Deep Convolutional Neural Network for CBCT Artifact Reduction
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current methods for reducing artifacts in cone beam computed tomography (CBCT) images are inefficient and time-consuming, failing to effectively suppress all types of noise and artifacts, which hinders adaptive replanning and diagnosis in radiation therapy and other imaging applications.
Innovation Solution
A deep convolutional neural network (DCNN) is trained using a collection of view-adjacent artifact-contaminated and artifact-reduced CBCT projection space images to learn correlations and reduce artifacts in real-time, specifically targeting scattering, noise, and other artifacts such as beam hardening and ring artifacts.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Productivity
If traditional artifact reduction methods are used, then processing time is reduced, but image quality and artifact suppression effectiveness deteriorate
Solution Approach 1:
The patent replaces traditional mechanical/mathematical artifact reduction algorithms with a deep learning-based neural network system. The DCNN learns artifact patterns from training data and automatically suppresses them during inference, achieving both high processing speed and superior image quality without the trade-off present in traditional methods.
Solution Approach 2:
The patent transforms the artifact reduction problem from a deterministic parameter-based mathematical operation into a data-driven parameter optimization problem. By training the neural network on diverse CBCT images with varying artifacts, the system learns optimal parameter adjustments dynamically, achieving high image quality while maintaining fast processing during clinical use.
2Manufacturing precision
If multiple different artifact reduction methods are employed, then artifact suppression effectiveness improves, but system complexity and processing time increase
Solution Approach 1:
The patent creates a universal artifact reduction system where a single DCNN model handles multiple types of CBCT artifacts (scattering, beam hardening, ring artifacts, noise) simultaneously. The neural network learns to identify and suppress different artifact types through unified training, eliminating the need for multiple separate processing methods and reducing system complexity.
Solution Approach 2:
The patent combines multiple artifact reduction capabilities into a single integrated neural network framework. Instead of applying separate methods for different artifacts in sequence, the DCNN merges all artifact suppression functions into one model that processes images in a single pass, reducing both computational complexity and processing time.
3Manufacturing precision
If multiple different artifact reduction methods are employed, then artifact suppression effectiveness improves, but processing time and computational resources increase
Solution Approach 1:
The patent performs preliminary action by training the DCNN model offline on large datasets containing various artifact types. This pre-learning phase captures artifact patterns and reduction strategies, so that during clinical processing, the network can rapidly apply learned knowledge without time-consuming iterative calculations, achieving high image quality with minimal processing time.
Solution Approach 2:
The patent substitutes time-consuming iterative mathematical optimization algorithms with a trained neural network that performs artifact reduction in a single forward pass. This replacement of mechanical computation with learned pattern recognition dramatically reduces processing time while maintaining or improving artifact suppression effectiveness.
Data Source
Figure 1
Figure 2
Figure 3
AI summary
Systems and methods can include training a DCNN to reduce one or more artifacts using a projection space approach or an image space approach. A projection space approach can include collecting an artifact contaminated CBCT projection space image, and a corresponding artifact reduced, CBCT projection space image from each patient in a group of patients, and using the artifact contaminated CBCT projection space image and the corresponding artifact reduced, CBCT projection space image collected from each patient in the group of patients to train a DCNN to reduce one or more artifacts in a projection space image. An image space approach can include collecting a plurality of CBCT patient anatomical images and corresponding registered CT anatomical images from a group of patients, and using the plurality of CBCT anatomical images and corresponding artifact reduced CT anatomical images to train a DCNN to remove artifacts from a CBCT anatomical image.