X-Ray CT Contribution Mapping for Interpretable Texture Features
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing medical imaging technologies struggle to effectively utilize texture feature quantities for accurate disease diagnosis and prognosis, as the significance of these features is often unclear to health professionals, and existing methods may present misleading frequency-based information rather than influence-based insights.
Innovation Solution
An X-ray CT apparatus and medical image processing system that calculates texture feature quantities, specifies a contribution region with high influence on the diagnosis, and displays this region to provide clear, influence-based grounds for disease condition estimation using a pre-trained model.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If texture feature quantities are extracted and input to machine learning models for disease diagnosis, then diagnostic accuracy can be improved, but the interpretability and clarity of feature significance remain poor
Solution Approach 1:
The patent segments the medical image into multiple regions and calculates texture feature quantities for each region separately. It then determines the influence degree of each region on the overall diagnostic result by analyzing the contribution of regional texture features to the machine learning model's output. This segmentation allows the system to maintain high diagnostic accuracy while providing interpretable information about which specific regions drive the diagnostic conclusion.
Solution Approach 2:
The patent introduces an intermediary mechanism that calculates the influence degree of each region by measuring the correlation between regional texture feature quantities and the diagnostic output. This intermediary layer bridges the gap between the raw image data and the diagnostic model, providing transparent explanation about region importance without compromising the accuracy of the machine learning-based diagnosis.
2Quantity of substance
If frequency-based information is presented to show feature distribution, then quantitative analysis is provided, but misleading information may be presented rather than influence-based insights
Solution Approach 1:
The patent changes the parameter used for analysis from frequency-based statistics to influence-degree-based metrics. Instead of presenting how often certain texture patterns occur, the system calculates and presents the influence degree of each region, which measures the actual impact of that region's texture features on the diagnostic result. This parameter change eliminates misleading frequency information and provides accurate influence-based insights.
3Measurement precision
If the entire medical image is processed to extract texture features, then comprehensive analysis is achieved, but processing time and computational complexity increase
Solution Approach 1:
The patent divides the medical image into multiple smaller regions and processes each region independently to extract texture feature quantities. By segmenting the image, the system can parallelize the processing across different regions, significantly reducing total processing time while maintaining comprehensive analysis through the aggregation of regional results. This segmentation approach also reduces the computational complexity of processing the entire image at once.
Data Source
AI summary
An X-ray CT apparatus according to an embodiment includes a photographing system configured to photograph a subject, and processing circuitry. The processing circuitry generates X-ray CT image data by performing reconstruction processing on projection data output from the photographing system, calculates an image feature quantity indicating a feature of a relation between one pixel on the X-ray CT image data and one or a plurality of pixels other than the former pixel, specifies a contribution region indicating a region on the X-ray CT image data delineating a feature having a high influence degree for a calculation result of the image feature quantity, and causes a display apparatus to display the X-ray CT image data representing the contribution region.