Deep Learning Virtual Monochromatic X-ray Image Correction
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Dual Energy CT (DECT) technology suffers from image quality degradation, particularly increased noise and deteriorated image texture when reconstructing low-energy virtual monochromatic X-ray images, limiting its effectiveness in medical diagnostics.
Innovation Solution
An image generation device utilizing deep learning to infer high-quality virtual monochromatic X-ray images of the first energy by inputting reconstructed images of the second energy into a trained model, thereby inheriting better image quality characteristics from the second energy images to improve the quality of the first energy images.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Adaptability or versatility
If DECT technique is used to reconstruct virtual monochromatic X-ray images, then material distinction capability is improved, but image quality deteriorates due to increased noise and deterioration in image texture
Solution Approach 1:
A trained model (intermediary) is introduced between the DECT reconstruction process and the final image output. The model takes the reconstructed virtual monochromatic X-ray image as input and generates a corrected image with improved quality, effectively mediating the trade-off between material distinction capability and image quality
Solution Approach 2:
The invention changes the parameter of image quality by applying a correction process that modifies the reconstructed image. The trained model adjusts parameters such as noise levels and texture characteristics while preserving the material distinction information, thereby improving overall image quality without sacrificing the core DECT capability
2Ease of manufacture
If deep learning is used to infer virtual monochromatic X-ray images from SECT images, then DECT equipment requirement is eliminated, but image quality degradation occurs
Solution Approach 1:
A trained model serves as an intermediary that processes SECT images and generates corrected virtual monochromatic X-ray images. This intermediary layer improves the quality of deep learning-based inference images without requiring DECT hardware, thus maintaining ease of implementation while improving image quality
Solution Approach 2:
The invention replaces the need for complex DECT hardware (mechanical system) with a software-based deep learning model. The trained model performs the function of material distinction and image quality enhancement purely through computational processing, eliminating the need for specialized equipment
Data Source
AI summary
A computed tomography (CT) system with one or a plurality of processors to perform operations. The operations include reconstructing a series of virtual monochromatic X-ray images of the first energy and a series of virtual monochromatic X-ray images of the second energy based on data collected from an imaging subject, inputting the input image generated based on the reconstructed virtual monochromatic X-ray images of the second energy to the trained model and using the trained model to infer a series of virtual monochromatic X-ray image of the first energy, and generating a corrected series of virtual monochromatic X-ray images of the first energy based on the first reconstructed series of virtual monochromatic X-ray images of the first energy and the inferred series of virtual monochromatic X-ray images of the first energy.