derives at least one added finding feature amount indicating an image feature for the added finding, and derives a query normal feature amount indicating an image feature for a normal region included in the part in the query base image. The processor derives similarities between the query image and a plurality of reference images including findings on the basis of comparisons between the added finding feature amount and the query normal feature amount and the like with reference to an image database in which the plurality of reference images have been registered and extracts the reference image that is similar to the query image as a similar image from the image database on the basis of the similarities.