MRI Image Reconstruction for Correlated Noise and Aliasing
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Challenges exist in accurately reconstructing magnetic resonance imaging (MRI) data to reduce noise and artifacts, particularly in low-field MRI systems where noise distribution is non-independent and identically distributed, making it difficult to train effective denoising and dealiasing models without clean reference images.
Innovation Solution
A two-stage training process for a denoising and dealiasing machine-learning model is employed, using a supervised approach with a first training step on high-field MRI data and a second data augmentation step to enhance the model's robustness, allowing it to effectively remove correlated noise in low-field MRI systems without requiring clean reference images.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Object-affected harmful factors
If traditional denoising methods are used on low-field MRI data, then noise reduction is achieved, but image quality and anatomical structure observability deteriorate due to non-independent and identically distributed noise characteristics
Solution Approach 1:
The patent transforms the training approach by changing the parameter distribution of training data to match the non-independent and non-identically distributed noise characteristics of low-field MRI. This involves using realistic noise models and data augmentation techniques that preserve the statistical properties of low-field noise, enabling the neural network to learn effective denoising patterns without sacrificing image quality.
Solution Approach 2:
The patent applies preliminary data processing and augmentation steps to synthetic MRI data before training the neural network. This includes adding realistic noise patterns, applying coil sensitivity maps, and simulating the specific artifacts of low-field systems in advance, so the network learns to handle these conditions during training rather than encountering them for the first time during inference.
2Reliability
If more training data is used to improve model robustness, then denoising performance improves, but training time and computational resources increase
Solution Approach 1:
The patent creates multiple copies and variations of synthetic training data through data augmentation techniques. Instead of collecting large amounts of real low-field MRI data with known ground truth, the system generates numerous synthetic copies with varied noise patterns, coil configurations, and anatomical structures, achieving robust training with limited computational resources and time.
Solution Approach 2:
The patent employs periodic data augmentation and mixed-precision training strategies that cycle through different training configurations. This includes alternating between full-precision and reduced-precision training, applying different augmentation transformations in sequences, and using learning rate schedules that periodically adjust training intensity, thereby improving model robustness while managing training time.
Data Source
AI summary
Systems and methods for training a machine-learning model to generate denoised and dealiased image data are provided. The present disclosure provides techniques for training a machine-learning (ML) model to generate denoised and dealiased imaging data. A method includes (1) training a first ML model using a first training dataset comprising first image data to obtain a second ML model; and (2) training (a) the second ML model or (b) a third ML model using a second training dataset to obtain a fourth ML model. The second training dataset includes (i) the first image data and (ii) training image data obtained by applying at least one of the second ML model or the third ML model to second image data. The denoising and dealiasing ML model may be either the fourth ML model or derived from the fourth ML model.