MRI Reconstruction via Neural Network and Compressed Sensing
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Magnetic resonance imaging (MRI) systems using undersampling techniques face challenges in reconstructing high-quality images due to artifacts and noise, with compressed sensing methods often resulting in loss of detailed information.
Innovation Solution
A system and method that utilize a target neural network model to generate estimated images from k-space data, exceeding the original sampling rate, and combine these with a compressed sensing model to reconstruct images, incorporating a consistency term and regularization terms to improve image quality through iterative updates.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Productivity
If undersampling technique is used to reduce scan time, then productivity is improved, but measurement precision deteriorates due to artifacts and noise in reconstructed images
Solution Approach 1:
A neural network model is introduced as an intermediary between the undersampled k-space data and the final reconstructed image. The neural network processes the undersampled data to generate a preliminary reconstructed image, which then serves as input for the compressed sensing model. This two-stage intermediary approach allows the system to maintain high productivity while improving measurement precision by systematically addressing artifacts and noise through sequential processing stages.
2Productivity
If compressed sensing technique is used to reconstruct images from undersampled data, then productivity is improved, but manufacturing precision deteriorates due to loss of detailed information
Solution Approach 1:
The patent merges two different reconstruction techniques into a unified hybrid approach. The neural network model and compressed sensing model are combined in sequence, where the neural network generates an initial reconstruction and the compressed sensing model refines it. This merging of methods allows the system to achieve both fast reconstruction (productivity) and high detail quality (manufacturing precision) by leveraging the complementary strengths of each technique.
Solution Approach 2:
The neural network model performs preliminary reconstruction action on the undersampled k-space data before the compressed sensing model is applied. This preliminary action generates a reconstructed image that serves as the input for the subsequent compressed sensing refinement stage. By performing this preliminary reconstruction first, the system prepares the data in a form that enables the compressed sensing model to effectively recover detailed information while maintaining fast overall reconstruction speed.
3Measurement precision
If higher sampling rate is used to improve image quality, then measurement precision is improved, but loss of time increases due to longer scan duration
Solution Approach 1:
The neural network model is trained on fully sampled (high-quality) image data and learns to copy or replicate the detailed information present in those high-quality images from the undersampled input. During actual reconstruction, the neural network uses this learned knowledge to generate a preliminary image that mimics the quality of fully sampled images, thereby achieving high measurement precision without requiring the time-consuming high sampling rate acquisition.
Data Source
AI summary
The disclosure relates to systems and methods for magnetic resonance imaging (MRI). A method may include obtaining k-space data associated with MR signals acquired by an MR scanner. The k-space data may corresponding to a first sampling rate. The method may also include generating one or more estimated images based on the k-space data and a target neural network model. The one or more estimated images may correspond to a second sampling rate that exceeds the first sampling rate. The method may further include determining one or more target images based on the one or more estimated images and the k-space data using a compressed sensing model. The compressed sensing model may be constructed based on the one or more estimated images.