Neural Network CT Slice Denoising for Low-Dose Image Quality
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing CT image reconstruction methods for low-dose scans are computationally intensive and require expensive hardware, leading to suboptimal image quality due to factors like high quantum noise, scanning geometry, and non-ideal physical phenomena, which are challenging to model accurately.
Innovation Solution
Utilizing deep learning networks, particularly convolutional neural networks (CNNs), to process reconstructed images or sinograms, tailored for specific CT scanning methods and conditions, optimizing noise reduction and artifact mitigation through offline training and customized networks for different noise levels and anatomical structures.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Manufacturing precision
If model based iterative image reconstruction or sinogram restoration methods are used to improve low-dose CT image quality, then image quality is improved, but computational time increases and hardware costs increase
Solution Approach 1:
The patent applies preliminary action by training deep learning networks in advance using high-dose CT images as ground truth. The trained network is then deployed to rapidly process low-dose images without requiring intensive computational resources during actual reconstruction, thus achieving high image quality with reduced computational time.
Solution Approach 2:
The patent uses copying by training the deep learning network to learn the mapping from low-dose to high-dose image characteristics. The network copies the noise reduction and artifact suppression capabilities from high-dose reference images, enabling fast reconstruction with quality comparable to expensive iterative methods.
2Manufacturing precision
If model based iterative image reconstruction or sinogram restoration methods are used to improve low-dose CT image quality, then image quality is improved, but hardware costs increase
Solution Approach 1:
The patent replaces complex hardware-based iterative reconstruction systems with a software-based deep learning approach. The trained neural network runs on standard computing hardware, eliminating the need for expensive specialized hardware while maintaining image quality through learned patterns from training data.
Solution Approach 2:
The patent uses computationally inexpensive deep learning inference after training, replacing expensive iterative reconstruction hardware. The trained model can be deployed on cost-effective hardware platforms, making high-quality low-dose reconstruction accessible without expensive equipment.
3Productivity
If deep learning networks are used to process reconstructed images, then computational time is reduced and hardware costs decrease, but image quality may be degraded without proper training
Solution Approach 1:
The patent ensures high image quality by performing preliminary training of the deep learning network using paired low-dose and high-dose images. This offline training phase captures the complex mappings needed to preserve image quality, while the actual reconstruction benefits from fast inference.
Solution Approach 2:
The patent implements feedback by using high-dose CT images as ground truth during the training phase. The network learns to minimize the difference between its output and the reference high-dose images, ensuring that the fast reconstruction maintains diagnostic quality through supervised learning.
Data Source
AI summary
A method and apparatus is provided that uses a deep learning (DL) network to reduce noise and artifacts in reconstructed medical images, such as images generated using computed tomography, positron emission tomography, and magnetic resonance imaging. The DL network can operate either on pre-reconstruction data or on a reconstructed image. The DL network can be an artificial neural network or a convolutional neural network (e.g., using a three-channel volumetric kernel architecture). Different neural networks can be trained depending on the noise level, scanning protocol, or the anatomic, diagnostic or clinical objective of the reconstructed image (e.g., by partitioning the training data into noise-level range and training respective DL networks for each range). Further, the DL networks can be trained to mitigate artifacts, such as the cone-beam artifact.