Deep Convolutional Network for MRI Reconstruction
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current magnetic resonance imaging (MRI) reconstruction methods, such as GRAPPA and SPIRiT, are slow and produce images with significant noise, leading to unsatisfactory visual quality when using triple one-dimensional uniform undersampling patterns.
Innovation Solution
A one-dimensional partial Fourier parallel MRI method based on a deep convolutional neural network (DCNN) is developed, which creates a sample set from undersampled MRI images, trains a deep convolutional network model using gradient descent, and applies this model to reconstruct full-sampled images, enhancing image processing speed and reducing noise.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If GRAPPA or SPIRiT algorithms are used for MRI reconstruction, then coil sensitivity information can be utilized for parallel imaging, but the reconstruction speed is too slow and the reconstructed images contain large amounts of noise
Solution Approach 1:
The patent replaces traditional iterative mathematical algorithms (GRAPPA/SPIRiT) with a deep convolutional neural network model. The DCNN learns the mapping from undersampled to fully sampled k-space data through training, substituting the mechanical iterative computation process with a trained neural network inference process that achieves both high speed and high quality reconstruction simultaneously
Solution Approach 2:
The patent performs preliminary training of the DCNN model using paired datasets of undersampled and fully sampled MRI images before actual reconstruction. This preliminary action allows the network to learn optimal reconstruction patterns in advance, enabling fast and high-quality reconstruction during actual clinical use without requiring slow iterative computations at that stage
2Productivity
If traditional parallel imaging reconstruction algorithms are used, then undersampling can be performed to accelerate scanning, but the reconstructed images have unsatisfactory visual effect with significant noise
Solution Approach 1:
The patent replaces traditional noise-reduction filtering methods with a deep convolutional neural network that learns to reconstruct high-quality images from undersampled data. The DCNN's convolutional layers automatically learn to suppress noise while preserving image features, achieving both acceleration and noise reduction simultaneously through the trained network architecture
Solution Approach 2:
The patent changes the reconstruction approach from direct mathematical inversion to learned parameter mapping. The DCNN learns optimal parameters for reconstructing fully sampled k-space data from undersampled inputs, transforming the reconstruction process into a parameter optimization problem that can be solved efficiently while maintaining image quality and reducing noise
Data Source
AI summary
The present disclosure relates to a 1D partial Fourier parallel magnetic resonance imaging method with a deep convolutional network and belongs to the technical field of magnetic resonance imaging. The method includes steps of: creating a sample set and a sample label set for training; constructing an initial deep convolutional network model; inputting a training sample of the sample set to the initial deep convolutional network model for forward process, comparing an output result of the forward process with an expected result in the sample label set, and performing training with a gradient descent method until a parameter of each layer which enables consistency between the output result and the expected result to be maximum is obtained; creating an optimal deep convolutional network model by using the obtained parameter of the each layer; and inputting a multi-coil undersampled image sampled online to the optimal deep convolutional network model, performing the forward process on the optimal deep convolutional network model, and outputting a reconstructed single-channel full-sampled image. The present disclosure can well remove the noise of the reconstructed image, reconstruct a magnetic resonance image with a better visual effect, and has high practical value.