Learned Model for Energy-Subtraction Image Generation
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Spectral imaging of the time-division type requires multiple images of the same site in a short period, leading to increased exposure dose and noise in medical imaging, while existing noise reduction methods fail to effectively reduce original noise in individual energy images.
Innovation Solution
An image processing apparatus that generates energy-subtraction images using a learned model, which improves image quality by combining images obtained with different radiation energies and artificially calculated noise, reducing the exposure dose and noise in medical imaging.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If multiple images are taken with equivalent dose for normal imaging, then spectral imaging can be performed, but the exposure dose of the subject increases
Solution Approach 1:
The patent applies partial action by taking multiple images with a reduced dose per image (less than normal imaging dose) rather than taking fewer images with full dose. The cumulative dose across multiple images is controlled to be comparable to or less than a single normal imaging dose, while still obtaining sufficient signal for spectral analysis through the learning model processing
Solution Approach 2:
The patent uses a learning model trained on pairs of low-dose and high-quality images to generate synthetic high-quality images from the acquired low-dose images. This creates a copy of the desired high-quality image output without requiring the subject to receive the high dose that would be needed to capture the reference high-quality image
2Object-affected harmful factors
If the radiation dose per image is reduced, then the exposure dose increases less, but the noise intensity increases and image quality decreases
Solution Approach 1:
The patent introduces a learning model as an intermediary between the acquired low-dose images and the final output images. This learning model processes multiple low-dose images and generates high-quality output images by learning the mapping from low-dose to high-quality images, effectively mediating the trade-off between dose reduction and image quality maintenance
Solution Approach 2:
The patent combines multiple low-dose images through the learning model to produce a single high-quality output image. By merging information from multiple low-dose acquisitions and processing them through the learned model, the system achieves image quality comparable to high-dose single images while keeping the cumulative exposure dose lower
3Object-affected harmful factors
If existing noise reduction methods are used on individual energy images, then noise can be reduced, but the original noise of individual energy images cannot be effectively reduced
Solution Approach 1:
The patent performs preliminary action by training the learning model in advance using pairs of low-dose images and corresponding high-quality reference images. This pre-training enables the model to learn effective noise reduction patterns and image quality enhancement strategies before actual imaging, allowing it to effectively reduce noise in the acquired low-dose images without requiring post-processing of individual energy images
Data Source
AI summary
An image processing apparatus is provided that includes: an obtaining unit configured to obtain a plurality of images relating to different radiation energies; and a generating unit configured to generate at least one of energy-subtraction images based on the plurality of images using a learned model, wherein the learned model is obtained using a first image obtained using a radiation and a second image obtained by improving image-quality of the first image or by adding a noise which has been artificially calculated to the first image.