WO2019109410A1 - 用于分割 mri 图像中异常信号区的全卷积网络模型训练方法 - Google Patents

用于分割 mri 图像中异常信号区的全卷积网络模型训练方法 Download PDF

Info

Publication number
WO2019109410A1
WO2019109410A1 PCT/CN2017/118298 CN2017118298W WO2019109410A1 WO 2019109410 A1 WO2019109410 A1 WO 2019109410A1 CN 2017118298 W CN2017118298 W CN 2017118298W WO 2019109410 A1 WO2019109410 A1 WO 2019109410A1
Authority
WO
WIPO (PCT)
Prior art keywords
mri
image
abnormal signal
signal region
network model
Prior art date
Application number
PCT/CN2017/118298
Other languages
English (en)
French (fr)
Inventor
马迪亚
Original Assignee
深圳博脑医疗科技有限公司
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 深圳博脑医疗科技有限公司 filed Critical 深圳博脑医疗科技有限公司
Publication of WO2019109410A1 publication Critical patent/WO2019109410A1/zh

Links

Classifications

    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/08—Learning methods
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T7/00—Image analysis
    • G06T7/10—Segmentation; Edge detection
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T7/00—Image analysis
    • G06T7/10—Segmentation; Edge detection
    • G06T7/11—Region-based segmentation
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T7/00—Image analysis
    • G06T7/10—Segmentation; Edge detection
    • G06T7/143—Segmentation; Edge detection involving probabilistic approaches, e.g. Markov random field [MRF] modelling

Definitions

  • the invention belongs to the technical field of image processing, and in particular relates to a method and a device for training a full convolutional network model for segmenting an abnormal signal region in an MRI image.
  • Magnetic resonance imaging Magnetic Resonance Imaging (MRI) can display the internal information of the brain graphically. It is a powerful tool for medical workers to analyze intracranial conditions. The abnormal signal area indicates that the MRI image of brain damage is different from the normal brain MRI image. Image segmentation based on MRI abnormal signal regions is important for the assessment of brain damage.
  • the manual segmentation mark method was used. Manual segmentation marks are cumbersome and time consuming, and are susceptible to subjective factors and mis-segmentation. Therefore, it is necessary to design an automatic and accurate segmentation algorithm to solve the problem of manual segmentation marks. Therefore, the present invention provides a more efficient and accurate method for training a full convolutional network model for segmenting anomalous signal regions in an MRI image.
  • the embodiment of the present invention provides a full convolutional network model training method for segmenting an abnormal signal region in an MRI image, so as to solve the problem that the manual segmentation flag in the prior art is very cumbersome and time-consuming, and is susceptible to subjective factors. The problem of mis-segmentation.
  • a first aspect of the embodiments of the present invention provides a method for training a full convolutional network model for segmenting an abnormal signal region in an MRI image, including:
  • the MRI sample image and the abnormal signal region segmentation sample image are trained as training samples to train the full convolution network model, and a full convolution network model for segmenting the abnormal signal region in the MRI image is obtained.
  • the structure of the full convolution network model includes: a downsampling channel and an upsampling channel.
  • the structure of the downsampling channel comprises: two three-dimensional convolution layers, one three-dimensional pooling layer, two three-dimensional convolution layers, one three-dimensional pooling layer, and two three-dimensional convolution layers;
  • the structure of the sampling channel includes: 2 three-dimensional convolution layers, one deconvolution layer, two three-dimensional convolution layers, one deconvolution layer, and two three-dimensional convolution layers.
  • the method further includes: updating the weight of the full convolution network model by using a batch random gradient descent method during the training process.
  • the initializing the weight parameter of the full convolution network model comprises: initializing a weight parameter of the full convolution network model by using a random initialization method obeying a Gaussian distribution.
  • a second aspect of the embodiments of the present invention provides a method for performing an abnormal signal region segmentation on an MRI image, which includes:
  • the server acquires an MRI sample image, and performs an abnormal signal region segmentation sample image obtained by performing an abnormal signal region segmentation on the MRI sample image; initializing a weight parameter of the full convolution network model; and extracting the MRI sample image and The abnormal signal region segmentation sample image is used as a training sample to train the full convolution network model, and a full convolution network model for segmenting the abnormal signal region in the MRI image is obtained;
  • Image segmentation the user terminal acquires the MRI image; and uses the trained full convolution network model to perform the segmentation of the abnormal signal region to obtain the segmentation image of the abnormal signal region in the MRI image.
  • a third aspect of the embodiments of the present invention provides a full convolutional network model training apparatus for an MRI image segmentation abnormal signal region, which includes:
  • a sample acquiring unit configured to acquire an MRI sample image, and perform an abnormal signal region segmentation sample image obtained by performing an abnormal signal segmentation on the MRI sample image;
  • a model initializing unit configured to initialize a weight parameter of the full convolution network model
  • the model training unit trains the MRI sample image and the abnormal signal region segmentation sample image as training samples to train the full convolution network model, and obtains a full convolution network model for segmenting the abnormal signal region in the MRI image.
  • a fourth aspect of the embodiments of the present invention provides a system for performing an abnormal signal region segmentation on an MRI image, including: a server and a user terminal, where the server includes:
  • a sample acquiring unit configured to acquire an MRI sample image, and perform an abnormal signal region segmentation sample image obtained by performing an abnormal signal region segmentation on the MRI sample image;
  • a model initializing unit configured to initialize a weight parameter of the full convolution network model
  • a model training unit training the MRI sample image and the abnormal signal region segmentation sample image as a training sample to train the full convolution network model, and obtaining a full convolution network model for segmenting an abnormal signal region in the MRI image;
  • the user terminal includes:
  • An image acquisition unit configured to acquire an MRI image
  • the image segmentation unit is configured to perform segmentation of the abnormal signal region by using the trained full convolution network model to obtain a segmentation image of the abnormal signal region in the MRI image.
  • a fifth aspect of an embodiment of the present invention provides a terminal device comprising: a memory, a processor, and a computer program stored in the memory and executable on the processor. The steps of the method of the first aspect and the second aspect are implemented when the processor executes the computer program.
  • a sixth aspect of the embodiments of the present invention provides a computer readable storage medium comprising: the computer readable storage medium storing a computer program. The steps of the method as described in the first aspect and the second aspect are implemented when the computer program is executed by a processor.
  • the trained full convolution network model of the invention directly outputs the complete segmentation probability map by using the brain MRI as an input image, and can efficiently generate accurate segmentation results without any image pre-processing and post-processing steps.
  • FIG. 1 is a schematic flowchart of an implementation process of a full convolutional network model training method for segmenting an abnormal signal region in an MRI image according to an embodiment of the present invention
  • FIG. 2 is a sample image of a brain MRI micro-oval low signal segmentation according to an embodiment of the present invention
  • FIG. 3 is a sample image of a brain MRI white signal high signal segmentation provided by an embodiment of the present invention.
  • FIG. 4 is a schematic flowchart of an implementation method of performing an abnormal signal region segmentation on an MRI image according to an embodiment of the present invention
  • FIG. 5 is a schematic diagram of a full convolutional network model training apparatus 500 for segmenting an abnormal signal region in an MRI image according to an embodiment of the present invention
  • FIG. 6 is a schematic diagram of a system 60 for performing an abnormal signal region segmentation on an MRI image according to an embodiment of the present invention
  • FIG. 7 is a schematic diagram of a terminal device according to an embodiment of the present invention.
  • FIG. 1 is a schematic flowchart of an implementation process of a full convolutional network model training method for segmenting an abnormal signal region in an MRI image, including the following steps:
  • Step S101 Acquire an MRI sample image, and perform an abnormal signal region segmentation sample image obtained by performing an abnormal signal region segmentation on the MRI sample image.
  • the abnormal signal region indicates an area in the MRI sample image in which the pixel values are different from the normal MRI image.
  • the white matter high signal and the liquid attenuation inversion (FLAIR) MRI are the abnormal signal regions.
  • the training device acquires a training sample, which is an MRI sample image and an abnormal signal region segmentation sample image obtained by performing an abnormal signal region segmentation on the MRI sample image.
  • a training sample which is an MRI sample image and an abnormal signal region segmentation sample image obtained by performing an abnormal signal region segmentation on the MRI sample image.
  • the training sample is a brain micro-oval low signal segmentation image and a corresponding MRI sample image manually labeled, as shown in FIG. 2, FIG. 2 is a brain MRI provided by an embodiment of the present invention.
  • FIG. 2 is the brain MRI sample image
  • the right image is the brain micro-oval low signal segmentation image which is manually labeled on the MRI sample image; in the segmentation of white matter damage,
  • the training sample is a manually labeled white matter high signal segmentation map and a corresponding MRI sample image, as shown in FIG. 3,
  • FIG. 3 is a brain MRI white matter high signal segmentation sample image provided by an embodiment of the present invention, and FIG. 3 is a left brain image.
  • the MRI sample image, the right image is a white matter high signal segmentation image in which the MRI sample image is manually labeled.
  • the scale is enlarged.
  • Step S102 initializing a weight parameter of the full convolution network model.
  • the structure of the full convolution network model of the present invention includes: a downsampling channel and an upsampling channel.
  • the structure of the downsampling channel comprises: two three-dimensional convolutional layers, one three-dimensional pooling layer, two three-dimensional convolutional layers, one three-dimensional pooling layer and two three-dimensional convolutional layers;
  • the structure includes: 2 three-dimensional convolution layers, one deconvolution layer, two three-dimensional convolution layers, one deconvolution layer, and two three-dimensional convolution layers.
  • the convolutional layer can only linearly transform the feature map, an activation function is added after each convolutional layer to make nonlinear changes to the feature map, which can increase the expressive power of the neural network.
  • the downsampling channel is mainly responsible for progressively extracting high-level, abstract, invariant image and semantic features. Its input is the original 3D medical image, which can be single-channel or multi-channel composed of multi-modality. 3D image.
  • all convolutional layers are three-dimensional, the step size is 1 ⁇ 1 ⁇ 1, and the convolution kernel size of the last convolutional layer is 1 ⁇ 1 ⁇ 1, and all other volumes
  • the size of the laminated convolution kernel is 3 ⁇ 3 ⁇ 3.
  • a 3 ⁇ 3 ⁇ 3 size convolution kernel is used to extract three-dimensional image features, and a 1 ⁇ 1 ⁇ 1 size convolution kernel is used to change the number of channels of the feature map.
  • the three-dimensional convolutional layer of the present invention can better extract the spatial information of the three-dimensional medical image and construct a higher level semantic abstraction. Then, using the three-dimensional pooling layer to reduce the size of the three-dimensional image features, not only can reduce the computational complexity, improve the operating efficiency of the algorithm, but also can give the network the invariance of the local features of the image.
  • the pooled kernel size of the three-dimensional pooling layer of the present invention is 2 x 2 x 2
  • the step size is 2 x 2 x 2
  • the image fill is zero. Therefore, each time the pooling operation is performed, the size of the feature map is reduced to 1/2 ⁇ 1/2 ⁇ 1/2 before the pooling.
  • the upsampling channel is composed of a plurality of successive alternating convolution layers and deconvolution layers, the convolution layer is responsible for extracting image features, and the deconvolution layer is responsible for restoring the details of the feature map. After each deconvolution, the upsampling channel combines the same size image features from the downsampling layer and upsampling across the conjoining layer as input to the next convolutional layer, which allows for better fusion.
  • Hierarchical information which complements high-level semantic information and low-level image features.
  • the deconvolution parameters in the upsampling channel are as follows, the convolution kernel size is 2 x 2 x 2, the step size is 0.5 x 0.5 x 0.5, and the image fill is zero.
  • the final layer of the full convolutional neural network of the present invention is a 3 x convolutional layer with a convolution kernel of 1 x 1 x 1 which is used to convert the three dimensional feature map into a final segmentation probability map.
  • the weighting parameter of the full convolutional network model is initialized by using a random initialization method obeying a Gaussian distribution. Specifically, the weighting of the trainable parameter of the full convolutional network model is initially described using a random initialization method obeying a Gaussian distribution. These parameters are mainly concentrated in the convolutional layer. It is assumed that the initial parameters of the full convolutional network model obey the Gaussian distribution with a mean of 0 and a variance of 0.01, and the initial values of the parameters of the full convolutional network model are assigned accordingly.
  • Step S103 The MRI sample image and the abnormal signal region segmentation sample image are used as training samples to train the full convolution network model, and a full convolution network model for segmenting the abnormal signal region in the MRI image is obtained.
  • the training samples acquired in step S101 are trained in the full convolutional network model.
  • the training samples are manually labeled micro-oval low signal segmentation maps and corresponding brain MRI images; in the segmentation of white matter damage, the training samples are manually labeled white matter high signal segmentation.
  • Figure and corresponding brain MRI images are manually labeled white matter high signal segmentation.
  • the batch random gradient descent method is used to update the parameters of the full convolutional network model.
  • the full convolutional network model randomly selects several training samples for forward propagation during each training process. And use cross entropy as a loss function to measure the accuracy of the current full convolutional network model on the training set, then calculate the partial derivative of the loss function for each full convolutional network model parameter, and then update these parameters according to the gradient descent method. Value.
  • a full convolutional network model for segmenting the abnormal signal region in the MRI image is obtained.
  • the trained full convolution network model can be directly used for the detection task of the abnormal signal region of the brain.
  • FIG. 4 is a schematic flowchart showing an implementation of a method for performing an abnormal signal region segmentation on an MRI image, including the following steps:
  • Step S401 offline training: the server acquires an MRI sample image, and an abnormal signal region segmentation sample image obtained by performing an abnormal signal region segmentation on the MRI sample image; initializing a weight parameter of the full convolution network model; and the MRI
  • the sample image and the abnormal signal region segmentation sample image are used as training samples to train the full convolution network model, and a full convolutional network model for segmenting the abnormal signal region in the MRI image is obtained.
  • the process of model training is performed in a server, and the trained model is pushed to a user terminal, such as a user terminal such as a nuclear magnetic resonance detector, and the MRI image is segmented at the user terminal.
  • a user terminal such as a nuclear magnetic resonance detector
  • the model can also be copied directly from the technician to the user terminal.
  • the server can communicate with the user terminal, and the MRI image acquired by the user terminal and the segmentation result thereof are uploaded to the server, and the server optimizes the model according to the preset period by the acquired MRI image and the corresponding segmentation result, and optimizes the model.
  • the model is pushed to the user terminal in real time for model update.
  • Step S402 image segmentation: the user terminal acquires the MRI image; and uses the trained full convolution network model to perform the segmentation of the abnormal signal region to obtain the segmentation image of the abnormal signal region in the MRI image.
  • the trained full convolution network can be directly used for the detection of abnormal brain signal areas.
  • the training samples are manually labeled low micro-element low signal segmentation maps and corresponding original ones.
  • MRI after training, obtains a full convolutional neural network for detecting micro-oval low signal in the brain, using the original brain MRI as the input image, without any pre-processing and post-processing, the full convolutional neural network can be directly
  • the segmentation probability map of the brain's tiny egg-type low signal is output, and then the position of the low-signal of each brain micro-oval is located by processing the probability map, and the best result is obtained in the positioning accuracy.
  • the same training method and network structure are used.
  • a full convolutional neural network of signals Only by changing the training samples to the manually labeled white matter high signal segmentation map and the corresponding MRI images, one can be used to segment the white matter high.
  • a full convolutional neural network of signals Using the brain MRI as the input image, the full convolutional neural network can directly output the segmentation probability map, and then obtain the final segmentation result and the volume measurement result of the white matter high signal region.
  • the full convolutional neural network of the present invention is an end-to-end (image to image) trainable network model.
  • the entire network takes the original nuclear magnetic resonance image (MRI) as input, and undergoes multiple convolution and pooling processes directly.
  • MRI nuclear magnetic resonance image
  • the complete segmentation probability map is output, and accurate segmentation results can be efficiently generated without any image pre-processing and post-processing steps.
  • the full convolutional network model training device 500 for segmenting anomalous signal regions in an MRI image includes a sample acquiring unit 501, a model initializing unit 502, and a model training unit 503.
  • a sample acquiring unit 501 configured to acquire an MRI sample image, and an abnormal signal region segmentation sample image obtained by performing an abnormal signal segmentation on the MRI sample image;
  • a model initialization unit 502 configured to initialize a weight parameter of the full convolution network model
  • the model training unit 503 trains the MRI sample image and the abnormal signal region segmentation sample image as training samples to train the full convolution network model to obtain a full convolution network model for segmenting the abnormal signal region in the MRI image.
  • FIG. 6 is a schematic diagram of a system 60 for performing an abnormal signal region segmentation on an MRI image.
  • the system for performing an abnormal signal region segmentation on an MRI image includes: a server 61 and a user terminal 62, wherein The server 61 includes:
  • a sample acquiring unit 611 configured to acquire an MRI sample image, and perform an abnormal signal region segmentation sample image obtained by performing an abnormal signal region segmentation on the MRI sample image;
  • a model initializing unit 612 configured to initialize a weight parameter of the full convolution network model
  • the model training unit 613 trains the MRI sample image and the abnormal signal region segmentation sample image as training samples to train the full convolution network model, and obtains a full convolution network model for segmenting the abnormal signal region in the MRI image.
  • the user terminal 62 includes:
  • An image obtaining unit 621 configured to acquire an MRI image
  • the image segmentation unit 622 is configured to perform the segmentation of the abnormal signal region by using the trained full convolution network model to obtain a segmentation image of the abnormal signal region in the MRI image.
  • FIG. 7 is a schematic diagram of a terminal device according to an embodiment of the present invention.
  • the terminal device 7 of this embodiment includes a processor 70, a memory 71, and a computer program 72 stored in the memory 71 and operable on the processor 70.
  • the processor 70 when executing the computer program 52, implements the steps in the various method embodiments described above, such as steps 101 through 103 shown in FIG.
  • the processor 70 executes the computer program 72, the functions of the modules/units in the foregoing device embodiments are implemented, such as the functions of the modules 501 to 503 shown in FIG.
  • the computer program 72 can be partitioned into one or more modules/units that are stored in the memory 71 and executed by the processor 70 to complete this invention.
  • the one or more modules/units may be a series of computer program instruction segments capable of performing a particular function, the instruction segments being used to describe the execution of the computer program 72 in the terminal device 7.
  • the terminal device 7 may be a computing device such as a desktop computer, a notebook, a palmtop computer, and a cloud server.
  • the terminal device may include, but is not limited to, a processor 70 and a memory 71. It will be understood by those skilled in the art that FIG. 7 is only an example of the terminal device 7, and does not constitute a limitation of the terminal device 7, and may include more or less components than those illustrated, or combine some components or different components.
  • the terminal device may further include an input/output device, a network access device, a bus, and the like.
  • the so-called processor 70 can be a central processing unit (Central Processing Unit, CPU), can also be other general-purpose processors, digital signal processors (DSP), application specific integrated circuits (Application Specific Integrated Circuit (ASIC), Field-Programmable Gate Array (FPGA) or other programmable logic device, discrete gate or transistor logic device, discrete hardware components, etc.
  • the general purpose processor may be a microprocessor or the processor or any conventional processor or the like.
  • the memory 71 may be an internal storage unit of the terminal device 7, such as a hard disk or a memory of the terminal device 7.
  • the memory 71 may also be an external storage device of the terminal device 5, for example, a plug-in hard disk equipped on the terminal device 7, a smart memory card (SMC), and a secure digital (SD). Card, flash card (Flash Card) and so on.
  • the memory 71 may also include both an internal storage unit of the terminal device 7 and an external storage device.
  • the memory 71 is used to store the computer program and other programs and data required by the terminal device.
  • the memory 71 can also be used to temporarily store data that has been output or is about to be output.
  • each functional unit and module described above is exemplified. In practical applications, the above functions may be assigned to different functional units as needed.
  • the module is completed by dividing the internal structure of the device into different functional units or modules to perform all or part of the functions described above.
  • Each functional unit and module in the embodiment may be integrated into one processing unit, or each unit may exist physically separately, or two or more units may be integrated into one unit, and the integrated unit may be hardware.
  • Formal implementation can also be implemented in the form of software functional units.
  • the specific names of the respective functional units and modules are only for the purpose of facilitating mutual differentiation, and are not intended to limit the scope of protection of the present application.
  • For the specific working process of the unit and the module in the foregoing system reference may be made to the corresponding process in the foregoing method embodiment, and details are not described herein again.
  • the disclosed apparatus/terminal device and method may be implemented in other manners.
  • the device/terminal device embodiments described above are merely illustrative.
  • the division of the modules or units is only a logical function division.
  • there may be another division manner for example, multiple units.
  • components may be combined or integrated into another system, or some features may be omitted or not performed.
  • the mutual coupling or direct coupling or communication connection shown or discussed may be an indirect coupling or communication connection through some interface, device or unit, and may be in electrical, mechanical or other form.
  • the units described as separate components may or may not be physically separated, and the components displayed as units may or may not be physical units, that is, may be located in one place, or may be distributed to multiple network units. Some or all of the units may be selected according to actual needs to achieve the purpose of the solution of the embodiment.
  • each functional unit in each embodiment of the present invention may be integrated into one processing unit, or each unit may exist physically separately, or two or more units may be integrated into one unit.
  • the above integrated unit can be implemented in the form of hardware or in the form of a software functional unit.
  • the integrated modules/units if implemented in the form of software functional units and sold or used as separate products, may be stored in a computer readable storage medium. Based on such understanding, the present invention implements all or part of the processes in the foregoing embodiments, and may also be completed by a computer program to instruct related hardware.
  • the computer program may be stored in a computer readable storage medium. The steps of the various method embodiments described above may be implemented when the program is executed by the processor.
  • the computer program comprises computer program code, which may be in the form of source code, object code form, executable file or some intermediate form.
  • the computer readable medium may include any entity or device capable of carrying the computer program code, a recording medium, a USB flash drive, a removable hard disk, a magnetic disk, an optical disk, a computer memory, a read-only memory (ROM). Random access memory (Random Access Memory, RAM), electrical carrier signals, telecommunications signals, and software distribution media. It should be noted that the content contained in the computer readable medium may be appropriately increased or decreased according to the requirements of legislation and patent practice in a jurisdiction, for example, in some jurisdictions, according to legislation and patent practice, computer readable media Does not include electrical carrier signals and telecommunication signals.

Abstract

一种用于分割MRI图像中异常信号区的全卷积网络模型训练方法。包括：获取MRI样本图像，和对所述MRI样本图像进行异常信号区分割后得到的异常信号区分割样本图像（S101）；初始化所述全卷积网络模型的权重参数（S102）；将所述MRI样本图像和分割样本图像作为训练样本训练所述全卷积网络模型，得到用于分割MRI图像中异常信号区的全卷积网络模型（S103）。该方法可以解决现有技术中人工分割标记十分繁琐且耗时，容易受到主观因素的影响而产生误分割的问题，无需任何的图像预处理和后处理步骤就可以高效地产生精准的分割结果。

Description

用于分割MRI图像中异常信号区的全卷积网络模型训练方法 技术领域
本发明属于图像处理技术领域，尤其涉及用于分割MRI图像中异常信号区的全卷积网络模型训练方法及装置。
 背景技术
脑部损伤严重危及患者生命，例如脑肿瘤、脑出血、脑白质损伤等。核磁共振成像（Magnetic Resonance Imaging，MRI）可以以图像方式显示大脑内部信息，是医学工作者分析颅内情况的有力工具，异常信号区表示脑部损伤的MRI图像与正常的大脑MRI图像相比像素值不同的区域，基于MRI异常信号区图像分割对于脑部损伤的评估十分重要。早期，使用的是手工分割标记方法，人工分割标记十分繁琐且耗时，容易受到主观因素的影响而产生误分割。因此，设计一种自动的精准的分割算法来解决手工分割标记的不足是必需的。因此本发明提供了一种更高效、更准确的用于分割MRI图像中异常信号区的全卷积网络模型训练方法。
 技术问题
有鉴于此，本发明实施例提供了用于分割MRI图像中异常信号区的全卷积网络模型训练方法，以解决现有技术中人工分割标记十分繁琐且耗时，容易受到主观因素的影响而产生误分割的问题。
 技术解决方案
本发明实施例的第一方面提供了一种用于分割MRI图像中异常信号区的全卷积网络模型训练方法，其特征在于，包括：
获取MRI样本图像，和对所述MRI样本图像进行异常信号区分割后得到的异常信号区分割样本图像；
初始化所述全卷积网络模型的权重参数；
将所述MRI样本图像和异常信号区分割样本图像作为训练样本训练所述全卷积网络模型，得到用于分割MRI图像中异常信号区的全卷积网络模型。
进一步的，所述全卷积网络模型的结构包括：降采样通道及升采样通道。
进一步的，所述降采样通道的结构包括：2个三维卷积层、1个三维池化层、2个三维卷积层、1个三维池化层和2个三维卷积层；所述升采样通道的结构包括：2个三维卷积层、1个反卷积层、2个三维卷积层、1个反卷积层和2个三维卷积层。
进一步的，还包括：在训练过程中使用批量随机梯度下降法更新所述全卷积网络模型的权重。
进一步的，所述初始化所述全卷积网络模型的权重参数，包括：使用服从高斯分布的随机初始化方法来初始化所述全卷积网络模型的权重参数。
本发明实施例的第二方面提供了一种对MRI图像进行异常信号区分割的方法，其特征在于，包括：
离线训练：服务器获取MRI样本图像，和对所述MRI样本图像进行异常信号区分割后得到的异常信号区分割样本图像；初始化所述全卷积网络模型的权重参数；将所述MRI样本图像和异常信号区分割样本图像作为训练样本训练所述全卷积网络模型，得到用于分割MRI图像中异常信号区的全卷积网络模型；
图像分割：用户终端获取MRI图像；利用训练好的所述全卷积网络模型进行异常信号区分割，得到MRI图像中异常信号区的分割图像。
本发明实施例的第三方面提供了一种用于MRI图像分割异常信号区的全卷积网络模型训练装置，其特征在于，包括：
样本获取单元，用于获取MRI样本图像，和对所述MRI样本图像进行异常信号分割后得到的异常信号区分割样本图像；
模型初始化单元，用于初始化所述全卷积网络模型的权重参数；
模型训练单元，将所述MRI样本图像和异常信号区分割样本图像作为训练样本训练所述全卷积网络模型，得到用于分割MRI图像中异常信号区的全卷积网络模型。
本发明实施例的第四方面提供了一种对MRI图像进行异常信号区分割的系统，其特征在于，包括：服务器和用户终端，其中，所述服务器包括：
样本获取单元，用于获取MRI样本图像，和对所述MRI样本图像进行异常信号区分割后得到的异常信号区分割样本图像；
模型初始化单元，用于初始化所述全卷积网络模型的权重参数；
模型训练单元，将所述MRI样本图像和异常信号区分割样本图像作为训练样本训练所述全卷积网络模型，得到用于分割MRI图像中异常信号区的全卷积网络模型；
所述用户终端包括：
图像获取单元，用于获取MRI图像；
图像分割单元，用于利用训练好的所述全卷积网络模型进行异常信号区分割，得到MRI图像中异常信号区的分割图像。
本发明实施例的第五方面提供了一种终端设备，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第一方面和第二方面所述方法的步骤。
本发明实施例的第六方面提供了一种计算机可读存储介质，包括：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第一方面和第二方面所述方法的步骤。
 有益效果
本发明训练好的全卷积网络模型，以脑部MRI作为输入图像，直接输出完整的分割概率图，无需任何的图像预处理和后处理步骤就可以高效地产生精准的分割结果。
 附图说明
为了更清楚地说明本发明实施例中的技术方案，下面将对实施例或现有技术描述中所需要使用的附图作简单地介绍，显而易见地，下面描述中的附图仅仅是本发明的一些实施例，对于本领域普通技术人员来讲，在不付出创造性劳动性的前提下，还可以根据这些附图获得其他的附图。
图1是本发明实施例提供的一种用于分割MRI图像中异常信号区的全卷积网络模型训练方法的实现流程示意图；
图2是本发明实施例提供的大脑MRI微小卵型低信号分割样本图像；
图3是本发明实施例提供的大脑MRI脑白质高信号分割样本图像；
图4是本发明实施例提供的一种对MRI图像进行异常信号区分割的方法的实现流程示意图；
图5是本发明实施例提供的一种用于分割MRI图像中异常信号区的全卷积网络模型训练装置500的示意图；
图6是本发明实施例提供的一种对MRI图像进行异常信号区分割的系统60的示意图；
图7是本发明实施例提供的终端设备的示意图。
 本发明的实施方式
以下描述中，为了说明而不是为了限定，提出了诸如特定系统结构、技术之类的具体细节，以便透彻理解本发明实施例。然而，本领域的技术人员应当清楚，在没有这些具体细节的其它实施例中也可以实现本发明。在其它情况中，省略对众所周知的系统、装置、电路以及方法的详细说明，以免不必要的细节妨碍本发明的描述。
为了说明本发明所述的技术方案，下面通过具体实施例来进行说明。
请参考图1所示，是本发明提供一种用于分割MRI图像中异常信号区的全卷积网络模型训练方法的实现流程示意图，包括以下步骤：
步骤S101，获取MRI样本图像，和对所述MRI样本图像进行异常信号区分割后得到的异常信号区分割样本图像。
在本发明实施例中，异常信号区表示MRI样本图像中与正常MRI图像相比像素值不同的区域。例如磁敏感加权成像（Susceptibility Weighted Imaging，SWI） MRI中的微小卵型低信号和液体衰减反转恢复（Fluid Attenuated Inversion Recovery,FLAIR）MRI中的白质高信号就是异常信号区。
在训练开始之前，训练装置获取训练样本，训练样本为MRI样本图像和对所述MRI样本图像进行异常信号区分割后得到的异常信号区分割样本图像。例如，在大脑微出血的检测中，训练样本为由手工标记好的大脑微小卵型低信号分割图像和对应的MRI样本图像，如图2所示，图2是本发明实施例提供的大脑MRI微小卵型低信号分割样本图像，图2左图为大脑MRI样本图像，右图为对所述MRI样本图像进行手工标记好的大脑微小卵型低信号分割图像；在脑白质损伤的分割中，训练样本为手工标记的脑白质高信号分割图和对应的MRI样本图像，如图3所示，图3是本发明实施例提供的大脑MRI脑白质高信号分割样本图像，图3左图为大脑MRI样本图像，右图为对所述MRI样本图像进行手工标记的脑白质高信号分割图像。此处为了让分割图像更清楚，比例进行了放大。
步骤S102，初始化所述全卷积网络模型的权重参数。
本发明所述的全卷积网络模型的结构包括：降采样通道及升采样通道。所述降采样通道的结构包括：2个三维卷积层、1个三维池化层、2个三维卷积层、1个三维池化层和2个三维卷积层；所述升采样通道的结构包括：2个三维卷积层、1个反卷积层、2个三维卷积层、1个反卷积层和2个三维卷积层。
由于卷积层只能对特征图进行线性变换，在每个卷积层后都添加了一个激活函数，用于对特征图做非线性变化，这样可以增加神经网络网络的表达能力。降采样通道主要负责渐进地提取高层级的、抽象的、不变的图像和语义特征，它的输入是原始的三维医学图像，可以是单通道的，也可以是由多模态构成的多通道的三维图像。在本发明的全卷积网络结构中，所有的卷积层都是三维的，步长是1×1×1，最后一个卷积层的卷积核大小是1×1×1，其余所有卷积层的卷积核的大小都是3×3×3。 3×3×3大小的卷积核用于提取三维图像特征，1×1×1大小的卷积核用于改变特征图的通道数。
相比于用于自然图像的二维卷积层，本发明的三维卷积层可以更好的提取三维医学图像的空间信息，构建更高层次的语义抽象。然后，使用三维的池化层来降低三维图像特征的大小，这样不仅可以降低计算的复杂度，提高算法的运行效率，也可以赋予网络对于图像局部特征的不变性。本发明的三维池化层的池化核大小是2×2×2，步长是2×2×2，图像填充为0。因此每经过一次池化操作，特征图的大小缩小为池化前的1/2×1/2×1/2。虽然降采样通道对于医学图像有比较好的抽象作用，但是经过多次卷积和核池化操作后，图像的细节信息损失比较大，因而本发明采用了和降采样通道完全对称的升采样通道来恢复图像的细节信息。升采样通道由多个连续交替的卷积层和反卷积层组成，卷积层负责提取图像特征，反卷积层负责恢复特征图的细节信息。在每次反卷积后，升采样通道通过跨连接层将来自降采样层和升采样中得到的相同大小的图像特征合并起来，作为接下来卷积层的输入，这样可以更好的融合多层级的信息，对于高级语义信息和低级图像特征有互补的作用。在升采样通道中的反卷积参数如下，卷积核大小是2×2×2，步长是0.5×0.5×0.5，图像填充是0。本发明的全卷积神经网络的最后一层是一个卷积核为1×1×1的三维卷积层，它用于把三维的特征图转换成最终的分割概率图。
本实施例使用服从高斯分布的随机初始化方法来初始化所述全卷积网络模型的权重参数，具体的，使用服从高斯分布的随机初始化方法来初始述全卷积网络模型的可训练参数的权值，这些参数主要集中在卷积层，假设全卷积网络模型的初始参数服从均值为0，方差为0.01的高斯分布，并据此为全卷积网络模型的参数赋初始值。
步骤S103 , 将所述MRI样本图像和异常信号区分割样本图像作为训练样本训练所述全卷积网络模型，得到用于分割MRI图像中异常信号区的全卷积网络模型。
将步骤S101获取的训练样本训练所述全卷积网络模型。例如，在大脑微出血的检测中，训练样本为手工标记好的大脑微小卵型低信号分割图和对应的大脑MRI图像；在脑白质损伤的分割中训练样本为手工标记的脑白质高信号分割图和对应的大脑MRI图像。
在训练全卷积网络模型时，使用批量随机梯度下降法来更新全卷积网络模型的参数，具体地，全卷积网络模型在每次训练过程中会随机选取几个训练样本进行前向传播，并使用交叉熵作为损失函数来衡量当前全卷积网络模型在训练集上的准确度，然后计算损失函数对于每个全卷积网络模型参数的偏导数，再根据梯度下降法来更新这些参数的值。
经过训练后得到用于分割MRI图像中异常信号区的全卷积网络模型，经过训练后的全卷积网络模型可以直接用于大脑异常信号区的检测任务。
图4所示为一种对MRI图像进行异常信号区分割的方法的实现流程示意图，包括以下步骤：
步骤S401，离线训练：服务器获取MRI样本图像，和对所述MRI样本图像进行异常信号区分割后得到的异常信号区分割样本图像；初始化所述全卷积网络模型的权重参数；将所述MRI样本图像和异常信号区分割样本图像作为训练样本训练所述全卷积网络模型，得到用于分割MRI图像中异常信号区的全卷积网络模型。
本发明实施例中，模型训练的过程在服务器中进行，将训练后的模型推送到用户终端，如核磁共振检测仪等用户终端，在用户终端对MRI图像进行分割。此外，模型也可以直接由技术人员拷贝至用户终端。
此外，服务器可以与用户终端进行通信，用户终端获取的MRI图像及其分割结果会上传至服务器，服务器通过获取的MRI图像及其对应的分割结果按照预设周期对模型进行优化，并将优化后的模型实时推送至用户终端进行模型更新。通过这种设置，能够进一步提高模型的分割精度，从而获得更准确的分割结果。
步骤S402，图像分割：用户终端获取MRI图像；利用训练好的所述全卷积网络模型进行异常信号区分割，得到MRI图像中异常信号区的分割图像。
经过训练后的全卷积网络可以直接用于大脑异常信号区的检测任务，在大脑微出血的检测任务中，训练样本为由手工标记好的大脑微小卵型低信号分割图和对应的原始的MRI，经过训练后得到一个用于检测大脑微小卵型低信号的全卷积神经网络，以原始的脑部MRI作为输入图像，无需经过任何预处理和后处理，全卷积神经网络就可以直接输出大脑微小卵型低信号的分割概率图，进而通过对概率图的处理来定位每个大脑微小卵型低信号的位置，在定位准确度上达到了目前最好的结果。在脑白质异常的分割中，使用了完全相同的训练方法和网络结构，仅仅将训练样本改变为手工标记的脑白质高信号分割图和对应的MRI图像就可以训练出一个用于分割脑白质高信号的全卷积神经网络。以脑部MRI作为输入图像，全卷积神经网络可以直接输出分割概率图，进而得到最终的分割结果和脑白质高信号区域的体积测量结果。
本发明的全卷积神经网络是一种端到端（图像到图像）的可训练网络模型，整个网络以原始的核磁共振图像（MRI）作为输入，经过多重的卷积和池化处理，直接输出完整的分割概率图，无需任何的图像预处理和后处理步骤就可以高效地产生精准的分割结果。
图5为一种用于分割MRI图像中异常信号区的全卷积网络模型训练装置500的示意图，此实施例中未详细描述之处，请参见前面方法的实施例。如图3所示，所述用于分割MRI图像中异常信号区的全卷积网络模型训练装置包括：样本获取单元501，模型初始化单元502，模型训练单元503。
样本获取单元501，用于获取MRI样本图像，和对所述MRI样本图像进行异常信号分割后得到的异常信号区分割样本图像；
模型初始化单元502，用于初始化所述全卷积网络模型的权重参数；
模型训练单元503，将所述MRI样本图像和异常信号区分割样本图像作为训练样本训练所述全卷积网络模型，得到用于分割MRI图像中异常信号区的全卷积网络模型。
图6为一种对MRI图像进行异常信号区分割的系统60的示意图，如图6所示，所述一种对MRI图像进行异常信号区分割的系统包括：服务器61和用户终端62，其中，所述服务器61包括：
样本获取单元611，用于获取MRI样本图像，和对所述MRI样本图像进行异常信号区分割后得到的异常信号区分割样本图像；
模型初始化单元612，用于初始化所述全卷积网络模型的权重参数；
模型训练单元613，将所述MRI样本图像和异常信号区分割样本图像作为训练样本训练所述全卷积网络模型，得到用于分割MRI图像中异常信号区的全卷积网络模型。
所述用户终端62包括：
图像获取单元621，用于获取MRI图像；
图像分割单元622，用于利用训练好的所述全卷积网络模型进行异常信号区分割，得到MRI图像中异常信号区的分割图像。
应理解，上述实施例中各步骤的序号的大小并不意味着执行顺序的先后，各过程的执行顺序应以其功能和内在逻辑确定，而不应对本发明实施例的实施过程构成任何限定。在上述实施例中，对各个实施例的描述都各有侧重，某个实施例中没有详述或记载的部分，可以参见其它实施例的相关描述。
图7是本发明一实施例提供的一种终端设备的示意图。如图5所示，该实施例的终端设备7包括：处理器70、存储器71以及存储在所述存储器71中并可在所述处理器70上运行的计算机程序72。所述处理器70执行所述计算机程序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101至103。或者，所述处理器70执行所述计算机程序72时实现上述各装置实施例中各模块/单元的功能，例如图5所示模块501至503的功能。
示例性的，所述计算机程序72可以被分割成一个或多个模块/单元，所述一个或者多个模块/单元被存储在所述存储器71中，并由所述处理器70执行，以完成本发明。所述一个或多个模块/单元可以是能够完成特定功能的一系列计算机程序指令段，该指令段用于描述所述计算机程序72在所述终端设备7中的执行过程。
所述终端设备7可以是桌上型计算机、笔记本、掌上电脑及云端服务器等计算设备。所述终端设备可包括，但不仅限于，处理器70、存储器71。本领域技术人员可以理解，图7仅仅是终端设备7的示例，并不构成对终端设备7的限定，可以包括比图示更多或更少的部件，或者组合某些部件，或者不同的部件，例如所述终端设备还可以包括输入输出设备、网络接入设备、总线等。
所称处理器70可以是中央处理单元(Central Processing Unit，CPU)，还可以是其他通用处理器、数字信号处理器 (Digital Signal Processor，DSP)、专用集成电路 (Application Specific Integrated Circuit，ASIC)、现成可编程门阵列 (Field-Programmable Gate Array，FPGA) 或者其他可编程逻辑器件、分立门或者晶体管逻辑器件、分立硬件组件等。通用处理器可以是微处理器或者该处理器也可以是任何常规的处理器等。
所述存储器71可以是所述终端设备7的内部存储单元，例如终端设备7的硬盘或内存。所述存储器71也可以是所述终端设备5的外部存储设备，例如所述终端设备7上配备的插接式硬盘，智能存储卡（Smart Media Card, SMC），安全数字（Secure Digital, SD）卡，闪存卡（Flash Card）等。进一步地，所述存储器71还可以既包括所述终端设备7的内部存储单元也包括外部存储设备。所述存储器71用于存储所述计算机程序以及所述终端设备所需的其他程序和数据。所述存储器71还可以用于暂时地存储已经输出或者将要输出的数据。
所属领域的技术人员可以清楚地了解到，为了描述的方便和简洁，仅以上述各功能单元、模块的划分进行举例说明，实际应用中，可以根据需要而将上述功能分配由不同的功能单元、模块完成，即将所述装置的内部结构划分成不同的功能单元或模块，以完成以上描述的全部或者部分功能。实施例中的各功能单元、模块可以集成在一个处理单元中，也可以是各个单元单独物理存在，也可以两个或两个以上单元集成在一个单元中，上述集成的单元既可以采用硬件的形式实现，也可以采用软件功能单元的形式实现。另外，各功能单元、模块的具体名称也只是为了便于相互区分，并不用于限制本申请的保护范围。上述系统中单元、模块的具体工作过程，可以参考前述方法实施例中的对应过程，在此不再赘述。
在上述实施例中，对各个实施例的描述都各有侧重，某个实施例中没有详述或记载的部分，可以参见其它实施例的相关描述。
本领域普通技术人员可以意识到，结合本文中所公开的实施例描述的各示例的单元及算法步骤，能够以电子硬件、或者计算机软件和电子硬件的结合来实现。这些功能究竟以硬件还是软件方式来执行，取决于技术方案的特定应用和设计约束条件。专业技术人员可以对每个特定的应用来使用不同方法来实现所描述的功能，但是这种实现不应认为超出本发明的范围。
在本发明所提供的实施例中，应该理解到，所揭露的装置/终端设备和方法，可以通过其它的方式实现。例如，以上所描述的装置/终端设备实施例仅仅是示意性的，例如，所述模块或单元的划分，仅仅为一种逻辑功能划分，实际实现时可以有另外的划分方式，例如多个单元或组件可以结合或者可以集成到另一个系统，或一些特征可以忽略，或不执行。另一点，所显示或讨论的相互之间的耦合或直接耦合或通讯连接可以是通过一些接口，装置或单元的间接耦合或通讯连接，可以是电性，机械或其它的形式。
所述作为分离部件说明的单元可以是或者也可以不是物理上分开的，作为单元显示的部件可以是或者也可以不是物理单元，即可以位于一个地方，或者也可以分布到多个网络单元上。可以根据实际的需要选择其中的部分或者全部单元来实现本实施例方案的目的。
另外，在本发明各个实施例中的各功能单元可以集成在一个处理单元中，也可以是各个单元单独物理存在，也可以两个或两个以上单元集成在一个单元中。上述集成的单元既可以采用硬件的形式实现，也可以采用软件功能单元的形式实现。
所述集成的模块/单元如果以软件功能单元的形式实现并作为独立的产品销售或使用时，可以存储在一个计算机可读取存储介质中。基于这样的理解，本发明实现上述实施例方法中的全部或部分流程，也可以通过计算机程序来指令相关的硬件来完成，所述的计算机程序可存储于一计算机可读存储介质中，该计算机程序在被处理器执行时，可实现上述各个方法实施例的步骤。其中，所述计算机程序包括计算机程序代码，所述计算机程序代码可以为源代码形式、对象代码形式、可执行文件或某些中间形式等。所述计算机可读介质可以包括：能够携带所述计算机程序代码的任何实体或装置、记录介质、U盘、移动硬盘、磁碟、光盘、计算机存储器、只读存储器（Read-Only Memory，ROM）、随机存取存储器（Random Access Memory，RAM）、电载波信号、电信信号以及软件分发介质等。需要说明的是，所述计算机可读介质包含的内容可以根据司法管辖区内立法和专利实践的要求进行适当的增减，例如在某些司法管辖区，根据立法和专利实践，计算机可读介质不包括电载波信号和电信信号。
以上所述实施例仅用以说明本发明的技术方案，而非对其限制；尽管参照前述实施例对本发明进行了详细的说明，本领域的普通技术人员应当理解：其依然可以对前述各实施例所记载的技术方案进行修改，或者对其中部分技术特征进行等同替换；而这些修改或者替换，并不使相应技术方案的本质脱离本发明各实施例技术方案的精神和范围，均应包含在本发明的保护范围之内。

Claims (10)

  1. 一种用于分割MRI图像中异常信号区的全卷积网络模型训练方法，其特征在于，包括：
    获取MRI样本图像，和对所述MRI样本图像进行异常信号区分割后得到的异常信号区分割样本图像；
    初始化所述全卷积网络模型的权重参数；
    将所述MRI样本图像和异常信号区分割样本图像作为训练样本训练所述全卷积网络模型，得到用于分割MRI图像中异常信号区的全卷积网络模型。
  2. 如权利要求1所述的方法，其特征在于，所述全卷积网络模型的结构包括：降采样通道及升采样通道。
  3. 如权利要求2所述的方法，其特征在于，所述降采样通道的结构包括：2个三维卷积层、1个三维池化层、2个三维卷积层、1个三维池化层和2个三维卷积层；所述升采样通道的结构包括：2个三维卷积层、1个反卷积层、2个三维卷积层、1个反卷积层和2个三维卷积层。
  4. 如权利要求1-3任一项所述的方法，其特征在于，还包括：在训练过程中使用批量随机梯度下降法更新所述全卷积网络模型的权重。
  5. 如权利要求1-3任一项所述的方法，其特征在于，所述初始化所述全卷积网络模型的权重参数，包括：使用服从高斯分布的随机初始化方法来初始化所述全卷积网络模型的权重参数。
  6. 一种对MRI图像进行异常信号区分割的方法，其特征在于，包括：
    离线训练：服务器获取MRI样本图像，和对所述MRI样本图像进行异常信号区分割后得到的异常信号区分割样本图像；初始化所述全卷积网络模型的权重参数；将所述MRI样本图像和异常信号区分割样本图像作为训练样本训练所述全卷积网络模型，得到用于分割MRI图像中异常信号区的全卷积网络模型；
    图像分割：用户终端获取MRI图像；利用训练好的所述全卷积网络模型进行异常信号区分割，得到MRI图像中异常信号区的分割图像。
  7. 一种用于分割MRI图像中异常信号区的全卷积网络模型训练装置，其特征在于，包括：
    样本获取单元，用于获取MRI样本图像，和对所述MRI样本图像进行异常信号区分割后得到的异常信号区分割样本图像；
    模型初始化单元，用于初始化所述全卷积网络模型的权重参数；
    模型训练单元，将所述MRI样本图像和异常信号区分割样本图像作为训练样本训练所述全卷积网络模型，得到用于分割MRI图像中异常信号区的全卷积网络模型。
  8. 一种对MRI图像进行异常信号区分割的系统，其特征在于，包括：服务器和用户终端，其中，所述服务器包括：
    样本获取单元，用于获取MRI样本图像，和对所述MRI样本图像进行异常信号区分割后得到的异常信号区分割样本图像；
    模型初始化单元，用于初始化所述全卷积网络模型的权重参数；
    模型训练单元，将所述MRI样本图像和异常信号区分割样本图像作为训练样本训练所述全卷积网络模型，得到用于分割MRI图像中异常信号区的全卷积网络模型；
    所述用户终端包括：
    图像获取单元，用于获取MRI图像；
    图像分割单元，用于利用训练好的所述全卷积网络模型进行异常信号区分割，得到MRI图像中异常信号区的分割图像。
  9. 一种终端设备，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5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5任一项所述方法的步骤。
PCT/CN2017/118298 2017-12-06 2017-12-25 用于分割 mri 图像中异常信号区的全卷积网络模型训练方法 WO2019109410A1 (zh)

Applications Claiming Priority (2)

Application Number Priority Date Filing Date Title
CN201711275383.3 2017-12-06
CN201711275383.3A CN109886992A (zh) 2017-12-06 2017-12-06 用于分割mri图像中异常信号区的全卷积网络模型训练方法

Publications (1)

Publication Number Publication Date
WO2019109410A1 true WO2019109410A1 (zh) 2019-06-13

Family

ID=66750715

Family Applications (1)

Application Number Title Priority Date Filing Date
PCT/CN2017/118298 WO2019109410A1 (zh) 2017-12-06 2017-12-25 用于分割 mri 图像中异常信号区的全卷积网络模型训练方法

Country Status (2)

Country Link
CN (1) CN109886992A (zh)
WO (1) WO2019109410A1 (zh)

Cited By (1)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
ES2813777A1 (es) * 2019-09-23 2021-03-24 Quibim S L Metodo y sistema para la segmentacion automatica de hiperintensidades de sustancia blanca en imagenes de resonancia magnetica cerebral

Families Citing this family (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN111127395B (zh) * 2019-11-19 2023-04-07 中国人民解放军陆军军医大学第一附属医院 一种基于swi图像和循环神经网络的血管识别方法
CN111275689A (zh) * 2020-01-20 2020-06-12 平安科技（深圳）有限公司 医学影像识别检测方法、装置及计算机可读存储介质

Citations (5)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN103714536A (zh) * 2013-12-17 2014-04-09 深圳先进技术研究院 基于稀疏表示的多模态磁共振图像的分割方法及装置
CN105809175A (zh) * 2014-12-30 2016-07-27 深圳先进技术研究院 一种基于支持向量机算法的脑水肿分割方法及系统
CN106296699A (zh) * 2016-08-16 2017-01-04 电子科技大学 基于深度神经网络和多模态mri图像的脑肿瘤分割方法
CN107016681A (zh) * 2017-03-29 2017-08-04 浙江师范大学 基于全卷积网络的脑部mri肿瘤分割方法
CN107274402A (zh) * 2017-06-27 2017-10-20 北京深睿博联科技有限责任公司 一种基于胸部ct影像的肺结节自动检测方法及系统

Family Cites Families (13)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US20030157187A1 (en) * 1996-12-02 2003-08-21 Angiotech Pharmaceuticals, Inc. Compositions and methods for treating or preventing inflammatory diseases
JP5438029B2 (ja) * 2008-01-31 2014-03-12 コーニンクレッカ フィリップス エヌ ヴェ 短軸遅延強調心臓ｍｒｉの自動三次元セグメント化
US10762623B2 (en) * 2014-11-12 2020-09-01 Materialise N.V. System and method of improving surgical devices using captured images for efficient surgical plan development
CN105447458B (zh) * 2015-11-17 2018-02-27 深圳市商汤科技有限公司 一种大规模人群视频分析系统和方法
EP3380859A4 (en) * 2015-11-29 2019-07-31 Arterys Inc. AUTOMATED SEGMENTATION OF CARDIAC VOLUME
CN106127794A (zh) * 2016-07-29 2016-11-16 天津大学 基于可能性fcm算法mri肿瘤图像分割方法和系统
CN106920243B (zh) * 2017-03-09 2019-12-17 桂林电子科技大学 改进的全卷积神经网络的陶瓷材质件序列图像分割方法
CN107203989A (zh) * 2017-04-01 2017-09-26 南京邮电大学 基于全卷积神经网络的端对端胸部ct图像分割方法
CN107169421B (zh) * 2017-04-20 2020-04-28 华南理工大学 一种基于深度卷积神经网络的汽车驾驶场景目标检测方法
CN107239751B (zh) * 2017-05-22 2020-11-03 西安电子科技大学 基于非下采样轮廓波全卷积网络的高分辨sar图像分类方法
CN107239797A (zh) * 2017-05-23 2017-10-10 西安电子科技大学 基于全卷积神经网络的极化sar地物分类方法
CN107220980B (zh) * 2017-05-25 2019-12-03 重庆师范大学 一种基于全卷积网络的mri图像脑肿瘤自动分割方法
CN107229918B (zh) * 2017-05-26 2020-11-03 西安电子科技大学 一种基于全卷积神经网络的sar图像目标检测方法

Patent Citations (5)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN103714536A (zh) * 2013-12-17 2014-04-09 深圳先进技术研究院 基于稀疏表示的多模态磁共振图像的分割方法及装置
CN105809175A (zh) * 2014-12-30 2016-07-27 深圳先进技术研究院 一种基于支持向量机算法的脑水肿分割方法及系统
CN106296699A (zh) * 2016-08-16 2017-01-04 电子科技大学 基于深度神经网络和多模态mri图像的脑肿瘤分割方法
CN107016681A (zh) * 2017-03-29 2017-08-04 浙江师范大学 基于全卷积网络的脑部mri肿瘤分割方法
CN107274402A (zh) * 2017-06-27 2017-10-20 北京深睿博联科技有限责任公司 一种基于胸部ct影像的肺结节自动检测方法及系统

Cited By (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
ES2813777A1 (es) * 2019-09-23 2021-03-24 Quibim S L Metodo y sistema para la segmentacion automatica de hiperintensidades de sustancia blanca en imagenes de resonancia magnetica cerebral
WO2021058843A1 (es) * 2019-09-23 2021-04-01 Quibim, S.L. Método y sistema para la segmentación automática de hiperintensidades de sustancia blanca en imágenes de resonancia magnética cerebral

Also Published As

Publication number Publication date
CN109886992A (zh) 2019-06-14

Similar Documents

Publication Publication Date Title
CN107622240B (zh) 人脸检测方法和装置
CN110909611B (zh) 一种关注区域检测方法、装置、可读存储介质及终端设备
CN111080660B (zh) 一种图像分割方法、装置、终端设备及存储介质
WO2022001623A1 (zh) 基于人工智能的图像处理方法、装置、设备及存储介质
CN111161269B (zh) 图像分割方法、计算机设备和可读存储介质
CN108765447B (zh) 一种图像分割方法、图像分割装置及电子设备
WO2021120961A1 (zh) 大脑成瘾结构图谱评估方法及装置
CN111860398A (zh) 遥感图像目标检测方法、系统及终端设备
WO2019109410A1 (zh) 用于分割 mri 图像中异常信号区的全卷积网络模型训练方法
WO2021115061A1 (zh) 图像分割方法、装置及服务器
CN111210444A (zh) 多模态磁共振影像的分割方法、装置及介质
Sarica et al. A dense residual U-net for multiple sclerosis lesions segmentation from multi-sequence 3D MR images
CN111161348B (zh) 一种基于单目相机的物体位姿估计方法、装置及设备
US20230386067A1 (en) Systems and methods for segmenting 3d images
CN111281355B (zh) 一种用于确定脉搏采集位置的方法与设备
CN110288691B (zh) 渲染图像的方法、装置、电子设备和计算机可读存储介质
CN110210314B (zh) 人脸检测方法、装置、计算机设备及存储介质
CN115689947B (zh) 一种图像锐化的方法、系统、电子装置和存储介质
CN110827291A (zh) 一种用于自动化脑mri定量分析的方法和装置
CN113409324B (zh) 一种融合微分几何信息的脑分割方法
CN113781653B (zh) 对象模型生成方法、装置、电子设备及存储介质
CN115239655A (zh) 一种甲状腺超声影像肿瘤分割分类方法及装置
CN115147434A (zh) 图像处理方法、装置、终端设备及计算机可读存储介质
CN114821216A (zh) 图片去网纹神经网络模型建模、使用方法及相关设备
CN110689112A (zh) 数据处理的方法及装置

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: 17933947

Country of ref document: EP

Kind code of ref document: A1

NENP Non-entry into the national phase

Ref country code: DE

122 Ep: pct application non-entry in european phase

Ref document number: 17933947

Country of ref document: EP

Kind code of ref document: A1