Convolutional neural network-based low-dosage CT image noise inhibition method

A convolutional neural network and CT image technology, applied in machine learning and pattern recognition, low-dose CT image noise suppression based on convolutional neural network, to avoid cumbersome processes and preserve image details

Pending Publication Date: 2018-09-21
SOUTHERN MEDICAL UNIVERSITY
View PDF4 Cites 23 Cited by
  • Summary
  • Abstract
  • Description
  • Claims
  • Application Information

AI Technical Summary

Problems solved by technology

[0007] The present invention overcomes the shortcomings in the prior art, and provides a low-dose CT image noise suppression method based on a convolutional neural network. The network (Convolutional Neural Network, ConvNet) learns CT image data, automatically learns image features...

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
  • Convolutional neural network-based low-dosage CT image noise inhibition method
  • Convolutional neural network-based low-dosage CT image noise inhibition method
  • Convolutional neural network-based low-dosage CT image noise inhibition method

Examples

Experimental program
Comparison scheme
Effect test

Embodiment 1

[0074] In the training process of the convolutional neural network, different purposes can be achieved by changing the preprocessing process of the data input to the network: (1) by training the convolutional neural network to predict the noise image corresponding to the low-dose CT image, after processing A denoised image can be obtained. (2) Also according to the purpose and principle of denoising low-dose CT images, the same neural network can be used to directly predict high-dose CT images corresponding to low-dose CT images through different training methods. Low-dose CT denoising is basically the same, but the detail process is slightly different. The following will describe the process of realizing the two goals.

[0075] Objective 1: Input low-dose CT images to suppress artifacts and noise in low-dose CT images, and suppress or remove noise and artifacts in low-dose CT images through the predictive ability of convolutional neural networks

[0076] The present inventi...

Embodiment 2

[0118] In the above experiments, the convolutional neural network used is a plain network formed by stacking volume-based layers. In the specific implementation, the application method of the present invention can be extended by introducing connection shortcuts between the volume-based layers in the plain network. To further improve the denoising effect, the method of introducing connection shortcuts in this experiment is relatively flexible. Only two examples are given, but the network architecture involving such connection shortcuts belongs to the protection scope of this patent.

[0119] 1. Residual network connection (multi-scale cascade) method:

[0120] In the "flat" network (and the network introduced above), when the number of network layers increases to a certain extent, the network's

[0121] The output of the layer-number convolutional neural network not only did not improve, but showed significant degradation. In order for the network to reach a deeper level witho...

Embodiment 3

[0133] The above experiments dealt with two-dimensional CT images, which can be extended to the processing of three-dimensional low-dose CT images.

[0134] It is also feasible to use convolutional neural networks to denoise 3D images or predict iteratively reconstructed high-dose CT images at the 3D level. The principle is the same as operating on a two-dimensional plane. It uses the 3D convolution kernel to extract the features of the input 3D CT image data. These convolution kernels can be operated on different levels, and more spatial details of the image are retained. A 3D convolution is constructed based on the 3D convolution feature kernel. product neural network. Compared with the two-dimensional convolutional neural network, this network architecture can generate multi-channel information from successive layers, and finally combine the information of all channels to form the final feature description.

[0135] The basic process of generating high-dose iteratively re...

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 relates to a convolutional neural network-based low-dosage CT image noise inhibition method. The convolutional neural network-based low-dosage CT image noise inhibition method comprisesthe following steps: (1) performing normalization processing on the input original low-dosage CT image L by utilizing the low-dosage CT image obtained through low tube current tube voltage scanning, evaluating the mean value and the standard deviation of the gray level of all the pixels of the low-dosage CT image, and subtracting the mean value from the L and dividing the standard deviation to obtain a CT image L0; (2) taking the acquired preprocessed low-dosage CT image L0 as input of the convolutional neural network and predicting a noise CT image D0 corresponding to a low-dosage CT image I;and (3) subtracting the predicted noise image D0 from the L0, multiplying the standard deviation of the low-dosage CT image and adding the mean value of the low-dosage CT image to acquire the denoised image H0. The low-dosage CT image is subjected to denoising processing by the convolutional neural network, so that the image is guaranteed to meet the diagnosis quality, the irradiation dosage of asubject is reduced, the detection rate of the focus is increased and the disease is diagnosed early.

Description

technical field [0001] The invention belongs to the field of digital image processing, and also belongs to the category of machine learning and pattern recognition. Specifically, it relates to a low-dose CT image noise suppression method based on a convolutional neural network. Background technique [0002] With the continuous development of CT technology and the popularization of CT equipment, CT examination has been more and more widely used in the diagnosis of various diseases. In recent years, the emergence of multi-slice spiral CT has improved the time resolution, spatial resolution and density resolution of CT images, and the development of image post-processing software has greatly improved the detection rate of lesions and enabled early diagnosis of diseases. However, with the upgrading of multi-slice spiral CT, the volume acquisition greatly increases the amount of data and information, which not only provides more high-quality image data for diagnosticians, but als...

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
IPC IPC(8): G06T5/00
CPCG06T5/002G06T2207/10081G06T2207/20081G06T2207/20084
Inventor 阳维陈阳张慧娟蔡广威冯前进
Owner SOUTHERN MEDICAL UNIVERSITY
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