Looking for breakthrough ideas for innovation challenges? Try Patsnap Eureka!

3D medical image super-resolution reconstruction method based based on dense convolution neural network

A super-resolution reconstruction, convolutional neural network technology, applied in the field of image processing, can solve a large number of parameters and other problems, achieve the effect of less weight, robustness and small model

Pending Publication Date: 2019-02-19
TIANJIN UNIV
View PDF3 Cites 20 Cited by
  • Summary
  • Abstract
  • Description
  • Claims
  • Application Information

AI Technical Summary

Problems solved by technology

Secondly, the fast super-resolution convolutional neural network is to directly stack multiple neural networks. Direct conversion to 3D may result in a large number of parameters, and it will also face challenges in memory allocation.

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
  • 3D medical image super-resolution reconstruction method based based on dense convolution neural network
  • 3D medical image super-resolution reconstruction method based based on dense convolution neural network

Examples

Experimental program
Comparison scheme
Effect test

Embodiment 1

[0034] A 3D medical image super-resolution reconstruction method based on dense convolutional neural network, see figure 1 , the method includes the following steps:

[0035] 101: Randomly divide the public data set of the Human Connectome Project into four parts 7:1:1:1, the first part is used as the training set, the second part is used to verify and optimize the network weights, and the third part is used to evaluate the selected hyperparameters , the fourth part is used as the test set;

[0036] 102: Divide the low-resolution three-dimensional image 320*320*256 into small blocks of 64*64*64, and use it as network input based on dense connection structure;

[0037] Wherein, the network in step 102 includes: a normalization layer, a nonlinear mapping part composed of ELU, and a convolutional layer with a growth rate k of 24.

[0038] 103: Calculate the error between the reconstructed image and the actual image, and propagate the error layer by layer from the output layer t...

Embodiment 2

[0045] The scheme in embodiment 1 is further introduced below in conjunction with specific examples, see the following description for details:

[0046] 201: Data preparation;

[0047] The first step: divide the data set, the data source is the data set released by the Human Connectome Project (Human Connectome Project).

[0048] The entire data set is randomly divided into four parts 7:1:1:1, the first part is used as the training set, the second part is used to verify and optimize the network weights, the third part is used to evaluate the selected hyperparameters, and the fourth part is used as the test set . Among them, the data sets for training, verification and testing are all composed of high-resolution images and corresponding low-resolution images. The low-resolution images are converted from high-resolution images to frequency domain by Fourier transform, and the Values ​​in other directions other than the vector direction are set to zero to obtain a low-resolutio...

Embodiment 3

[0060] Below in conjunction with concrete data, the scheme in embodiment 1 and 2 is further introduced, see the following description for details:

[0061] 301: data preparation;

[0062] (a) Divide the data set, the data source is the HCP public data set, which contains the brain T1w structural images of 1113 patients, the image size is 320*320*256, and the spatial resolution is 0.7mm.

[0063] (b) The entire data set is randomly divided into four parts 7:1:1:1, 780 images are used as training sets, 111 images are used for verification and optimization of network weights, 111 images are used to evaluate selected hyperparameters, 111 image as a test set.

[0064] 302: Network construction;

[0065] The network structure of the embodiment of the present invention mainly includes: two convolutional layers and a densely connected structure, which will be combined with the following figure 1 , to describe in detail the network structure built by the embodiment of the present in...

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 3D medical image super-resolution reconstruction method based on a dense convolution neural network, which includes: randomly dividing the data set disclosed by human connection group project into four parts, wherein the first part is used as training set, the second part is used for verifying and optimizing network weight, the third part is used for evaluating selected super-parameters, and the fourth part is used as test set; dividing a low-resolution three-dimensional image 320*320*256 into small blocks of 64*64*64 as network input based on a dense connection structure; The error between the reconstructed image and the actual image is calculated, and the error is propagated backward from the output layer to the hidden layer until it is propagated to the input layer. After continuous feedback optimization until the error is no longer reduced, the optimal model of super-resolution reconstruction is obtained, and the optimal model is trained. The medical three-dimensional low-resolution image is input, the trained model is loaded, and the reconstructed high-resolution image is output.

Description

technical field [0001] The invention relates to the field of image processing, in particular to a method for super-resolution reconstruction of three-dimensional medical images based on a dense convolutional neural network. Background technique [0002] Due to its extensive practical value and theoretical value, image super-resolution reconstruction technology has become a research hotspot in the field of computer vision and image processing. Image super-resolution refers to using one or more low-resolution images and using related algorithms to obtain A crisp, high-resolution image. High-resolution medical images can have rich structural information, supporting image analysis and quantitative measurements. [0003] High-resolution images of medical images are rich in structural details, enabling precise image analysis and quantitative measurements. However, the high-resolution image generation of MRI (Nuclear Magnetic Resonance Imaging, nuclear magnetic resonance imaging)...

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): G06T3/40G06T17/00
CPCG06T3/4053G06T17/00G06T2210/41
Inventor 吕卫张国帅褚晶辉
Owner TIANJIN UNIV
Who we serve
  • R&D Engineer
  • R&D Manager
  • IP Professional
Why Patsnap Eureka
  • Industry Leading Data Capabilities
  • Powerful AI technology
  • Patent DNA Extraction
Social media
Patsnap Eureka Blog
Learn More
PatSnap group products

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

© 2024 PatSnap. All rights reserved.Legal|Privacy policy|Modern Slavery Act Transparency Statement|Sitemap|About US| Contact US: help@patsnap.com