Medical Image Classification Apparatus with Dynamic Conversion Definitions
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current medical image classification methods struggle to accurately classify pixels into case regions according to individual doctors' interpretations and purposes, as they rely on predetermined criteria and require extensive teacher data for learning, making it difficult to adapt to varying medical practices and diagnostic goals.
Innovation Solution
A medical image classification apparatus that outputs evaluation values for each pixel, allowing for conversion of initial classification results using customizable conversion definition information, which can be tailored for specific doctors, purposes, or diseases, enabling flexible classification into multiple types of case regions.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Extent of automation
If predetermined criteria such as sample histograms and statistics are used for classification, then classification can be performed automatically, but it becomes difficult to classify pixels according to individual doctors' interpretations and purposes
Solution Approach 1:
The patent makes the classification criteria dynamic by allowing conversion between different classification definitions. The system stores multiple conversion definition information sets corresponding to different doctors' interpretations and purposes, and can dynamically switch between them. This enables the classification system to adapt to individual doctors' needs while maintaining automated operation, resolving the contradiction between automation and adaptability.
2Measurement precision
If a discriminator is used for classification, then classification accuracy can be improved through learning, but it requires a large amount of teacher data which is difficult to prepare according to individual doctors or purposes
Solution Approach 1:
The patent performs preliminary classification using a discriminator to generate initial classification results, and then applies conversion definition information to adjust these results according to specific doctors' interpretations. This two-stage approach allows the system to benefit from the discriminator's learning capability while avoiding the need to prepare extensive teacher data for each individual doctor or purpose, as the conversion definitions can be prepared in advance.
3Adaptability or versatility
If multiple conversion definition information are stored for different doctors and purposes, then classification can be tailored to individual needs, but the system complexity increases
Solution Approach 1:
The patent implements a universal conversion mechanism that can handle multiple different classification definitions through a single unified system. The conversion definition information is stored in a standardized format that can accommodate various doctors' interpretations and purposes without requiring separate processing systems. This multi-functional approach enables high adaptability while maintaining relatively simple system architecture.
Data Source
AI summary
A first classification unit outputs a plurality of evaluation values indicating the possibility of being each of a plurality of types of case regions for each pixel of a three-dimensional image. Based on selected conversion definition information, a second classification unit converts a first classification result for each pixel of the three-dimensional image based on the plurality of evaluation values, and outputs a second classification result for each pixel of the three-dimensional image.