WO2020118826A1 - 一种左心室图像分割方法、装置、设备及存储介质 - Google Patents

一种左心室图像分割方法、装置、设备及存储介质 Download PDF

Info

Publication number
WO2020118826A1
WO2020118826A1 PCT/CN2019/071075 CN2019071075W WO2020118826A1 WO 2020118826 A1 WO2020118826 A1 WO 2020118826A1 CN 2019071075 W CN2019071075 W CN 2019071075W WO 2020118826 A1 WO2020118826 A1 WO 2020118826A1
Authority
WO
WIPO (PCT)
Prior art keywords
image
layer
network
feature
sampling
Prior art date
Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
Ceased
Application number
PCT/CN2019/071075
Other languages
English (en)
French (fr)
Inventor
胡战利
梁栋
贺阳素
杨永峰
刘新
郑海荣
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
Shenzhen Institute of Advanced Technology of CAS
Original Assignee
Shenzhen Institute of Advanced Technology of CAS
Priority date (The priority date is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the date listed.)
Filing date
Publication date
Application filed by Shenzhen Institute of Advanced Technology of CAS filed Critical Shenzhen Institute of Advanced Technology of CAS
Publication of WO2020118826A1 publication Critical patent/WO2020118826A1/zh
Anticipated expiration legal-status Critical
Ceased legal-status Critical Current

Links

Images

Classifications

    • G—PHYSICS
    • G06—COMPUTING OR CALCULATING; COUNTING
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T7/00—Image analysis
    • G06T7/10—Segmentation; Edge detection
    • G—PHYSICS
    • G06—COMPUTING OR CALCULATING; COUNTING
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T2207/00—Indexing scheme for image analysis or image enhancement
    • G06T2207/20—Special algorithmic details
    • G06T2207/20081—Training; Learning
    • G—PHYSICS
    • G06—COMPUTING OR CALCULATING; COUNTING
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T2207/00—Indexing scheme for image analysis or image enhancement
    • G06T2207/20—Special algorithmic details
    • G06T2207/20084—Artificial neural networks [ANN]
    • G—PHYSICS
    • G06—COMPUTING OR CALCULATING; COUNTING
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T2207/00—Indexing scheme for image analysis or image enhancement
    • G06T2207/30—Subject of image; Context of image processing
    • G06T2207/30004—Biomedical image processing
    • G06T2207/30048—Heart; Cardiac

Definitions

  • the invention belongs to the technical field of medical image processing, and particularly relates to a left ventricular image segmentation method, device, equipment and storage medium.
  • the purpose of the present invention is to provide a method, device, equipment and storage medium for left ventricular image segmentation, aiming to solve the problem that the prior art cannot provide an effective left ventricular image segmentation method, resulting in low efficiency and effect of left ventricular image segmentation Bad question.
  • the present invention provides a left ventricular image segmentation method.
  • the method includes the following steps:
  • the image segmentation network is a deep learning network
  • the image segmentation network includes a downsampling part and an upsampling part
  • the downsampling part includes a first A convolution layer and a down-sampling network layer
  • the up-sampling part includes a second convolution layer and an up-sampling network layer;
  • the left ventricular segmented image processed by the image segmentation network is obtained and output.
  • the present invention provides a left ventricular image segmentation device, the device comprising:
  • An image-to-be-divided image receiving unit used to receive the left ventricle image to be divided
  • An image segmentation unit for inputting the left ventricle image into a pre-trained image segmentation network for segmentation the image segmentation network is a deep learning network, and the image segmentation network includes a downsampling part and an upsampling part, the The down-sampling section includes a first convolution layer and a down-sampling network layer, and the up-sampling section includes a second convolution layer and an up-sampling network layer;
  • the divided image output unit is used to obtain and output the left ventricular divided image obtained by the image division network processing.
  • the present invention also provides a computer device, including a memory, a processor, and a computer program stored in the memory and executable on the processor, which is implemented when the processor executes the computer program
  • a computer device including a memory, a processor, and a computer program stored in the memory and executable on the processor, which is implemented when the processor executes the computer program
  • the present invention also provides a computer-readable storage medium that stores a computer program, and when the computer program is executed by a processor, the computer program can be implemented as described in the above-mentioned left ventricular image segmentation method A step of.
  • the invention receives the left ventricle image to be segmented, and the trained image segmentation network divides the left ventricular image to obtain the segmented left ventricular segmented image and outputs it.
  • the image segmentation network is a deep learning network, including an upsampling part And the down-sampling part, the up-sampling part includes the first convolution layer and the down-sampling network layer, the up-sampling part includes the second convolution layer and the up-sampling network layer, so that the left ventricular image is automatically segmented by the image segmentation network of the above structure , Effectively improve the efficiency and effect of left ventricular image segmentation.
  • FIG. 1 is an implementation flowchart of a left ventricular image segmentation method according to Embodiment 1 of the present invention
  • FIG. 2 is a flowchart of an implementation of a left ventricular image segmentation method provided by Embodiment 2 of the present invention
  • FIG. 3 is a schematic structural diagram of a left ventricular image segmentation device provided in Embodiment 3 of the present invention.
  • FIG. 4 is a schematic diagram of a preferred structure of a left ventricular image segmentation device provided in Embodiment 3 of the present invention.
  • FIG. 5 is a schematic structural diagram of a computer device according to Embodiment 4 of the present invention.
  • FIG. 1 shows an implementation process of a method for segmenting a left ventricle image provided by Embodiment 1 of the present invention. For ease of description, only the parts related to the embodiment of the present invention are shown. The details are as follows:
  • step S101 receive the left ventricle image to be segmented
  • the embodiments of the present invention are applicable to devices such as computers and servers that support image processing.
  • the left ventricular image to be segmented may be a left ventricular image obtained from a medical database published on the Internet, or a left ventricular image provided by a hospital or doctor, or a left ventricular image scanned by a medical scanning device such as CT or MRI.
  • the embodiment of the present invention does not involve the scanning process of the left ventricle image, and will not be described here.
  • the left ventricle image is input into a pre-trained image segmentation network for segmentation.
  • the image segmentation network is a deep learning network, including a down-sampling part and an up-sampling part, and the down-sampling part includes a first convolutional layer and down-sampling For the network layer, the upsampling part includes a second convolution layer and an upsampling network layer.
  • the image segmentation network is pre-trained, where the image segmentation network is a deep network, including a downsampling part and an upsampling part, and the downsampling part includes a first convolution layer and a downsampling network layer, and the upsampling part Including the second convolutional layer and the upsampling network layer.
  • the left ventricular image After obtaining the left ventricular image, input the left ventricular image to the first convolutional layer to extract the image features of the left ventricular image, and then input the image features to the down-sampling network layer, which will be processed by each layer of the down-sampling network layer in turn
  • the image features of the image are input to the upsampling network layer for processing to obtain the image features processed by each layer of the downsampling network layer and each layer of the downsampling network layer, and then the second convolution layer performs a convolution operation to realize the left ventricular image Of division.
  • each down-sampling network layer includes a first feature convolution layer and a second feature convolution layer
  • each up-sampling network layer includes a feature deconvolution layer and a first feature convolution layer, thereby passing different convolution layers
  • each image features processed by the down-sampling network layer of each layer enter the first-layer up-sampling network layer, these image features are processed through the feature deconvolution layer in the first-layer up-sampling network layer, and then the The first feature convolutional layer in an up-sampling network layer processes the image features processed by the feature deconvolution layer.
  • the downsampling part in the image segmentation network includes a first convolution layer and a three-layer downsampling network layer
  • the upsampling part in the image segmentation network includes a three-layer upsampling network layer and a three-layer second convolution Layers
  • the first and second convolution layers have different convolution kernel sizes.
  • the first feature convolution layer and the second feature convolution layer have different step sizes to improve the image segmentation network. The processing effect of the left ventricular image features.
  • the convolution kernel of the first convolutional layer has a size of 5*5
  • the convolution kernel of the second convolutional layer has a size of 1*1
  • the convolution kernel of the first feature convolutional layer has a size of 5*5 and
  • the step size is 3
  • the convolution kernel of the second feature convolutional layer is 5*5 in size and the step size is 1
  • the convolution kernel size of the feature deconvolution layer is 4*4 and the step size is 3.
  • step S103 the left ventricular segmented image obtained by the image segmentation network processing is obtained and output.
  • the left ventricular segmented image processed by the image segmentation network is a left ventricular contour image segmented from the left ventricular image to be segmented, which can help the doctor visually observe the patient's left ventricular status.
  • the convolution calculation formula is:
  • I represents the left ventricle image to be segmented
  • m and n are the width and height of the convolution kernel in the first convolution layer
  • K l represents the lth convolution kernel in the first convolution layer
  • i and j are The position of the image pixel in the left ventricular image.
  • the formula for processing the image features output by the upper layer up-sampling network through the second convolution layer is also the above formula.
  • I represents the image features output by the one-layer up-sampling network
  • m and n are the second The width and height of the convolution kernel in the convolution layer
  • K l represents the lth convolution kernel in the second convolution layer
  • i and j are the positions of image pixels in the image feature.
  • these image features are nonlinearly activated to improve the segmentation effect of the image segmentation network on the left ventricle image.
  • the formula for nonlinear activation is:
  • S l (i, j) is the image feature extracted from the first convolution layer
  • z is the number of extracted image features
  • relu() is the nonlinear activation function
  • f(x) is the nonlinear activation function.
  • the image features received by the first feature convolution layer in the down-sampling network layer, the second feature convolution layer in the down-sampling network layer, or the first feature convolution layer in the up-sampling network When processing, it includes the feature convolution operation and the non-linear change operation on the image features to improve the processing effect on the image features.
  • feature convolution operations are performed on the image features input to the first feature convolution layer, and the routing coefficients are performed by the dynamic routing algorithm Update, and then input the image features processed by the feature convolution operation into a preset nonlinear change formula, and perform a nonlinear change operation to improve the processing effect of the first feature convolution on the image features.
  • S xy is the image feature processed by the feature convolution operation
  • N is the number of these image features
  • p is the number of network layers in the image segmentation network where the current down-sampling network layer (or current up-sampling network layer) is located
  • x and y are the image features in Position in the left ventricle image.
  • V xy is the image feature processed by the nonlinear change operation.
  • routing coefficient When the routing coefficient is updated by the dynamic routing algorithm, it is updated iteratively, and the d-th iteration process is as follows:
  • the image features input to the layer network are subjected to feature convolution processing, and the processing formula is:
  • the left ventricular image is segmented by a trained image segmentation network to obtain the segmented left ventricular segmented image and output
  • the image segmentation network is a deep learning network, including an upsampling part and a downsampling Partially, the upsampling part includes the first convolutional layer and the downsampling network layer, the upsampling part includes the second convolutional layer and the upsampling network layer, and the downsampling network layer includes the first feature convolutional layer and the second feature convolutional layer
  • the up-sampling network layer includes a feature deconvolution layer and a first feature deconvolution layer, so that the left ventricular image is automatically segmented by the image segmentation network with the above structure, which effectively improves the efficiency and effect of left ventricular image segmentation.
  • FIG. 2 shows an implementation flow of a method for segmenting a left ventricle image provided by Embodiment 2 of the present invention. For ease of description, only parts related to the embodiment of the present invention are shown, and details are as follows:
  • step S201 the acquired left ventricular training image is acquired.
  • step S202 the left ventricle training image is input into the image segmentation network, and the image segmentation network is trained through a preset optimization algorithm, which is an Adam learning rate adaptive optimization algorithm.
  • a preset optimization algorithm which is an Adam learning rate adaptive optimization algorithm.
  • the network structure of the image segmentation network has been described in detail in the first embodiment.
  • the image segmentation network is constructed according to the above network results. Collect multiple left ventricular training images and use them as training samples for the image segmentation network, input them into the image segmentation network one by one, and use Adam (adaptive moment estimation) learning rate adaptive optimization algorithm to train the image segmentation network to obtain training Good image segmentation network.
  • Adam adaptive moment estimation
  • step S203 receive the left ventricle image to be segmented
  • step S204 the left ventricular image is input into a pre-trained image segmentation network for segmentation.
  • the image segmentation network includes a downsampling part and an upsampling part.
  • the downsampling part includes a first convolutional layer and a downsampling network layer.
  • Upsampling Part includes the second convolution layer and the up-sampling network layer;
  • step S205 the left ventricular segmented image obtained by the image segmentation network processing is obtained and output.
  • the image segmentation network is trained by the Adam learning rate adaptive optimization algorithm, and the trained image segmentation network segments the left ventricle image to obtain the segmented left ventricular segmented image and output it.
  • the image segmentation The network is a deep learning network, including an upsampling part and a downsampling part.
  • the upsampling part includes a first convolutional layer and a downsampling network layer.
  • the upsampling part includes a second convolutional layer and an upsampling network layer.
  • the downsampling network layer includes The first feature convolution layer and the second feature convolution layer, the up-sampling network layer includes a feature deconvolution layer and a first feature deconvolution layer, so as to automatically optimize the left ventricular image segmentation and effectively improve the left ventricular image Efficiency and effect of segmentation.
  • FIG. 3 shows a structure of a left ventricle image segmentation device provided in Embodiment 3 of the present invention. For ease of explanation, only parts related to the embodiment of the present invention are shown, including:
  • the image-to-be-divided receiving unit 31 is configured to receive the left ventricle image to be divided.
  • the image segmentation unit 32 is used to input the left ventricle image into a pre-trained image segmentation network for segmentation.
  • the image segmentation network is a deep learning network.
  • the image segmentation network includes a downsampling part and an upsampling part.
  • the downsampling part includes the first volume For the accumulation layer and the down-sampling network layer
  • the up-sampling part includes the second convolution layer and the up-sampling network layer.
  • the image segmentation network is pre-trained. After obtaining the left ventricular image, input the left ventricular image to the first convolutional layer to extract the image features of the left ventricular image, and then input the image features to the down-sampling network layer, which will be processed by each layer of the down-sampling network layer in turn The image features of the image are input to the upsampling network layer for processing to obtain the image features processed by each layer of the downsampling network layer and each layer of the downsampling network layer, and then the second convolution layer performs a convolution operation to realize the left ventricular image Of division.
  • each down-sampling network layer includes a first feature convolution layer and a second feature convolution layer
  • each up-sampling network layer includes a feature deconvolution layer and a first feature convolution layer, thereby passing different convolution layers
  • each image features processed by the down-sampling network layer of each layer enter the first-layer up-sampling network layer, these image features are processed through the feature deconvolution layer in the first-layer up-sampling network layer, and then the The first feature convolutional layer in an up-sampling network layer processes the image features processed by the feature deconvolution layer.
  • the downsampling part in the image segmentation network includes a first convolution layer and a three-layer downsampling network layer
  • the upsampling part in the image segmentation network includes a three-layer upsampling network layer and a three-layer second convolution Layers
  • the first and second convolution layers have different convolution kernel sizes.
  • the first feature convolution layer and the second feature convolution layer have different step sizes to improve the image segmentation network. The processing effect of the left ventricular image features.
  • the divided image output unit 33 is used to obtain and output a left ventricular divided image obtained by image segmentation network processing.
  • the left ventricular segmented image processed by the image segmentation network is a left ventricular contour image segmented from the left ventricular image to be segmented, which can help the doctor visually observe the patient's left ventricular status.
  • the convolution calculation formula is:
  • I represents the left ventricle image to be segmented
  • m and n are the width and height of the convolution kernel in the first convolution layer
  • K l represents the lth convolution kernel in the first convolution layer
  • i and j are The position of the image pixel in the left ventricular image.
  • the formula for processing the image features output by the upper layer up-sampling network through the second convolution layer is also the above formula.
  • I represents the image features output by the one-layer up-sampling network
  • m and n are the second The width and height of the convolution kernel in the convolution layer
  • K l represents the lth convolution kernel in the second convolution layer
  • i and j are the positions of image pixels in the image feature.
  • these image features are non-linearly activated to improve the segmentation effect of the image segmentation network on the left ventricle image.
  • the formula for nonlinear activation is:
  • S l (i, j) is the image feature extracted from the first convolution layer
  • z is the number of extracted image features
  • relu() is the nonlinear activation function
  • f(x) is the nonlinear activation function.
  • the image features received by the first feature convolution layer in the down-sampling network layer, the second feature convolution layer in the down-sampling network layer, or the first feature convolution layer in the up-sampling network When processing, it includes the feature convolution operation and the non-linear change operation on the image features to improve the processing effect on the image features.
  • feature convolution operations are performed on the image features input to the first feature convolution layer, and the routing coefficients are performed by the dynamic routing algorithm Update, and then input the image features processed by the feature convolution operation into a preset nonlinear change formula, and perform a nonlinear change operation to improve the processing effect of the first feature convolution on the image features.
  • S xy is the image feature processed by the feature convolution operation
  • N is the number of these image features
  • p is the number of network layers in the image segmentation network where the current down-sampling network layer (or current up-sampling network layer) is located
  • x and y are the image features in Position in the left ventricle image.
  • V xy is the image feature processed by the nonlinear change operation.
  • routing coefficient When the routing coefficient is updated by the dynamic routing algorithm, it is updated iteratively, and the d-th iteration process is as follows:
  • the image features input to the layer network are subjected to feature convolution processing, and the processing formula is:
  • the image segmentation unit 32 further includes:
  • the image feature extraction unit 421 is configured to perform feature extraction on the left ventricular image through the first convolution layer to generate image features corresponding to the left ventricular image;
  • the image feature processing unit 422 is used to input the image features to the down-sampling network layer connected to the first convolutional layer, and sequentially process through each layer of the down-sampling network layer, each layer of the up-sampling network layer, and the second convolutional layer to obtain Split image of left ventricle.
  • an image segmentation network is constructed according to the above network results.
  • Collect multiple left ventricular training images and use them as training samples for the image segmentation network input them into the image segmentation network one by one, and use Adam (adaptive moment estimation) learning rate adaptive optimization algorithm to train the image segmentation network to obtain training A good image segmentation network, so as to improve the automatic segmentation effect of the image segmentation network on the left ventricle image through the adaptive optimization algorithm.
  • Adam adaptive moment estimation
  • the image segmentation network is trained by the Adam learning rate adaptive optimization algorithm, and the trained image segmentation network segments the left ventricle image to obtain the segmented left ventricular segmented image and output it.
  • the image segmentation The network is a deep learning network, including an upsampling part and a downsampling part.
  • the upsampling part includes a first convolutional layer and a downsampling network layer.
  • the upsampling part includes a second convolutional layer and an upsampling network layer.
  • the downsampling network layer includes The first feature convolution layer and the second feature convolution layer, the up-sampling network layer includes a feature deconvolution layer and a first feature deconvolution layer, so as to automatically optimize the left ventricular image segmentation and effectively improve the left ventricular image Efficiency and effect of segmentation.
  • each unit of a left ventricular image segmentation device may be implemented by a corresponding hardware or software unit, and each unit may be an independent software and hardware unit, or may be integrated into one software and hardware unit, which is not used here To limit the invention.
  • FIG. 5 shows a structure of a computer device provided by Embodiment 5 of the present invention. For convenience of description, only parts related to the embodiment of the present invention are shown.
  • the automobile 5 of the embodiment of the present invention includes a processor 50, a memory 51, and a computer program 52 stored in the memory 51 and executable on the processor 50.
  • the processor 50 executes the computer program 52, the steps in the above method embodiments are implemented, for example, steps S101 to S103 shown in FIG. 1.
  • the processor 50 executes the computer program 52, the functions of the units in the foregoing device embodiments are realized, for example, the functions of the units 31 to 33 shown in FIG.
  • the left ventricular image is segmented by a trained image segmentation network to obtain the segmented left ventricular segmented image and output
  • the image segmentation network is a deep learning network, including an upsampling part and a downsampling Partially, the upsampling part includes the first convolutional layer and the downsampling network layer, the upsampling part includes the second convolutional layer and the upsampling network layer, and the downsampling network layer includes the first feature convolutional layer and the second feature convolutional layer
  • the up-sampling network layer includes a feature deconvolution layer and a first feature deconvolution layer, so that the left ventricular image is automatically segmented by the image segmentation network with the above structure, which effectively improves the efficiency and effect of left ventricular image segmentation.
  • a computer-readable storage medium stores a computer program, and when the computer program is executed by a processor, the steps in the foregoing method embodiments are implemented, for example, FIG. 1 Steps S101 to S103 shown.
  • the functions of the units in the foregoing device embodiments are realized, for example, the functions of the units 31 to 33 shown in FIG. 3.
  • the left ventricular image is segmented by the trained image segmentation network to obtain the segmented left ventricular segmented image and output.
  • the image segmentation network is a deep learning network, including an upsampling part and a downsampling Partially, the upsampling part includes the first convolutional layer and the downsampling network layer, the upsampling part includes the second convolutional layer and the upsampling network layer, and the downsampling network layer includes the first feature convolutional layer and the second feature convolutional layer
  • the upsampling network layer includes a feature deconvolution layer and a first feature deconvolution layer, so that the left ventricular image is automatically segmented by the image segmentation network with the above structure, which effectively improves the efficiency and effect of left ventricular image segmentation.
  • the computer-readable storage medium in the embodiments of the present invention may include any entity or device capable of carrying computer program code, and a recording medium, such as ROM/RAM, magnetic disk, optical disk, flash memory, and other memories.

Landscapes

  • Engineering & Computer Science (AREA)
  • Computer Vision & Pattern Recognition (AREA)
  • Physics & Mathematics (AREA)
  • General Physics & Mathematics (AREA)
  • Theoretical Computer Science (AREA)
  • Image Processing (AREA)
  • Image Analysis (AREA)

Abstract

一种左心室图像分割方法、装置、设备及存储介质，该方法包括：接收待分割的左心室图像（S101），将左心室图像输入训练好的图像分割网络中进行分割，图像分割网络为深度学习网络，包括下采样部分和上采样部分，下采样部分包括第一卷积层和下采样网络层，上采样部分包括第二卷积层和上采样网络层（S102），获得图像分割网络处理得到的左心室分割图像并输出（S103），从而实现左心室图像的自动分割，提高了左心室图像分割的效率和效果，适用于医学图像处理技术领域。

Description

一种左心室图像分割方法、装置、设备及存储介质 技术领域
本发明属于医学图像处理技术领域，尤其涉及一种左心室图像分割方法、装置、设备及存储介质。
背景技术
在心血管疾病的早期诊断中，需要对患者心脏进行扫描成像，并在扫描到的图像上对相应的左心室轮廓进行分割。扫描过程中，血液流动会带来成像伪影，加上人体结构的复杂性、软组织的不规则性，如何对左心室图像进行准确分割有着非常重要的临床价值。
传统对左心室图像进行分割的方式是由专业医生进行手动分割，操作繁琐，效率不高，而且手动分割容易受人工主观影响，导致左心室图像分割的准确度不高。
发明内容
本发明的目的在于提供一种左心室图像分割方法、装置、设备及存储介质，旨在解决由于现有技术无法提供一种有效的左心室图像分割方法，导致左心室图像分割效率不高且效果不佳的问题。
一方面，本发明提供了一种左心室图像分割方法，所述方法包括下述步骤：
接收待分割的左心室图像；
将所述左心室图像输入预先训练好的图像分割网络中进行分割，所述图像分割网络为深度学习网络，所述图像分割网络包括下采样部分和上采样部分，所述下采样部分包括第一卷积层和下采样网络层，所述上采样部分包括第二卷积层和上采样网络层；
获得所述图像分割网络处理得到的左心室分割图像并输出。
另一方面，本发明提供了一种左心室图像分割装置，所述装置包括：
待分割图像接收单元，用于接收待分割的左心室图像；
图像分割单元，用于将所述左心室图像输入预先训练好的图像分割网络中进行分割，所述图像分割网络为深度学习网络，所述图像分割网络包括下采样部分和上采样部分，所述下采样部分包括第一卷积层和下采样网络层，所述上采样部分包括第二卷积层和上采样网络层；以及
分割图像输出单元，用于获得所述图像分割网络处理得到的左心室分割图像并输出。
另一方面，本发明还提供了一种计算机设备，包括存储器、处理器以及存储在所述存储器中并可在所述处理器上运行的计算机程序，所述处理器执行所述计算机程序时实现如上述一种左心室图像分割方法所述的步骤。
另一方面，本发明还提供了一种计算机可读存储介质，所述计算机可读存储介质存储有计算机程序，所述计算机程序被处理器执行时实现如上述一种左心室图像分割方法所述的步骤。
本发明接收待分割的左心室图像，由训练好的图像分割网络对左心室图像进行分割，获得分割后得到的左心室分割图像并输出，其中，图像分割网络为深度学习网络，包括上采样部分和下采样部分，上采样部分包括第一卷积层和下采样网络层，上采样部分包括第二卷积层和上采样网络层，从而借助上述结构的图像分割网络对左心室图像进行自动分割，有效地提高左心室图像分割的效率和效果。
附图说明
图1是本发明实施例一提供的一种左心室图像分割方法的实现流程图；
图2是本发明实施例二提供的一种左心室图像分割方法的实现流程图；
图3是本发明实施例三提供的一种左心室图像分割装置的结构示意图；
图4是本发明实施例三提供的一种左心室图像分割装置的优选结构示意图；以及
图5是本发明实施例四提供的计算机设备的结构示意图。
具体实施方式
为了使本发明的目的、技术方案及优点更加清楚明白，以下结合附图及实施例，对本发明进行进一步详细说明。应当理解，此处所描述的具体实施例仅仅用以解释本发明，并不用于限定本发明。
以下结合具体实施例对本发明的具体实现进行详细描述：
实施例一：
图1示出了本发明实施例一提供的一种左心室图像分割方法的实现流程，为了便于说明，仅示出了与本发明实施例相关的部分，详述如下：
在步骤S101中，接收待分割的左心室图像；
本发明实施例适用于计算机、服务器等支持图像处理的设备。待分割的左心室图像可为从网上公开的医学数据库中获取的左心室图像，也可以为医院或医生提供的左心室图像，或者由CT、MRI等医学扫描设备扫描到的左心室图像。本发明实施例并不涉及左心室图像的扫描过程，在此不作描述。
在步骤S102中，将左心室图像输入预先训练好的图像分割网络中进行分割，图像分割网络为深度学习网络，包括下采样部分和上采样部分，下采样部分包括第一卷积层和下采样网络层，上采样部分包括第二卷积层和上采样网络层。
在本发明实施例中，预先训练好图像分割网络，其中，图像分割网络为深度网络，包括下采样部分和上采样部分，下采样部分包括第一卷积层和下采样网络层，上采样部分包括第二卷积层和上采样网络层。在获得左心室图像后，将左心室图像输入至第一卷积层，提取得到左心室图像的图像特征，再将图像特征输入至下采样网络层，将依次经每层下采样网络层处理后的图像特征输入 至上采样网络层进行处理，获得依次将每层下采样网络层、每层下采样网络层处理后的图像特征，再由第二卷积层进行卷积操作，实现对左心室图像的分割。
优选地，每层下采样网络层包括第一特征卷积层和第二特征卷积层，每层上采样网络层包括特征反卷积层和第一特征卷积层，从而通过不同卷积层的结合，提高图像分割网络对左心室图像特征的处理效果。其中，将每层下采样网络层处理后的图像特征进入第一层上采样网络层时，先通过第一层上采样网络层中的特征反卷积层对这些图像特征进行处理，再由第一层上采样网络层中的第一特征卷积层对该特征反卷积层处理的图像特征进行处理。
进一步优选地，图像分割网络中的下采样部分包括一层第一卷积层和三层下采样网络层，图像分割网络中的上采样部分包括三层上采样网络层和三层第二卷积层，第一卷积层和第二卷积层的卷积核大小不同，在上采样网络层中，第一特征卷积层与第二特征卷积层的步长不同，以提高图像分割网络对左心室图像特征的处理效果。作为示例地，第一卷积层的卷积核为5*5大小，第二卷积层的卷积核为1*1大小，第一特征卷积层的卷积核为5*5大小且步长为3，第二特征卷积层的卷积核为5*5大小且步长为1，特征反卷积层的卷积核大小为4*4且步长为3。
在步骤S103中，获得图像分割网络处理得到的左心室分割图像并输出。
在本发明实施例中，由图像分割网络处理得到的左心室分割图像，为待分割的左心室图像中分割出左心室轮廓图像，可帮助医生直观地观察到患者的左心室状况。
优选地，在通过第一卷积层对到待分割的左心室图像进行处理时，卷积计算公式为：
Figure PCTCN2019071075-appb-000001
其中，I表示待分割的左心室图像，m、n为第一卷积层中卷积核的宽与高，K l表示第一卷积层中的第l个卷积核，i、j为左心室图像中图像像素的位置。同样地，通过第二卷积层对上一层上采样网络输出的图像特征进行处理的公式也为上述公式， 此时，I表示一层上采样网络输出的图像特征，m、n为第二卷积层中卷积核的宽与高，K l表示第二卷积层中的第l个卷积核，i、j为图像特征中图像像素的位置。
进一步优选地，在获得经过第一卷积层提取到的图像特征后，对这些图像特征进行非线性激活，以提高图像分割网络对左心室图像的分割效果。其中，非线性激活的公式为：
Figure PCTCN2019071075-appb-000002
其中，S l(i,j)为第一卷积层提取到的图像特征，z为提取到的图像特征的数量，relu()为非线性激活函数，f(x)为非线性激活函数的输出结果。
优选地，在通过下采样网络层中的第一特征卷积层、下采样网络层中的第二特征卷积层、或者上采样网络中的第一特征卷积层对其接收到的图像特征进行处理时，包括对图像特征进行特征卷积操作和非线性变化操作，以提高对图像特征的处理效果。
进一步优选地，根据第一特征卷积层对应的网络权重和动态路由算法中的路由系数，对输入至该第一特征卷积层的图像特征进行特征卷积操作，路由系数通过动态路由算法进行更新，再将特征卷积操作处理后的图像特征输入预设的非线性变化公式，进行非线性变化操作，以提高第一特征卷积对图像特征的处理效果。
其中，特征卷积操作的公式为：
Figure PCTCN2019071075-appb-000003
其中，
Figure PCTCN2019071075-appb-000004
为输入该第一特征卷积层的图像特征，
Figure PCTCN2019071075-appb-000005
为第一特征卷积层对应的网络权重，S xy为经特征卷积操作处理后的图像特征，
Figure PCTCN2019071075-appb-000006
为动态路由算法中的路由系数，
Figure PCTCN2019071075-appb-000007
为图像特征所对应特征编号的集合，N为这些图像特征的数目，p为当前下采样网络层(或当前上采样网络层)位于图像分割网络中的网络层数，x、y为图像特征在左心室图像中的位置。
其中，非线性变化公式为：
Figure PCTCN2019071075-appb-000008
V xy为经非线性变化操作处理后的图像特征。
在通过动态路由算法更新路由系数时，通过迭代方式进行更新，其中第d次迭代过程如下：
(1)获取位于图像分割网络p层的下采样网络(或上采样网络)的路由系数：
Figure PCTCN2019071075-appb-000009
其中，k＝k h×k w为图像特征在位于图像分割网络p层的下采样网络(或上采样网络)中对应的网络权重，
Figure PCTCN2019071075-appb-000010
为预设参数，可将第一次迭代中的
Figure PCTCN2019071075-appb-000011
设为0。
(2)在位于图像分割网络p+1层的下采样网络(或上采样网络)中，对输入该层网络的图像特征进行特征卷积处理，处理公式为：
Figure PCTCN2019071075-appb-000012
(3)在位于图像分割网络p+1层的下采样网络(或上采样网络)中，对经特征卷积处理的图像特征进行非线性变化操作，公式为：
Figure PCTCN2019071075-appb-000013
(4)更新参数
Figure PCTCN2019071075-appb-000014
更新公式如下：
Figure PCTCN2019071075-appb-000015
在本发明实施例中，由训练好的图像分割网络对左心室图像进行分割，获得分割后得到的左心室分割图像并输出，其中，图像分割网络为深度学习网络，包括上采样部分和下采样部分，上采样部分包括第一卷积层和下采样网络层，上采样部分包括第二卷积层和上采样网络层，下采样网络层包括第一特征卷积层和第二特征卷积层，上采样网络层包括特征反卷积层和第一特征反卷积层，从而借助上述结构的图像分割网络对左心室图像进行自动分割，有效地提高左心室图像分割的效率和效果。
实施例二：
图2示出了本发明实施例二提供的一种左心室图像分割方法的实现流程，为了便于说明，仅示出了与本发明实施例相关的部分，详述如下：
在步骤S201中，获取采集到的左心室训练图像。
在步骤S202中，将左心室训练图像输入图像分割网络中，并通过预设的优化算法对图像分割网络进行训练，优化算法为Adam学习率自适应优化算法。
在本发明实施例中，在实施例一中已详细描述了图像分割网络的网络结构，在训练之前，按照上述网络结果构建图像分割网络。采集多张左心室训练图像，用作图像分割网络的训练样本，一一输入图像分割网络之中，并采用Adam(适应性矩估计)学习率自适应优化算法对图像分割网络进行训练，获得训练好的图像分割网络。
在步骤S203中，接收待分割的左心室图像；
在步骤S204中，将左心室图像输入预先训练好的图像分割网络中进行分割，图像分割网络包括下采样部分和上采样部分，下采样部分包括第一卷积层和下采样网络层，上采样部分包括第二卷积层和上采样网络层；
在步骤S205中，获得图像分割网络处理得到的左心室分割图像并输出。
在本发明实施例中，步骤S203～步骤S205可参照实施例一中相应步骤的详细描述，在此不再赘述。
在本发明实施例中，通过Adam学习率自适应优化算法训练图像分割网络，由训练好的图像分割网络对左心室图像进行分割，获得分割后得到的左心室分割图像并输出，其中，图像分割网络为深度学习网络，包括上采样部分和下采样部分，上采样部分包括第一卷积层和下采样网络层，上采样部分包括第二卷积层和上采样网络层，下采样网络层包括第一特征卷积层和第二特征卷积层，上采样网络层包括特征反卷积层和第一特征反卷积层，从而实现对左心室图像进行自动优化分割，有效地提高左心室图像分割的效率和效果。
实施例三：
图3示出了本发明实施例三提供的一种左心室图像分割装置的结构，为了便于说明，仅示出了与本发明实施例相关的部分，其中包括：
待分割图像接收单元31，用于接收待分割的左心室图像。
图像分割单元32，用于将左心室图像输入预先训练好的图像分割网络中进行分割，图像分割网络为深度学习网络，图像分割网络包括下采样部分和上采样部分，下采样部分包括第一卷积层和下采样网络层，上采样部分包括第二卷积层和上采样网络层。
在本发明实施例中，预先训练好图像分割网络。在获得左心室图像后，将左心室图像输入至第一卷积层，提取得到左心室图像的图像特征，再将图像特征输入至下采样网络层，将依次经每层下采样网络层处理后的图像特征输入至上采样网络层进行处理，获得依次将每层下采样网络层、每层下采样网络层处理后的图像特征，再由第二卷积层进行卷积操作，实现对左心室图像的分割。
优选地，每层下采样网络层包括第一特征卷积层和第二特征卷积层，每层上采样网络层包括特征反卷积层和第一特征卷积层，从而通过不同卷积层的结合，提高图像分割网络对左心室图像特征的处理效果。其中，将每层下采样网络层处理后的图像特征进入第一层上采样网络层时，先通过第一层上采样网络层中的特征反卷积层对这些图像特征进行处理，再由第一层上采样网络层中的第一特征卷积层对该特征反卷积层处理的图像特征进行处理。
进一步优选地，图像分割网络中的下采样部分包括一层第一卷积层和三层下采样网络层，图像分割网络中的上采样部分包括三层上采样网络层和三层第二卷积层，第一卷积层和第二卷积层的卷积核大小不同，在上采样网络层中，第一特征卷积层与第二特征卷积层的步长不同，以提高图像分割网络对左心室图像特征的处理效果。
分割图像输出单元33，用于获得图像分割网络处理得到的左心室分割图像并输出。
在本发明实施例中，由图像分割网络处理得到的左心室分割图像，为待分割的左心室图像中分割出左心室轮廓图像，可帮助医生直观地观察到患者的左心室状况。
优选地，在通过第一卷积层对到待分割的左心室图像进行处理时，卷积计算公式为：
Figure PCTCN2019071075-appb-000016
其中，I表示待分割的左心室图像，m、n为第一卷积层中卷积核的宽与高，K l表示第一卷积层中的第l个卷积核，i、j为左心室图像中图像像素的位置。同样地，通过第二卷积层对上一层上采样网络输出的图像特征进行处理的公式也为上述公式，此时，I表示一层上采样网络输出的图像特征，m、n为第二卷积层中卷积核的宽与高，K l表示第二卷积层中的第l个卷积核，i、j为图像特征中图像像素的位置。
进一步优选地，在获得经过第一卷积层提取到的图像特征后，对这些图像 特征进行非线性激活，以提高图像分割网络对左心室图像的分割效果。其中，非线性激活的公式为：
Figure PCTCN2019071075-appb-000017
其中，S l(i,j)为第一卷积层提取到的图像特征，z为提取到的图像特征的数量，relu()为非线性激活函数，f(x)为非线性激活函数的输出结果。
优选地，在通过下采样网络层中的第一特征卷积层、下采样网络层中的第二特征卷积层、或者上采样网络中的第一特征卷积层对其接收到的图像特征进行处理时，包括对图像特征进行特征卷积操作和非线性变化操作，以提高对图像特征的处理效果。
进一步优选地，根据第一特征卷积层对应的网络权重和动态路由算法中的路由系数，对输入至该第一特征卷积层的图像特征进行特征卷积操作，路由系数通过动态路由算法进行更新，再将特征卷积操作处理后的图像特征输入预设的非线性变化公式，进行非线性变化操作，以提高第一特征卷积对图像特征的处理效果。
其中，特征卷积操作的公式为：
Figure PCTCN2019071075-appb-000018
其中，
Figure PCTCN2019071075-appb-000019
为输入该第一特征卷积层的图像特征，
Figure PCTCN2019071075-appb-000020
为第一特征卷积层对应的网络权重，S xy为经特征卷积操作处理后的图像特征，
Figure PCTCN2019071075-appb-000021
为动态路由算法中的路由系数，
Figure PCTCN2019071075-appb-000022
为图像特征所对应特征编号的集合，N为这些图像特征的数目，p为当前下采样网络层(或当前上采样网络层)位于图像分割网络中的网络层数，x、y为图像特征在左心室图像中的位置。
其中，非线性变化公式为：
Figure PCTCN2019071075-appb-000023
V xy为经非线性变化操作处理后的图像特征。
在通过动态路由算法更新路由系数时，通过迭代方式进行更新，其中第d次迭代过程如下：
(1)获取位于图像分割网络p层的下采样网络(或上采样网络)的路由系数：
Figure PCTCN2019071075-appb-000024
其中，k＝k h×k w为图像特征在位于图像分割网络p层的下采样网络(或上采样网络)中对应的网络权重，
Figure PCTCN2019071075-appb-000025
为预设参数，可将第一次迭代中的
Figure PCTCN2019071075-appb-000026
设为0。
(2)在位于图像分割网络p+1层的下采样网络(或上采样网络)中，对输入该层网络的图像特征进行特征卷积处理，处理公式为：
Figure PCTCN2019071075-appb-000027
(3)在位于图像分割网络p+1层的下采样网络(或上采样网络)中，对经特征卷积处理的图像特征进行非线性变化操作，公式为：
Figure PCTCN2019071075-appb-000028
(4)更新参数
Figure PCTCN2019071075-appb-000029
更新公式如下：
Figure PCTCN2019071075-appb-000030
优选地，如图4所示，图像分割单元32还包括：
图像特征提取单元421，用于通过第一卷积层对左心室图像进行特征提取，生成左心室图像对应的图像特征；以及
图像特征处理单元422，用于将图像特征输入与第一卷积层连接的下采样网络层，依次经过每层下采样网络层、每层上采样网络层和第二卷积层的处理，获得左心室分割图像。
优选地，在训练之前，按照上述网络结果构建图像分割网络。采集多张左心室训练图像，用作图像分割网络的训练样本，一一输入图像分割网络之中，并采用Adam(适应性矩估计)学习率自适应优化算法对图像分割网络进行训练，获得训练好的图像分割网络，从而通过自适应优化算法提高图像分割网络对左心室图像的自动分割效果。
在本发明实施例中，通过Adam学习率自适应优化算法训练图像分割网络，由训练好的图像分割网络对左心室图像进行分割，获得分割后得到的左心室分割图像并输出，其中，图像分割网络为深度学习网络，包括上采样部分和下采样部分，上采样部分包括第一卷积层和下采样网络层，上采样部分包括第二卷积层和上采样网络层，下采样网络层包括第一特征卷积层和第二特征卷积层，上采样网络层包括特征反卷积层和第一特征反卷积层，从而实现对左心室图像进行自动优化分割，有效地提高左心室图像分割的效率和效果。
在本发明实施例中，一种左心室图像分割装置的各单元可由相应的硬件或软件单元实现，各单元可以为独立的软、硬件单元，也可以集成为一个软、硬件单元，在此不用以限制本发明。
实施例四：
图5示出了本发明实施例五提供的一种计算机设备的结构，为了便于说明，仅示出了与本发明实施例相关的部分。
本发明实施例的汽车5包括处理器50、存储器51以及存储在存储器51中并可在处理器50上运行的计算机程序52。该处理器50执行计算机程序52时实现上述各个方法实施例中的步骤，例如图1所示的步骤S101至S103。或者，处理器50执行计算机程序52时实现上述各装置实施例中各单元的功能，例如图3所示单元31至33的功能。
在本发明实施例中，由训练好的图像分割网络对左心室图像进行分割，获得分割后得到的左心室分割图像并输出，其中，图像分割网络为深度学习网络，包括上采样部分和下采样部分，上采样部分包括第一卷积层和下采样网络层，上采样部分包括第二卷积层和上采样网络层，下采样网络层包括第一特征卷积层和第二特征卷积层，上采样网络层包括特征反卷积层和第一特征反卷积层，从而借助上述结构的图像分割网络对左心室图像进行自动分割，有效地提高左心室图像分割的效率和效果。
实施例五：
在本发明实施例中，提供了一种计算机可读存储介质，该计算机可读存储介质存储有计算机程序，该计算机程序被处理器执行时实现上述各个方法实施例中的步骤，例如，图1所示的步骤S101至S103。或者，该计算机程序被处理器执行时实现上述各装置实施例中各单元的功能，例如图3所示单元31至33的功能。
在本发明实施例中，由训练好的图像分割网络对左心室图像进行分割，获得分割后得到的左心室分割图像并输出，其中，图像分割网络为深度学习网络，包括上采样部分和下采样部分，上采样部分包括第一卷积层和下采样网络层，上采样部分包括第二卷积层和上采样网络层，下采样网络层包括第一特征卷积层和第二特征卷积层，上采样网络层包括特征反卷积层和第一特征反卷积层，从而借助上述结构的图像分割网络对左心室图像进行自动分割，有效地提高左心室图像分割的效率和效果。
本发明实施例的计算机可读存储介质可以包括能够携带计算机程序代码的 任何实体或装置、记录介质，例如，ROM/RAM、磁盘、光盘、闪存等存储器。
以上所述仅为本发明的较佳实施例而已，并不用以限制本发明，凡在本发明的精神和原则之内所作的任何修改、等同替换和改进等，均应包含在本发明的保护范围之内。

Claims (10)

  1. 一种左心室图像分割方法，其特征在于，所述方法包括下述步骤：
    接收待分割的左心室图像；
    将所述左心室图像输入预先训练好的图像分割网络中进行分割，所述图像分割网络为深度学习网络，所述图像分割网络包括下采样部分和上采样部分，所述下采样部分包括第一卷积层和下采样网络层，所述上采样部分包括第二卷积层和上采样网络层；
    获得所述图像分割网络处理得到的左心室分割图像并输出。
  2. 如权利要求1所述的方法，其特征子在于，将所述左心室图像输入预先训练好的图像分割网络中进行分割的步骤，包括：
    通过所述第一卷积层对所述左心室图像进行特征提取，生成所述左心室图像对应的图像特征；
    将所述图像特征输入与所述第一卷积层连接的下采样网络层，依次经过每层所述下采样网络层、每层所述上采样网络层和所述第二卷积层的处理，获得所述左心室分割图像。
  3. 如权利要求2所述的方法，其特征在于，每层所述下采样网络层包括第一特征卷积层和第二特征卷积层，每层所述上采样层包括特征反卷积层和所述第一特征卷积层。
  4. 如权利要求3所述的方法，其特征在于，所述第一特征卷积层和所述第二特征卷积层包括特征卷积操作和非线性变换操作。
  5. 如权利要求4所述的方法，其特征在于，将所述图像特征输入与所述第一卷积层连接的下采样网络层的步骤，包括：
    在所述下采样网络层中，根据所述第一特征卷积层对应的网络权重和动态路由算法中的路由系数，对输入至所述第一特征卷积层的图像特征进行所述特征卷积操作，所述路由系数通过所述动态路由算法进行更新；
    将经所述特征卷积操作处理后的图像特征输入预设的非线性变换公式，进 行所述非线性变化操作。
  6. 如权利要求1所述的方法，其特征在于，接收待分割的左心室图像的步骤之前，所述方法还包括：
    获取采集到的左心室训练图像；
    将所述左心室训练图像输入所述图像分割网络中，并通过预设的优化算法对所述图像分割网络进行训练，所述优化算法为Adam学习率自适应优化算法。
  7. 一种左心室图像分割装置，其特征在于，所述装置包括：
    待分割图像接收单元，用于接收待分割的左心室图像；
    图像分割单元，用于将所述左心室图像输入预先训练好的图像分割网络中进行分割，所述图像分割网络为深度学习网络，所述图像分割网络包括下采样部分和上采样部分，所述下采样部分包括第一卷积层和下采样网络层，所述上采样部分包括第二卷积层和上采样网络层；以及
    分割图像输出单元，用于获得所述图像分割网络处理得到的左心室分割图像并输出。
  8. 如权利要求7所述的装置，其特征在于，所述图像分割单元包括：
    图像特征提取单元，用于通过所述第一卷积层对所述左心室图像进行特征提取，生成所述左心室图像对应的图像特征；以及
    图像特征处理单元，用于将所述图像特征输入与所述第一卷积层连接的下采样网络层，依次经过每层所述下采样网络层、每层所述上采样网络层和所述第二卷积层的处理，获得所述左心室分割图像。
  9. 一种计算机设备，包括存储器、处理器以及存储在所述存储器中并可在所述处理器上运行的计算机程序，其特征在于，所述处理器执行所述计算机程序时实现如权利要求1至6任一项所述方法的步骤。
  10. 一种计算机可读存储介质，所述计算机可读存储介质存储有计算机程序，其特征在于，所述计算机程序被处理器执行时实现如权利要求1至6任一项所述方法的步骤。
PCT/CN2019/071075 2018-12-14 2019-01-10 一种左心室图像分割方法、装置、设备及存储介质 Ceased WO2020118826A1 (zh)

Applications Claiming Priority (2)

Application Number Priority Date Filing Date Title
CN201811535090.9A CN109685803B (zh) 2018-12-14 2018-12-14 一种左心室图像分割方法、装置、设备及存储介质
CN201811535090.9 2018-12-14

Publications (1)

Publication Number Publication Date
WO2020118826A1 true WO2020118826A1 (zh) 2020-06-18

Family

ID=66187805

Family Applications (1)

Application Number Title Priority Date Filing Date
PCT/CN2019/071075 Ceased WO2020118826A1 (zh) 2018-12-14 2019-01-10 一种左心室图像分割方法、装置、设备及存储介质

Country Status (2)

Country Link
CN (1) CN109685803B (zh)
WO (1) WO2020118826A1 (zh)

Families Citing this family (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN110136111B (zh) * 2019-05-14 2021-08-10 深圳大学 一种心脏运动估计方法、系统及终端设备
CN110731777B (zh) * 2019-09-16 2023-07-25 平安科技（深圳）有限公司 基于图像识别的左心室测量方法、装置以及计算机设备

Citations (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN108062543A (zh) * 2018-01-16 2018-05-22 中车工业研究院有限公司 一种面部识别方法及装置
CN108537801A (zh) * 2018-03-29 2018-09-14 山东大学 基于生成对抗网络的视网膜血管瘤图像分割方法
CN108664974A (zh) * 2018-04-03 2018-10-16 华南理工大学 一种基于rgbd图像与全残差网络的语义分割方法
CN108921850A (zh) * 2018-04-16 2018-11-30 博云视觉（北京）科技有限公司 一种基于图像分割技术的图像局部特征的提取方法

Family Cites Families (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US9760807B2 (en) * 2016-01-08 2017-09-12 Siemens Healthcare Gmbh Deep image-to-image network learning for medical image analysis
CN108806793A (zh) * 2018-04-17 2018-11-13 平安科技（深圳）有限公司 病变监测方法、装置、计算机设备和存储介质

Patent Citations (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN108062543A (zh) * 2018-01-16 2018-05-22 中车工业研究院有限公司 一种面部识别方法及装置
CN108537801A (zh) * 2018-03-29 2018-09-14 山东大学 基于生成对抗网络的视网膜血管瘤图像分割方法
CN108664974A (zh) * 2018-04-03 2018-10-16 华南理工大学 一种基于rgbd图像与全残差网络的语义分割方法
CN108921850A (zh) * 2018-04-16 2018-11-30 博云视觉（北京）科技有限公司 一种基于图像分割技术的图像局部特征的提取方法

Also Published As

Publication number Publication date
CN109685803A (zh) 2019-04-26
CN109685803B (zh) 2020-10-23

Similar Documents

Publication Publication Date Title
CN110706246B (zh) 一种血管图像分割方法、装置、电子设备和存储介质
CN108537794B (zh) 医学图像数据处理方法、装置和计算机可读存储介质
Tang et al. CT image enhancement using stacked generative adversarial networks and transfer learning for lesion segmentation improvement
JP6505124B2 (ja) 適応性放射線治療における自動輪郭抽出システム及び方法
WO2019200753A1 (zh) 病变监测方法、装置、计算机设备和存储介质
WO2021017168A1 (zh) 图像分割方法、装置、设备及存储介质
CN110288611A (zh) 基于注意力机制和全卷积神经网络的冠状血管分割方法
WO2021017006A1 (zh) 图像处理方法及装置、神经网络及训练方法、存储介质
CN114155208B (zh) 一种基于深度学习的心房颤动评估方法和装置
CN111882503A (zh) 一种图像降噪方法及其应用
JP5709216B2 (ja) 画像処理プログラム、方法及び装置
CN108038840A (zh) 一种图像处理方法、装置、图像处理设备及存储介质
CN119417748B (zh) 一种肺部ct的病变与相关因素关联分析方法
CN115775219B (zh) 医学图像分割方法、系统、电子设备及介质
CN109685803B (zh) 一种左心室图像分割方法、装置、设备及存储介质
CN113538419B (zh) 一种图像处理方法和系统
CN109816665B (zh) 一种光学相干断层扫描图像的快速分割方法及装置
JP4885138B2 (ja) 一連の画像における動き修正のための方法およびシステム
US9454814B2 (en) PACS viewer and a method for identifying patient orientation
WO2022027216A1 (zh) 一种图像降噪方法及其应用
CN111489314A (zh) 图像增强方法、装置及终端设备
WO2023138273A1 (zh) 图像增强方法和系统
CN112766333B (zh) 医学影像处理模型训练方法、医学影像处理方法及装置
CN119810120B (zh) 一种基于ct图像的多脊柱结构分割方法及系统
CN111563940B (zh) 一种步进轴扫ct重建中拼接伪影的去除方法及电子介质

Legal Events

Date Code Title Description
121 Ep: the epo has been informed by wipo that ep was designated in this application

Ref document number: 19896720

Country of ref document: EP

Kind code of ref document: A1

NENP Non-entry into the national phase

Ref country code: DE

32PN Ep: public notification in the ep bulletin as address of the adressee cannot be established

Free format text: NOTING OF LOSS OF RIGHTS PURSUANT TO RULE 112(1) EPC (EPO FORM 1205A DATED 03.11.2021)

122 Ep: pct application non-entry in european phase

Ref document number: 19896720

Country of ref document: EP

Kind code of ref document: A1