X-Ray Image Decomposition Using CT-Based Pseudo Projection Training
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing methods for decomposing medical images, such as X-ray images, struggle to accurately separate bones from soft tissues due to superimposition, requiring specialized imaging equipment or machine learning techniques that are not effective in recognizing anatomical structures or diseases.
Innovation Solution
A learning method that generates pseudo X-ray images from CT images and compares them to reduce errors, using machine learning to update parameters and integrate discrimination results, allowing for accurate decomposition of X-ray images into anatomical regions.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Productivity
If a simple X-ray image is captured, then the imaging process is simple and quick, but the bone and soft tissue are superimposed making anatomical structures difficult to recognize
Solution Approach 1:
The patent applies segmentation by dividing the superimposed X-ray image into separate anatomical regions (bone and soft tissue) using a trained neural network model. The model processes the input X-ray image and outputs segmented images where bones and soft tissues are separated, enabling clear recognition of anatomical structures while maintaining the simplicity and speed of the original imaging process.
2Measurement precision
If dedicated imaging equipment with multiple energy bands is used, then bone and soft tissue can be decomposed, but the device complexity and cost increase
Solution Approach 1:
The patent replaces the mechanical/imaging system (dedicated multi-energy X-ray equipment) with an information processing system (neural network model). Instead of using complex physical equipment with multiple energy bands to decompose images, the system uses a trained neural network that processes standard X-ray images to achieve the same decomposition effect, thereby eliminating the need for specialized imaging equipment.
Solution Approach 2:
The patent creates a virtual model (CT-based pseudo X-ray image) that replicates the appearance of a real X-ray image but with the advantage of having separated anatomical structures. By training the neural network to map from this virtual copied image to the actual X-ray image, the system learns to decompose real X-ray images without requiring the physical characteristics of multi-energy imaging equipment.
3Device complexity
If machine learning decomposition methods are used, then the need for specialized equipment is reduced, but the accuracy in recognizing anatomical structures and diseases decreases
Solution Approach 1:
The patent applies preliminary action by pre-training the neural network model using CT images as ground truth data before deploying it for actual X-ray image decomposition. The training phase creates a learned mapping relationship between X-ray images and their corresponding anatomical structures, enabling the model to accurately decompose new X-ray images without requiring specialized equipment or manual annotation during operation.
Solution Approach 2:
The patent uses CT-based pseudo X-ray images as an intermediary during the training process. These pseudo images serve as a bridge between the input X-ray images and the ground truth anatomical structures, allowing the neural network to learn the decomposition mapping more effectively. The intermediary helps the model understand the relationship between the projected X-ray appearance and the actual three-dimensional anatomical structures.
Data Source
Figure 1
Figure 2
Figure 3
AI summary
An embodiment of the present invention provides a learning method, a learning apparatus, and a learning program for decomposing a two-dimensional X-ray image into a variety of regions, and an image processing method, an image processing apparatus, and an image processing program using a result of learning. A learning method according to an aspect of the present invention includes generating a masked CT image from a three-dimensional CT image, and generating a third pseudo X-ray image from the masked CT image. Then, a second pseudo X-ray image generated by projecting an original CT image is compared with the third pseudo X-ray image as a reference, and a parameter for use in generating the second pseudo X-ray image is updated based on a comparison result. A processor repeats the processing of each step until a predetermined condition is satisfied.