ML-Based Iterative Image Reconstruction Noise Reduction
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Iterative reconstruction algorithms in tomographic imaging introduce statistical noise and require numerous iterations to converge, consuming time and computational resources, while existing regularization techniques often lead to artificial noise textures and reduced quantitative accuracy, especially for small structures.
Innovation Solution
A system utilizing a trained machine learning module that reduces noise in correction data rather than intermediate images, leveraging the relationship between noise levels learned from historical patient records to improve convergence speed and reduce noise, employing larger receptive fields than previous methods to effectively separate noise from real features.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If iterative reconstruction algorithms are used to obtain cross-sectional imagery, then image quality and diagnostic detail are improved, but statistical noise is introduced and computational time is increased
Solution Approach 1:
The machine learning model is trained in advance on a large dataset of projection data and corresponding ground truth images. This preliminary training phase enables the model to learn complex reconstruction patterns and noise characteristics, allowing it to perform rapid inference during actual reconstruction without requiring multiple iterative computations.
Solution Approach 2:
The patent replaces the traditional iterative mechanical computation process with a trained machine learning model that performs reconstruction in a single pass. The ML model substitutes the iterative algorithmic approach, transforming the reconstruction from a multi-step computational process into a direct prediction task that significantly reduces computation time.
2Reliability
If more iterations are performed in iterative reconstruction, then convergence to final result is improved, but statistical noise propagates and computational resources are consumed
Solution Approach 1:
The machine learning model replaces the iterative reconstruction process that generates noise through repeated computations. By using a pre-trained neural network, the system achieves convergence in a single inference step without the noise propagation that occurs during multiple iterative updates of traditional algorithms.
3Object-generated harmful factors
If regularization techniques are applied to reduce noise in iterative reconstruction, then noise corruption is curbed, but artificial noise textures and reduced quantitative accuracy occur
Solution Approach 1:
The patent replaces traditional regularization techniques with a machine learning-based approach. The ML model, trained on diverse data, learns to distinguish between actual image features and noise patterns, enabling effective noise reduction without introducing the artificial textures and quantitative inaccuracies that characterize conventional regularization methods.
Data Source
AI summary
A system (CDD) and related method for facilitating an iterative reconstruction operation. In iterative reconstruction, imagery in image domain is reconstructed in plural steps from measured projection data in projection domain. The system and methods use a trained machine learning module (MLM). The system receives input correction data generated in the iterative reconstruction operation. The system predicts, based on the input correction data, output correction data. The output correction data is provided for facilitating correcting a current image, as reconstructed in a given step, into a new image.