Compressed sensing magnetic resonance reconstruction method of AttISTANet based on attention mechanism

A compressed sensing and attention technology, applied in neural learning methods, 2D image generation, image data processing, etc., can solve problems such as lack of attention, indistinguishable learning features, etc., and achieve good generalization ability, PSNR value and SSIM The effect of increasing the value

Pending Publication Date: 2022-01-14
HARBIN UNIV OF SCI & TECH
View PDF0 Cites 0 Cited by
  • Summary
  • Abstract
  • Description
  • Claims
  • Application Information

AI Technical Summary

Problems solved by technology

However, ISTANet + The model lacks attention to the frequency and feature information contained in different regions and channels, and learns features indiscriminately

Method used

the structure of the environmentally friendly knitted fabric provided by the present invention; figure 2 Flow chart of the yarn wrapping machine for environmentally friendly knitted fabrics and storage devices; image 3 Is the parameter map of the yarn covering machine
View more

Image

Smart Image Click on the blue labels to locate them in the text.
Viewing Examples
Smart Image
  • Compressed sensing magnetic resonance reconstruction method of AttISTANet based on attention mechanism
  • Compressed sensing magnetic resonance reconstruction method of AttISTANet based on attention mechanism
  • Compressed sensing magnetic resonance reconstruction method of AttISTANet based on attention mechanism

Examples

Experimental program
Comparison scheme
Effect test

specific Embodiment approach 1

[0052] A kind of AttISTANet compressed sensing magnetic resonance reconstruction method based on attention mechanism described in the present embodiment is characterized in that, the method comprises the steps:

[0053] Step 1: ISTANet + Construction of network structure;

[0054] Added H(·) and L(·) operators on the basis of ISTANet, where H k ( ) and L k ( ) is a 1×1 convolution kernel, which are linear operators for increasing the number of channels and reducing the number of channels, respectively, and are respectively linear operators for increasing the number of channels and reducing the number of channels, and use the residual structure to reduce the difficulty of network training ;

[0055] Step 2: the MRI reconstruction model construction based on the attention module;

[0056] on ISTANet + The attention module is introduced into the network, and in each reconstructed image x k The improved channel attention module and spatial attention module were added before,...

specific Embodiment approach 2

[0061] This embodiment is a further description of the AttISTANet compressed sensing magnetic resonance reconstruction method based on the attention mechanism described in the first embodiment. The specific process of the first step is: on the basis of ISTANet, H(·) and L(·) operator, where H k ( ) and L k ( ) is a 1×1 convolution kernel, which are linear operators for increasing the number of channels and reducing the number of channels, respectively, and are respectively linear operators for increasing the number of channels and reducing the number of channels, and use the residual structure to reduce the difficulty of network training , ISTANet + The iteration of the kth stage in the method is as follows:

[0062] r k =x k-1 -ρ k Φ T (Φx k-1 -y) (1)

[0063] x k = L k (F k ) T (soft(F k (H k (r k )), θ k ))+r k (2)

[0064] In reconstructing MRI images, the K-space data features used are scattered in different areas, ADMMNet, ISTANet + The equal processi...

specific Embodiment approach 3

[0065] This embodiment is a further description of the AttISTANet compressed sensing magnetic resonance reconstruction method based on the attention mechanism described in the first embodiment, and the specific process of the second step is: in the ISTANet + The network introduces different attention modules, and in each reconstructed image x k reconstruction and intermediate variable r k Added the improved channel attention module and spatial attention module before, and used the attention module to recalibrate the channel and spatial features of the original K-space data features. After passing the channel attention module and spatial attention module, the expression is:

[0066] C k = L k (F k ) T (soft(F k (H k (r k )), θ k )) (3)

[0067] a k =sigmoid(relu(w(C k ))) (4)

[0068] the s k =sigmoid(f(a k )) (5)

[0069] x k =C k ×s k + r k (6)

the structure of the environmentally friendly knitted fabric provided by the present invention; figure 2 Flow chart of the yarn wrapping machine for environmentally friendly knitted fabrics and storage devices; image 3 Is the parameter map of the yarn covering machine
Login to view more

PUM

No PUM Login to view more

Abstract

The invention discloses a compressed sensing magnetic resonance reconstruction method of AttISTANet based on an attention mechanism. The method aims to solve the problem that an ISTANet + model lacks attention to frequency and feature information contained in different regions and channels and cannot learn features differentially. The method comprises the following steps: step 1, constructing an ISTANet + network structure; constructing an AttISTANet model; step 2, introducing an attention module into an ISTANet + network, adding AttISTANet of a channel attention module and a space attention module before each reconstructed image xk, and carrying out channel and space feature recalibration on features of original K space data; step 3, designing a loss function, and replacing symmetric constraint loss in ISTANet + with a smooth average absolute loss function according to the idea of a greedy algorithm; and step 4, testing the model. The method is used for compressed sensing MRI reconstruction based on the attention mechanism.

Description

technical field [0001] The invention relates to magnetic resonance image reconstruction, in particular to an AttISTANet compressive sensing magnetic resonance reconstruction method based on an attention mechanism. Background technique [0002] Traditional CS-MRI reconstruction methods mostly use structural sparsity as the image prior and solve the sparse regularization problem iteratively. These methods are based on the intrinsic properties of images and existing models of image formation. However, such methods usually not only need to manually set optimization parameters, but also need to perform complex iterative calculations, which makes image reconstruction time longer and difficult to meet the needs of medical diagnosis. [0003] In recent years, deep neural networks have gradually emerged. Using the powerful feature extraction and generalization capabilities of deep neural networks, they can not only handle the optimal transformation in CS, but also avoid manual setti...

Claims

the structure of the environmentally friendly knitted fabric provided by the present invention; figure 2 Flow chart of the yarn wrapping machine for environmentally friendly knitted fabrics and storage devices; image 3 Is the parameter map of the yarn covering machine
Login to view more

Application Information

Patent Timeline
no application Login to view more
Patent Type & Authority Applications(China)
IPC IPC(8): G06T11/00G06N3/04G06N3/08
CPCG06T11/008G06N3/08G06N3/045
Inventor 宋立新闫忠英
Owner HARBIN UNIV OF SCI & TECH
Who we serve
  • R&D Engineer
  • R&D Manager
  • IP Professional
Why Eureka
  • Industry Leading Data Capabilities
  • Powerful AI technology
  • Patent DNA Extraction
Social media
Try Eureka
PatSnap group products

Browse by: Latest US Patents, China's latest patents, Technical Efficacy Thesaurus, Application Domain, Technology Topic.

© 2024 PatSnap. All rights reserved.Legal|Privacy policy|Modern Slavery Act Transparency Statement|Sitemap