Medical CT Image Spectral Conversion via Machine Learning
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current X-ray CT imaging technologies require re-exposure to radiation to acquire images with different spectral information, which increases patient radiation burden and is inefficient.
Innovation Solution
A medical information processing method and apparatus that generates a trained model using machine learning to convert X-ray CT images from one spectral energy to another, allowing for the acquisition of images with different spectral information without additional radiation exposure.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Adaptability or versatility
If another X-ray CT image is acquired in another X-ray spectrum to obtain different spectral information, then the diagnostic capability is improved, but the patient's radiation exposure increases
Solution Approach 1:
The patent creates a virtual copy of the X-ray CT image in a different spectral domain by training a machine learning model to map between spectral spaces. The model learns the transformation from a first spectral space to a second spectral space, enabling generation of synthetic images that mimic what would be obtained from actual imaging at different energies without exposing the patient to additional radiation.
Solution Approach 2:
The patent changes the spectral parameter (X-ray energy) of the image data through machine learning transformation. By adjusting the spectral information in the digital domain rather than physically changing X-ray energy during imaging, the system generates images with different spectral characteristics while maintaining the same physical imaging conditions and radiation dose.
2Adaptability or versatility
If multiple X-ray CT images with different spectral information are acquired, then the diagnostic flexibility is improved, but the imaging time and efficiency are worsened
Solution Approach 1:
Instead of acquiring multiple physical images at different spectral settings, the system creates virtual copies through machine learning transformation. A single acquired image serves as the source, and the trained model generates multiple spectral variants computationally, eliminating the need for repeated imaging procedures and improving workflow efficiency.
Solution Approach 2:
The machine learning model is trained in advance on pairs of images with different spectral information. This preliminary training phase enables the system to quickly generate spectral-transformed images during clinical use without requiring actual re-imaging, thus improving productivity while maintaining diagnostic flexibility.
Data Source
AI summary
A medical information processing method according to an embodiment includes: acquiring an X-ray CT image (I1) and spectral information on imaging of the X-ray CT image (I1); acquiring sets of distribution data (D11, D12, and D13) on substances in the X-ray CT image by performing segmentation of the X-ray CT image (I1) according to substance; acquiring plural sets of forward projection data (P11, P12, and P13) on the respective substances by performing forward projection processes for the sets of distribution data (D11, D12, and D13) on the basis of the spectral information and attenuation coefficients for the respective substances; and generating a trained model (M1) by machine learning based on the plural sets of forward projection data (P11, P12, and P13) and raw data (R1) used in generation of the X-ray CT image (I1).