Pseudo 2D Breast Image Generation with Compression-Matched Training
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing image generation models trained with combinations of images from tomosynthesis and normal two-dimensional imaging in different compression states result in decreased accuracy of pseudo two-dimensional image generation due to lower correlation between input and output information.
Innovation Solution
An image generation apparatus and method that trains an image generation model using paired images from tomosynthesis and normal two-dimensional imaging performed in the same compression state, correcting magnification ratios and radiation conditions, and classifying images by categories such as person attributes and imaging settings to improve accuracy.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Productivity
If images from tomosynthesis and normal two-dimensional imaging in different compression states are used for training, then the training process can proceed with available images, but the generation accuracy of pseudo two-dimensional images decreases due to lower correlation between input and output information
Solution Approach 1:
The patent applies local quality by selectively matching images based on compression state. Specifically, it identifies and pairs tomosynthesis images with normal two-dimensional images that share the same compression state, creating high-quality training pairs where the breast anatomy is in consistent conditions. This selective matching ensures that only locally optimal image pairs (same compression state) are used for training, thereby maintaining high correlation between input and output while preserving training efficiency
Solution Approach 2:
The patent changes the parameter of image selection criteria from general availability to specific compression state matching. By introducing compression state as a key selection parameter, the system filters and pairs images based on this specific attribute, ensuring that training data consists of image pairs where the breast was imaged under identical compression conditions. This parameter-based selection resolves the contradiction by maintaining both training productivity and generation accuracy
2Quantity of substance
If multiple images are input to the image generation model, then more information is provided for processing, but the correlation between input information and output information decreases, reducing generation accuracy
Solution Approach 1:
The patent extracts and removes redundant or mismatched images from the input set. Specifically, it filters out tomosynthesis images that do not have corresponding normal two-dimensional images in the same compression state. By extracting only the relevant, correlated image pairs and discarding unrelated images, the system maintains high information quality and correlation, thereby preserving generation accuracy while still providing sufficient training information
Applied Scientific Principles
This section explains which scientific principles are used to turn an abstract innovation direction into a practical engineering solution.
Function Achieved in This Case
Enhances the generation accuracy of pseudo two-dimensional images by maintaining high correlation between input and output information, leading to improved reproduction of breast lesion shapes.
Implementation Method 1
an image generation model has been trained in advance using a plurality of combinations of the composite two-dimensional image and a normal two-dimensional image captured by irradiating, with radiation, the breast in a state of being compressed by the compression member during the tomosynthesis imaging for obtaining the composite two-dimensional image
Data Source
Figure 1~2
Figure 3
Figure 4~5
AI summary
Provided are an image generation apparatus, an image generation method, and a program capable of improving the generation accuracy of a pseudo two-dimensional image, as compared with a case of using an image generation model that has been trained using a combination of a composite two-dimensional image and a normal two-dimensional image captured separately from tomosynthesis imaging for obtaining the composite two-dimensional image. An image processing apparatus as an image generation apparatus includes an image generation model that has been trained in advance using a plurality of combinations of a composite two-dimensional image and a normal two-dimensional image captured by irradiating, with radiation, a breast in a state of being compressed by a compression member during tomosynthesis imaging for obtaining the composite two-dimensional image.