Medical Image Representative Selection via Wavelet Centroid Distance
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing medical image analysis systems face challenges in identifying representative images and radiographic interpretation information, leading to difficulties in diagnosing patient symptoms accurately, as they often retrieve a large number of similar images with varying symptoms, making it hard to determine a representative image for guideline purposes.
Innovation Solution
The system calculates wavelet features from past medical images, extracts keywords from radiographic interpretation information, classifies images into groups, calculates centroid vectors, and determines the image with the shortest spatial distance to the centroid as the representative image for each group, along with its associated radiographic interpretation.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If similar case search is performed using feature values from diagnosis object images, then the ability to find similar images is improved, but the difficulty in identifying the most representative image increases
Solution Approach 1:
The patent transforms the image selection problem from qualitative medical judgment to quantitative measurement by calculating spatial distances between centroid vectors and image feature vectors. This parameter change enables automatic identification of representative images through mathematical optimization rather than subjective evaluation.
Solution Approach 2:
The patent replaces the mechanical/subjective process of medical expert judgment with an automated computational system using wavelet features, centroid calculations, and spatial distance measurements. This substitution eliminates reliance on individual expertise while maintaining diagnostic quality.
2Quantity of substance
If multiple similar images are retrieved for case comparison, then the comprehensiveness of diagnostic information is improved, but the complexity of selecting a representative image increases
Solution Approach 1:
The patent simplifies the selection process by changing from complex qualitative assessment to simple quantitative comparison of spatial distances. The representative image is automatically identified as the one with the minimum spatial distance to the centroid vector, eliminating the need for complex selection algorithms.
Solution Approach 2:
The system performs self-service by automatically identifying representative images through mathematical optimization without requiring manual intervention. The centroid-based approach enables the system to autonomously select the most representative image from each group based on calculated spatial distances.
3Measurement precision
If representative images are selected based on medical expertise, then the diagnostic accuracy is improved, but the time required for image selection increases
Solution Approach 1:
The patent performs preliminary action by pre-calculating wavelet features and centroid vectors for all images in the database. This preprocessing enables rapid identification of representative images through simple spatial distance calculations, eliminating the need for time-consuming expert review while maintaining diagnostic accuracy.
Solution Approach 2:
The patent substitutes the time-consuming manual evaluation process with automated computational methods using wavelet transforms and vector calculations. This mechanical substitution dramatically reduces selection time while preserving diagnostic quality through objective mathematical criteria.
Data Source
AI summary
A technique for generating a representative image representing a case and radiographic interpretation information for each case includes calculating wavelet features of a plurality of images that have been taken and stored in the past. The calculated wavelet features and extracted keywords are stored in association with the stored images. The stored images are classified on the basis of the extracted keywords to generate a plurality of groups. For each of the generated groups, a centroid vector of wavelet feature-based feature vectors of respective images corresponding to the keywords is calculated and a spatial distance between the calculated centroid vector and each of the wavelet feature-based feature vectors is calculated. For each of the groups, the image for which the calculated spatial distance is the shortest and the radiographic interpretation information associated with the image is stored as a representative image of that group.