WO2023202231A1 - 一种图像重建方法、装置、电子设备及存储介质 - Google Patents

一种图像重建方法、装置、电子设备及存储介质 Download PDF

Info

Publication number
WO2023202231A1
WO2023202231A1 PCT/CN2023/079245 CN2023079245W WO2023202231A1 WO 2023202231 A1 WO2023202231 A1 WO 2023202231A1 CN 2023079245 W CN2023079245 W CN 2023079245W WO 2023202231 A1 WO2023202231 A1 WO 2023202231A1
Authority
WO
WIPO (PCT)
Prior art keywords
data
training
neural network
deep neural
measured data
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/CN2023/079245
Other languages
English (en)
French (fr)
Inventor
管明涛
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.)
Beijing Huarui Boshi Medical Imaging Technology Co Ltd
Original Assignee
Beijing Huarui Boshi Medical Imaging Technology Co Ltd
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 Beijing Huarui Boshi Medical Imaging Technology Co Ltd filed Critical Beijing Huarui Boshi Medical Imaging Technology Co Ltd
Priority to CN202380012111.1A priority Critical patent/CN117425920A/zh
Priority to US18/853,527 priority patent/US20250225687A1/en
Priority to EP23790894.2A priority patent/EP4513436A4/en
Priority to JP2024549760A priority patent/JP2025513161A/ja
Publication of WO2023202231A1 publication Critical patent/WO2023202231A1/zh
Anticipated expiration legal-status Critical
Ceased legal-status Critical Current

Links

Classifications

    • GPHYSICS
    • G06COMPUTING OR CALCULATING; COUNTING
    • G06TIMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T11/00Two-dimensional [2D] image generation
    • GPHYSICS
    • G06COMPUTING OR CALCULATING; COUNTING
    • G06TIMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T12/00Tomographic reconstruction from projections
    • G06T12/20Inverse problem, i.e. transformations from projection space into object space
    • GPHYSICS
    • G06COMPUTING OR CALCULATING; COUNTING
    • G06TIMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T7/00Image analysis
    • G06T7/10Segmentation; Edge detection
    • GPHYSICS
    • G06COMPUTING OR CALCULATING; COUNTING
    • G06TIMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T2207/00Indexing scheme for image analysis or image enhancement
    • G06T2207/20Special algorithmic details
    • G06T2207/20081Training; Learning
    • GPHYSICS
    • G06COMPUTING OR CALCULATING; COUNTING
    • G06TIMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T2207/00Indexing scheme for image analysis or image enhancement
    • G06T2207/20Special algorithmic details
    • G06T2207/20084Artificial neural networks [ANN]
    • GPHYSICS
    • G06COMPUTING OR CALCULATING; COUNTING
    • G06TIMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T2211/00Image generation
    • G06T2211/40Computed tomography
    • G06T2211/441AI-based methods, deep learning or artificial neural networks

Definitions

  • the present disclosure belongs to the technical field of image reconstruction, and specifically relates to an image reconstruction method, device, electronic equipment and storage medium.
  • Electromagnetic data imaging is widely used in biomedical imaging, non-destructive testing and other fields. It collects electromagnetic field data of the object to be measured through sensors, and then reconstructs the conductivity or dielectric constant distribution inside the object through a certain image reconstruction method.
  • the present disclosure proposes an image reconstruction method, device, electronic equipment and storage medium.
  • This disclosure uses a variational autoencoder deep neural network to minimize the inversion objective function based on measured data to obtain the target latent space parameters of the inversion objective function based on the measured data, and then uses the variational autoencoder deep neural network to invert the target latent space parameters.
  • the spatial parameters are decoded to obtain the reconstructed image.
  • the present disclosure provides an image reconstruction method, including: obtaining measured data of a target; constructing a measured data inversion objective function using the hidden space parameters of a variational autoencoder deep neural network as unknowns according to the measured data. ; Using the variational autoencoder deep neural network to minimize the measured data inversion objective function to obtain target latent space parameters; and using the variational autoencoder deep neural network to minimize the target latent space parameters Decode to obtain the target reconstructed image.
  • the variational autoencoder deep neural network includes: a decoder and an encoder; the variational autoencoder deep neural network is used to minimize the measured data inversion objective function,
  • Obtaining target latent space parameters includes: setting an initial model according to the target; using the encoder to encode the initial model to obtain the encoding of the initial model; using the decoder to decode and calculate the encoding to obtain simulation data; determine said Whether the difference between the simulation data and the measured data is greater than a first threshold; if the difference is greater than the first threshold, determine the update amount of the encoding; update the encoding according to the update amount , and continue to execute the step "Use the decoder to decode and calculate the encoding to obtain simulation data"; when the difference is less than or equal to the first threshold, use the encoding as the target hidden value. Spatial parameter output.
  • the training of the variational autoencoder deep neural network includes: obtaining training data and constructing a training set based on the training data; constructing a variational autoencoder deep neural network based on the training set; The training set constructs a training function of the variational autoencoder deep neural network; and the training function is used to train the variational autoencoder deep neural network.
  • the training data includes: image data, and obtaining the training data and constructing a training set based on the training data includes: segmenting the image data to obtain the target of interest in the image data; Allocate training parameters to the target of interest to form an initial training model; adjust different directions of the initial training model to obtain multiple deformation training models to obtain a training set.
  • the training function includes: where Q is the length of the latent space variable, L is the number of pixels in the model, is the q-th component of the variance vector output by the encoder, is the q-th component of the mean vector output by the encoder, m l is the l-th component of the encoder input, is the l-th component of the decoder output, and ⁇ is the regularization coefficient that adjusts the KL divergence of the variational autoencoder.
  • the measured data includes one of the following: time series differential electrical impedance data, absolute electrical impedance data, and microwave data.
  • the present disclosure provides an image reconstruction device, including: a first acquisition module configured to acquire measured data of a target; a first building module configured to construct a variational autoencoder based on the measured data.
  • the hidden space parameter of the deep neural network is an unknown measured data inversion objective function;
  • the first execution module is configured to use the variational autoencoder deep neural network to minimize the measured data inversion objective function, and obtain Target latent space parameters; and
  • a second execution module configured to use the variational autoencoder deep neural network to decode the target latent space parameters to obtain a reconstructed image.
  • the present disclosure provides an electronic device including a storage and a processor.
  • the storage stores a computer program.
  • the processor executes the computer program, the image reconstruction method described in the first aspect is implemented.
  • the present disclosure provides a storage medium that stores a computer program, and the computer program can be executed by one or more processors to implement the image reconstruction method described in the first aspect.
  • This disclosure uses a variational autoencoder deep neural network to minimize the inversion objective function based on measured data to obtain the target latent space parameters of the inversion objective function based on the measured data, and then uses the variational autoencoder deep neural network to invert the target latent space parameters.
  • the spatial parameters are decoded to obtain the reconstructed image. This greatly reduces the number of unknowns in the image reconstruction process, and at the same time improves the computational efficiency in the image reconstruction process.
  • Figure 1 is an overall flow chart of an image reconstruction method provided by an embodiment of the present disclosure.
  • Figure 2 is a structural block diagram of an image reconstruction device provided by an embodiment of the present disclosure.
  • first ⁇ second ⁇ third If similar descriptions of “first ⁇ second ⁇ third” appear in this disclosure, the following description will be added. In the following description, the terms “first ⁇ second ⁇ third” involved are only to distinguish between similar The objects do not represent a specific ordering of the objects. It can be understood that the “first ⁇ second ⁇ third” can be interchanged in a specific order or sequence if permitted, so that the embodiments of the disclosure described here can be Implementation in sequences other than those illustrated or described herein.
  • the present disclosure provides an image reconstruction method, which is applied to electronic devices.
  • the electronic devices can be servers, mobile terminals, computers, cloud platforms, etc.
  • the functions implemented by the device data processing provided by the embodiments of the present disclosure can be implemented by calling the program code by the processor of the electronic device.
  • the program code can be stored in a computer storage medium.
  • the image reconstruction method includes steps S1 to S4.
  • Step S1 Obtain the measured data of the target.
  • the target measured data in this disclosure may be any one of time series differential electrical impedance data, absolute electrical impedance data, and microwave data.
  • Step S2 Based on the measured data, construct the measured data inversion objective function using the latent space parameters of the variational autoencoder deep neural network as unknown variables.
  • the variational autoencoder deep neural network includes an encoder and a decoder.
  • the encoder outputs latent space parameters
  • the decoder is responsible for decoding the latent space parameters output by the encoder.
  • Step S3 Use the variational autoencoder deep neural network to minimize the inversion objective function of the measured data to obtain the target latent space parameters.
  • step S3 using a variational autoencoder deep neural network to minimize the measured data inversion objective function, and obtaining the target latent space parameters includes the following steps S31 to step S37.
  • Step S31 Set the initial model according to the target.
  • the initial model When setting up the initial model, it needs to be set according to the type of measured data. Preset different parameters for the initial model.
  • the measured data is time series differential electrical impedance data or absolute electrical impedance data
  • the conductivity is preset for the initial model.
  • the measured parameters are microwave data
  • the dielectric constant is preset for the initial model.
  • Step S32 Use the encoder to encode the initial model to obtain the encoding of the initial model.
  • the encoder of the variational autoencoder deep neural network is used to encode the initial model with preset parameters to obtain the encoding of the initial model.
  • Step S33 Use the decoder to decode and calculate the code to obtain simulation data.
  • the encoding result is decoded using the decoder of the variational autoencoder deep neural network.
  • the preset parameter model will be obtained, and then the preset parameter model will be calculated to obtain simulation data.
  • m is the discrete conductivity change
  • d is the time series differential electrical impedance data calculated based on the conductivity equation simulation
  • A is the forward modeling function, expressed here as a matrix.
  • m is the discrete organ conductivity
  • d is the absolute electrical impedance data calculated according to Maxwell's equation simulation
  • T is the mapping function, which is used to map the organ conductivity to the triangular finite element grid to generate the chest conductivity model
  • F is the forward modeling function.
  • m is the discrete dielectric constant
  • d is the scattering field data calculated based on Maxwell's equation simulation
  • F is the forward modeling function
  • Step S34 Determine whether the difference between the simulation data and the actual measured data is greater than the first threshold.
  • the inversion problem can be equated to finding the optimal parameters v that minimize the following objective function:
  • Step S35 When the difference is greater than the first threshold, determine the coding update amount.
  • the coding needs to be updated.
  • the update amount of the coding needs to be determined.
  • J is the Jacobian matrix
  • J H is the conjugate transpose of J
  • I is the identity matrix
  • Step S36 Update the encoding according to the update amount, and use the continue execution step to use the decoder to decode and calculate the encoding to obtain simulation data.
  • step S33 is continued on the updated encoding.
  • Step S37 When the difference is less than or equal to the first threshold, output the encoding as the target latent space parameter.
  • the encoding corresponding to the simulation data is output as the target latent space parameter.
  • Step S4 Use the variational autoencoder deep neural network to decode the target latent space parameters to obtain the target reconstructed image.
  • the decoder of the variational autoencoder deep neural network is used to decode the target latent space parameters to obtain a preset parameter model, and finally obtain the target reconstructed image based on the preset parameter model.
  • the dielectric constant model/conductivity model can be obtained as:
  • This disclosure uses a variational autoencoder deep neural network to minimize the inversion objective function based on measured data to obtain the target latent space parameters of the inversion objective function based on the measured data, and then uses the variational autoencoder deep neural network to invert the target latent space parameters.
  • the spatial parameters are decoded to obtain the reconstructed image. This greatly reduces the number of unknowns in the image reconstruction process, and at the same time improves the computational efficiency in the image reconstruction process.
  • the training required when training a variational autoencoder deep neural network includes steps S51 to S54.
  • Step S51 Obtain training data and construct a training set based on the training data.
  • the training data comes from image data obtained by other imaging methods, such as CT scan data. Of course, it can also be three-dimensional image data obtained from three-dimensional scan data.
  • the training set is constructed based on the obtained image data.
  • step S51 obtaining training data and constructing a training set based on the training data includes the steps S511 to step S513.
  • Step S511 Segment the image data to obtain objects of interest in the image data.
  • Step S512 Assign training parameters to the target of interest to form an initial training model.
  • the assigned training parameter is the conductivity of the target, and an initial training model is formed at this time.
  • Step S513 Adjust different orientations of the initial training model to obtain multiple deformed training models to obtain a training set.
  • lung images obtained by other detection methods are first obtained. Since the attributes obtained by different detection methods are different, such as lungs obtained by CT methods. The image itself does not have conductivity data or microwave data. Therefore, it is necessary to preset the conductivity of the obtained image. After presetting the conductivity, an initial training model of electrical impedance of the lungs was formed. Then the formed initial training model is data enhanced, that is, the lung lobes of the lung images are randomly resected. Since the lung lobes are resected, the electrical impedance in the entire initial training model will change, thus forming multiple deformation training models. The multiple deformation training models finally obtain the training set.
  • Step S52 Construct a variational autoencoder deep neural network based on the training set.
  • the designed variational autoencoder deep neural network includes an encoder and a decoder.
  • the size of the encoder's input and output is 32 ⁇ 48 ⁇ 48.
  • convolution and activation are applied alternately three times to output a tensor of size 4 ⁇ 6 ⁇ 6 ⁇ 32.
  • two dense layers are applied to generate mean and variance vectors respectively. Sampling a Gaussian distribution using a reparameterization method.
  • the structure of the decoder is symmetrical to that of the encoder, except for upsampling via transposed convolutional layers.
  • the network uses a rectified linear function (ReLU) for nonlinear activation.
  • Step S53 Construct the training function of the variational autoencoder deep neural network based on the training set.
  • the training function includes:
  • Q is the length of the latent space variable
  • L is the number of pixels in the model
  • q-th component of the variance vector output by the encoder is the q-th component of the mean vector output by the encoder
  • m l is the l-th component of the encoder input
  • is the regularization coefficient that adjusts the KL divergence of the variational autoencoder.
  • Step S54 Use the training function to train the variational autoencoder deep neural network.
  • the Adam optimization algorithm and training function are used to train the designed variational autoencoder deep neural network. practice. After the training is completed, the optimal variational autoencoder deep neural network parameters can be obtained.
  • This disclosure uses a variational autoencoder deep neural network to minimize the inversion objective function based on measured data to obtain the target latent space parameters of the inversion objective function based on the measured data, and then uses the variational autoencoder deep neural network to invert the target latent space parameters.
  • the spatial parameters are decoded to obtain the reconstructed image. This greatly reduces the number of unknowns in the image reconstruction process, and at the same time improves the computational efficiency in the image reconstruction process.
  • embodiments of the present disclosure provide an image reconstruction device.
  • Each module included in the device and each unit included in each module can be implemented by a processor in a computer device; of course, it can also be implemented by a logic circuit. ;
  • the processor can be a central processing unit (CPU, Central Processing Unit), a microprocessor (MPU, Microprocessor Unit), a digital signal processor (DSP, Digital Signal Processing) or a field programmable gate array ( FPGA, Field Programmable Gate Array), etc.
  • CPU Central Processing Unit
  • MPU Microprocessor Unit
  • DSP Digital Signal Processing
  • FPGA Field Programmable Gate Array
  • the second aspect provides an image reconstruction device, including: a first acquisition module 1 , a first construction module 2 , a first execution module 3 and a second execution module 4 .
  • the first acquisition module 1 is configured to acquire actual measured data of the target.
  • the first building module 2 is configured to construct a measured data inversion objective function using the latent space parameters of the variational autoencoder deep neural network as unknowns based on the measured data.
  • the first execution module 3 is configured to use a variational autoencoder deep neural network to minimize the measured data inversion objective function to obtain the target latent space parameters.
  • the second execution module 4 is configured to use a variational autoencoder deep neural network to decode the target latent space parameters to obtain a reconstructed image.
  • the first execution module 3 includes: a first setting module, a third execution module, a fourth execution module, a first determination module, a second determination module, and a first output module.
  • the first setting module is configured to set the initial model according to the target.
  • the third execution module is configured to use the encoder to encode the initial model to obtain the encoding of the initial model.
  • the fourth execution module is configured to use the decoder to decode and calculate the code to obtain simulation data.
  • the first determination module is configured to determine whether the difference between the simulated data and the measured data is greater than a first threshold.
  • the second determination module is configured to determine an update amount of the encoding when the difference is greater than the first threshold.
  • the first output module is configured to output the encoding as the target latent space parameter when the difference is less than or equal to the first threshold.
  • the image reconstruction device further includes: a second acquisition module, a fifth execution module, a sixth execution module and a seventh execution module.
  • the second acquisition module is configured to acquire training data and construct a training set based on the training data.
  • the fifth execution module is configured to build a variational autoencoder deep neural network based on the training set.
  • the sixth execution module is configured to construct a training function of the variational autoencoder deep neural network based on the training set.
  • the seventh execution module is configured to train the variational autoencoder deep neural network using the training function.
  • the fifth execution module includes: an eighth execution module, a ninth execution module, and a tenth execution module.
  • the eighth execution module is configured to segment the image data and obtain the target of interest in the image data.
  • the ninth execution module is configured to assign training parameters to the target of interest to form an initial training model.
  • the tenth execution module adjusts different orientations of the initial training model to obtain multiple deformed training models to obtain a training set.
  • Each module in the above-mentioned image reconstruction device can be implemented in whole or in part by software, hardware, and combinations thereof.
  • Each of the above modules may be embedded in or independent of the processor in the device in the form of hardware, or may be stored in the memory of the processing device in the form of software, so that the processor can call and execute the operations corresponding to each of the above modules.
  • the division of modules in the embodiment of the present disclosure is schematic and is only a logical function division. In actual implementation, there may be other division methods.
  • the third aspect provides an electronic device, including a storage and a processor.
  • the storage stores a computer program.
  • the processor executes the computer program, it implements the steps of an image reconstruction method.
  • the fourth aspect provides a storage medium.
  • the computer program stored in the storage medium can be executed by one or more processors.
  • the computer program can be used to implement the steps of any image reconstruction method in the first aspect.
  • Non-volatile memory can include read-only memory (ROM), magnetic tape, floppy disk, flash memory or optical memory, etc.
  • Volatile memory may include random access memory (RAM) or external cache memory.
  • RAM can be in many forms, such as static random access memory (Static Random Access Memory, SRAM) or dynamic random access memory (Dynamic Random Access Memory, DRAM).
  • the disclosed devices and methods can be implemented in other ways.
  • the device embodiments described above are only illustrative.
  • the division of units is only a logical function division.
  • the coupling, direct coupling, or communication connection between the components shown or discussed may be through some interfaces, and the indirect coupling or communication connection of the devices or units may be electrical, mechanical, or other forms. of.
  • the units described above as separate components may or may not be physically separated; the components shown as units may or may not be physical units; they may be located in one place or distributed to multiple network units; Some or all of the units can be selected according to actual needs to achieve the purpose of the solution of this embodiment.
  • each functional unit in each embodiment of the present disclosure can be all integrated into one processing unit, or each unit can be separately used as a unit, or two or more units can be integrated into one unit; the above-mentioned integration
  • the unit can be implemented in the form of hardware or in the form of hardware plus software functional units.
  • the aforementioned program can be stored in a computer-readable storage medium.
  • the execution includes: The steps of the above method embodiment; and the aforementioned storage media include: mobile storage devices, read-only memory (ROM, Read Only Memory), magnetic disks or optical disks and other various media that can store program codes.
  • the above-mentioned integrated units of the present disclosure are implemented in the form of software function modules and sold or used as independent products, they can also be stored in a computer-readable storage medium.
  • the technical solutions of the embodiments of the present disclosure are essentially or the parts that contribute to related technologies can be embodied in the form of software products.
  • the computer software products are stored in a storage medium and include a number of instructions to enable A controller performs all or part of the methods described in various embodiments of the present disclosure.
  • the aforementioned storage media include: mobile storage devices, ROMs, magnetic disks or optical disks and other media that can store program codes.

Landscapes

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

Abstract

本公开属于图像重建的技术领域,涉及一种图像重建方法、装置、电子设备及存储介质。本公开通过变分自编码器深度神经网络对实测数据反演目标函数进行最小化,以得到实测数据反演目标函数的目标隐空间参数,再利用变分自编码器深度神经网络对目标隐空间参数进行解码,得到重建图像。该图像重建方法,包括:获取目标的实测数据;根据实测数据构建以变分自编码器深度神经网络的隐空间参数为未知数的实测数据反演目标函数;利用变分自编码器深度神经网络对实测数据反演目标函数进行最小化,得到目标隐空间参数;利用变分自编码器深度神经网络对目标隐空间参数进行解码,得到目标重建图像。

Description

一种图像重建方法、装置、电子设备及存储介质
相关申请的交叉引用
本公开要求享有2022年4月20日提交的名称为“一种图像重建方法、装置、电子设备及存储介质”的中国专利申请CN202210417793.1的优先权,其全部内容通过引用并入本公开中。
技术领域
本公开属于图像重建技术领域,具体涉及一种图像重建方法、装置、电子设备及存储介质。
背景技术
电磁数据成像广泛应用于生物医学成像、无损探伤等领域,其通过传感器采集待测物体的电磁场数据,然后通过一定的图像重建方法重构物体内部的电导率或介电常数分布。
在图像重建方面,常见的方法将反演域分解为像素,然后通过最小化仿真数据和测量数据的残差重建离散的电导率或介电常数。像素的数量通常远大于数据的数量,因此这是一个非线性的病态问题,需要依赖先验知识进行合理的重建。
发明内容
本公开提出了一种图像重建方法、装置、电子设备及存储介质。本公开通过变分自编码器深度神经网络对根据实测数据反演目标函数进行最小化,以得到实测数据反演目标函数的目标隐空间参数,再利用变分自编码器深度神经网络对目标隐空间参数进行解码,得到重建图像。
第一方面,本公开提供了一种图像重建方法,包括:获取目标的实测数据;根据所述实测数据构建以变分自编码器深度神经网络的隐空间参数为未知数的实测数据反演目标函数;利用所述变分自编码器深度神经网络对所述实测数据反演目标函数进行最小化,得到目标隐空间参数;以及利用所述变分自编码器深度神经网络对所述目标隐空间参数进行解码,得到目标重建图像。
在一些实施例中,所述变分自编码器深度神经网络包括:解码器和编码器;所述利用所述变分自编码器深度神经网络对所述实测数据反演目标函数进行最小化,得到目标隐空间参数,包括:根据目标设定初始模型;利用所述编码器对所述初始模型进行编码,得到初始模型的编码;利用所述解码器对所述编码进行解码并计算,得到仿真数据;确定所述 仿真数据与所述实测数据之间的差值是否大于第一阈值;在所述差值大于第一阈值的情况下,确定所述编码的更新量;根据所述更新量对所述编码进行更新,并利用继续执行步骤“利用所述解码器对所述编码进行解码并计算,得到仿真数据”;在所述差值小于或等于第一阈值的情况下,将所述编码作为所述目标隐空间参数输出。
在一些实施例中,通过下式确定所述更新量:H(vk)·pk=-g(vk);其中,J为雅克比矩阵,JH为J的共轭转置;I为单位矩阵。
在一些实施例中,所述变分自编码器深度神经网络的训练包括:获取训练数据,并根据所述训练数据构建训练集;根据所述训练集构建变分自编码器深度神经网络;根据所述训练集构建所述变分自编码器深度神经网络的训练函数;以及利用所述训练函数对所述变分自编码器深度神经网络进行训练。
在一些实施例中,所述训练数据包括:图像数据,所述获取训练数据,并根据所述训练数据构建训练集,包括:对所述图像数据进行分割,获得图像数据中的感兴趣目标;对所述感兴趣目标分配训练参数,形成初始训练模型;对所述初始训练模型的不同方位分别进行调整,获得多个变形训练模型,以得到训练集。
在一些实施例中,所述训练函数包括: 其中Q为隐空间变量的长度,L为模型中像素点的个数,为编码器输出的方差向量的第q个分量,为编码器输出的均值向量的第q个分量,ml为编码器输入的第l个分量,为解码器输出的第l个分量,α为调整变分自编码器KL散度的正则化系数。
在一些实施例中,所述实测数据包括以下之一:时序差分电阻抗数据、绝对电阻抗数据和微波数据。
第二方面,本公开提供了一种图像重建装置,包括:第一获取模块,被配置为获取目标的实测数据;第一构建模块,被配置为根据所述实测数据构建以变分自编码器深度神经网络的隐空间参数为未知数的实测数据反演目标函数;第一执行模块,被配置为利用所述变分自编码器深度神经网络对所述实测数据反演目标函数进行最小化,得到目标隐空间参数;以及第二执行模块,被配置为利用所述变分自编码器深度神经网络对所述目标隐空间参数进行解码,得到重建图像。
第三方面,本公开提供了一种电子设备,包括储存器和处理器,所述储存器存储有计算机程序,所述处理器执行所述计算机程序时实现第一方面所述的图像重建方法。
第四方面,本公开提供了一种存储介质,该存储介质存储有计算机程序,所述计算机程序能够被一个或多个处理器执行,以实现第一方面所述的图像重建方法。
本公开通过变分自编码器深度神经网络对根据实测数据反演目标函数进行最小化,以得到实测数据反演目标函数的目标隐空间参数,再利用变分自编码器深度神经网络对目标隐空间参数进行解码,得到重建图像。使得在图像重建过程中的未知数大大减少,同时提高了图像重建过程中的计算效率。
附图说明
通过结合附图阅读下文示例性实施例的详细描述可更好地理解本公开的范围。其中所包括的附图是:
图1为本公开实施例提供的一种图像重建方法的整体流程图;以及
图2为本公开实施例提供的一种图像重建装置的结构框图。
具体实施方式
为了使本公开的目的、技术方案和优点更加清楚,下面将结合附图对本公开作进一步地详细描述,所描述的实施例不应视为对本公开的限制,本领域普通技术人员在没有做出创造性劳动前提下所获得的所有其它实施例,都属于本公开保护的范围。
在以下的描述中,涉及到“一些实施例”,其描述了所有可能实施例的子集,但是可以理解,“一些实施例”可以是所有可能实施例的相同子集或不同子集,并且可以在不冲突的情况下相互结合。
如果本公开中出现的“第一\第二\第三”的类似描述则增加以下的说明,在以下的描述中,所涉及的术语“第一\第二\第三”仅仅是是区别类似的对象,不代表针对对象的特定排序,可以理解,“第一\第二\第三”在允许的情况下可以互换特定的顺序或先后次序,以使这里描述的本公开实施例能够以除了在这里图示或描述的以外的顺序实施。
除非另有定义,本文所使用的所有的技术和科学术语与属于本公开的技术领域的技术人员通常理解的含义相同。本文中所使用的术语只是为了描述本公开实施例的目的,不是旨在限制本公开。
在图像重建方面,常见的方法所需像素的数量通常远大于数据的数量,因此这是一个非线性的病态问题,需要依赖先验知识进行合理的重建。由于未知数的数量通常为数千或数万,目标函数极小化的过程计算量巨大。此外,反演时利用先验信息的方式不够灵活。解释人员头脑中的先验知识难以用数学形式描述,因此无法约束反演过程,给模型重建和解释带来了挑战。
实施例1
如图1所示,本公开提供了一种图像重建方法,该方法应用于电子设备,电子设备可以服务器、移动终端、计算机、云平台等。本公开实施例提供的设备数据处理所实现的功能可以通过电子设备的处理器调用程序代码来实现,其中,程序代码可以保存在计算机存储介质中,图像重建方法包括步骤S1至步骤S4。
步骤S1:获取目标的实测数据。
本公开中的目标实测数据可以为时序差分电阻抗数据、绝对电阻抗数据和微波数据中的任意一种。
步骤S2:根据实测数据构建以变分自编码器深度神经网络的隐空间参数为未知数的实测数据反演目标函数。
变分自编码器深度神经网络包括有编码器和解码器,其中编码器输出的为隐空间参数,而解码器负责对编码器输出的隐空间参数进行解码。
步骤S3:利用变分自编码器深度神经网络对实测数据反演目标函数进行最小化,得到目标隐空间参数。
在一些实施例中,在步骤S3中,利用变分自编码器深度神经网络对实测数据反演目标函数进行最小化,得到目标隐空间参数包括以下步骤S31至步骤S37。
步骤S31:根据目标设定初始模型。
设置初始模型时需要根据实测数据的种类进行设置。为初始模型预设不同的参数。当实测数据为时序差分电阻抗数据或绝对电阻抗数据时,为初始模型预设电导率。当实测参数为微波数据时,为初始模型预设介电常数。
步骤S32:利用编码器对初始模型进行编码,得到初始模型的编码。
利用变分自编码器深度神经网络的编码器对预设了参数的初始模型进行编码,得到初始模型的编码。
步骤S33:利用解码器对编码进行解码并计算,得到仿真数据。
完成编码后,将编码结果利用变分自编码器深度神经网络的解码器进行解码,解码后会得到预设参数模型,再对预设参数模型进行计算得到仿真数据。
(1)在实测数据为时序差分电阻抗数据时,时序差分电阻抗数据通过下式仿真:
d=A·m,
其中m为离散的电导率变化量,d为根据电导率方程仿真计算出的时序差分电阻抗数据,A为正向建模函数,此处表示为一个矩阵。
离散电导率变化量m可由变分自编码器的解码器D对隐空间变量解码:
m=D(v);
因此,从隐空间变量计算电阻抗数据的过程可用S表示:
d=A·D(v)=S(v)。
(2)在实测数据为绝对电阻抗数据时,绝对电阻抗数据通过下式数值仿真:
d=F(T(m))
其中m为离散的器官电导率,d为根据麦克斯韦方程仿真计算出的绝对电阻抗数据,T为映射函数,用于将器官电导率映射到三角有限元网格上,生成胸腔电导率模型,F为正向建模函数。
器官电导率m可由变分自编码器的解码器D对隐空间变量v解码得到:
m=D(v)
因此,从隐空间变量v计算电压仿真数据的过程可用S表示:
d=F(T(D(v)))=S(v)。
(3)在微波数据反演中,电磁散射数据通过下式数值仿真:
d=F(m);
其中m为离散的介电常数,d为根据麦克斯韦方程仿真计算出的散射场数据,F为正向建模函数。
离散介电常数m可由变分自编码器的解码器D对隐空间变量解码:
m=D(v);
因此,从隐空间变量计算电磁数据的过程可用S表示:
d=F(D(v))=S(v)。
步骤S34:确定仿真数据与实测数据之间的差值是否大于第一阈值。
通过对比仿真数据和实测数据之间的差值确定最终结果是否收敛于第一阈值。
在确定性反演中,反演问题可等同于寻找使以下目标函数最小化的最优参数v:
其中||||表示L2范数,dobs为测量数据,λ为目标函数的正则化系数,用于稳定函数优化过程。
步骤S35:当差值大于第一阈值时,确定编码的更新量。
当结果不能收敛于第一阈值的时候,则需要对编码进行更新,在对编码进行跟新时,需要确定编码的更新量。
确定编码的更新量时,需要通过下式确定更新量:
H(vk)·pk=-g(vk)。
其中,

J为雅克比矩阵,JH为J的共轭转置。I为单位矩阵。
步骤S36:根据更新量对编码进行更新,并利用继续执行步骤利用解码器对编码进行解码并计算,得到仿真数据。
在确定了更新量之后,根据更新量对编码进行更新,并继续对更新后的编码执行步骤S33。
在确定了更新量后,根据下式对编码进行更新,
vk+1=vk+pk
步骤S37:当差值小于或等于第一阈值时,将编码作为目标隐空间参数输出。
当仿真数据和实测数据之间的差值确定最终结果收敛于第一阈值时,将与仿真数据相对应的编码作为目标隐空间参数输出。
步骤S4:利用变分自编码器深度神经网络对目标隐空间参数进行解码,得到目标重建图像。
得到目标隐空间参数后,利用变分自编码器深度神经网络的解码器对目标隐空间参数进行解码,得到预设参数模型,最终根据预设参数模型得到目标重建图像。
在得到vk+1后,用变分自编码器中的解码器D对其解码,可得介电常数模型/电导率模型为:
mk+1=D(vk+1)。
本公开通过变分自编码器深度神经网络对根据实测数据反演目标函数进行最小化,以得到实测数据反演目标函数的目标隐空间参数,再利用变分自编码器深度神经网络对目标隐空间参数进行解码,得到重建图像。使得在图像重建过程中的未知数大大减少,同时提高了图像重建过程中的计算效率。
在一些实施例中,在训练变分自编码器深度神经网络时所需要的训练包括步骤S51至步骤S54。
步骤S51:获取训练数据,并根据训练数据构建训练集。
训练数据来源于其他成像手段所得到的图像数据,比如CT扫描数据,当然也可以是三维扫描数据得到三维图像数据,根据得到的图像数据来构建训练集。
在一些实施例中,在步骤S51中,获取训练数据并根据训练数据构建训练集包括步骤 S511至步骤S513。
步骤S511:对图像数据进行分割,获得图像数据中的感兴趣目标。
步骤S512:对感兴趣目标分配训练参数,形成初始训练模型。
比如在进行电阻抗图像重建时,分配的训练参数便为目标的电导率,此时便形成初始训练模型。
步骤S513:对初始训练模型的不同方位分别进行调整,获得多个变形训练模型,以得到训练集。
为了增加训练模型数据量,可以通过修改目标的形态来增加训练模型。
比如:通过肺部的图像对变分自编码器深度神经网络进行训练时,先获得由其他探测手段获得的肺部图像,由于不同的探测手段所获得属性不相同,比如通过CT手段获得的肺部图像本身不具备电导率数据以及微波数据。所以需要对获得到的图像进行电导率预设。预设电导率后形成了关于肺部的电阻抗初始训练模型。然后再对形成的初始训练模型进行数据增强,即对肺部图像的肺叶进行随机切除,由于肺叶被切除,会使得整个初始训练模型中的电阻抗发生改变,进而形成了多个变形训练模型,而多个变形训练模型最终得到训练集。
步骤S52:根据训练集构建变分自编码器深度神经网络。
本实施例(时序差分电阻抗成像例)中,设计出来的变分自编码器深度神经网络包括有编码器和解码器。其中编码器的输入和输出的大小为32×48×48。在编码器中,卷积、激活交替应用三次以输出尺寸为4×6×6×32的张量。将张量重排列为向量后,分别应用两个密集层生成均值和方差向量。使用重新参数化方法对高斯分布进行采样。
本例(时序差分电阻抗成像例)中,隐空间变量的长度为32,实现了32/73728=0.043%的压缩率。除通过转置卷积层实现上采样外,解码器的结构与编码器的结构对称。网络使用线性整流函数(ReLU)进行非线性激活。
步骤S53:根据训练集构建变分自编码器深度神经网络的训练函数。
在一些实施例中,训练函数包括:
其中Q为隐空间变量的长度,L为模型中像素点的个数,为编码器输出的方差向量的第q个分量,为编码器输出的均值向量的第q个分量,ml为编码器输入的第l个分量,为解码器输出的第l个分量,α为调整变分自编码器KL散度的正则化系数。
步骤S54:利用训练函数对变分自编码器深度神经网络进行训练。
最后,利用Adam优化算法和训练函数对设计的变分自编码器深度神经网络进行训 练。训练完成后,可得到最优的变分自编码器深度神经网络参数。
本公开通过变分自编码器深度神经网络对根据实测数据反演目标函数进行最小化,以得到实测数据反演目标函数的目标隐空间参数,再利用变分自编码器深度神经网络对目标隐空间参数进行解码,得到重建图像。使得在图像重建过程中的未知数大大减少,同时提高了图像重建过程中的计算效率。
实施例2
基于前述的实施例,本公开实施例提供一种图像重建装置,该装置包括的各模块、以及各模块包括的各单元,可以通过计算机设备中的处理器来实现;当然也可通过逻辑电路实现;在实施的过程中,处理器可以为中央处理器(CPU,Central Processing Unit)、微处理器(MPU,Microprocessor Unit)、数字信号处理器(DSP,Digital Signal Processing)或现场可编程门阵列(FPGA,Field Programmable Gate Array)等。
如图5所示,第二方面提供了一种图像重建装置,包括:第一获取模块1、第一构建模块2、第一执行模块3和第二执行模块4。
第一获取模块1被配置为获取目标的实测数据。第一构建模块2被配置为根据实测数据构建以变分自编码器深度神经网络的隐空间参数为未知数的实测数据反演目标函数。第一执行模块3被配置为利用变分自编码器深度神经网络对实测数据反演目标函数进行最小化,得到目标隐空间参数。第二执行模块4被配置为利用变分自编码器深度神经网络对目标隐空间参数进行解码,得到重建图像。
在一些实施例中,第一执行模块3包括:第一设定模块、第三执行模块、第四执行模块、第一确定模块、第二确定模块、第一输出模块。
第一设定模块被配置为根据目标设定初始模型。第三执行模块被配置为利用编码器对初始模型进行编码,得到初始模型的编码。第四执行模块被配置为利用解码器对编码进行解码并计算,得到仿真数据。第一确定模块被配置为确定仿真数据与实测数据之间的差值是否大于第一阈值。第二确定模块被配置为当差值大于第一阈值时,确定编码的更新量。第一输出模块被配置为当差值小于或等于第一阈值时,将编码作为目标隐空间参数输出。
在一些实施例中,图像重建装置还包括:第二获取模块、第五执行模块、第六执行模块和第七执行模块。
第二获取模块被配置为获取训练数据,并根据训练数据构建训练集。第五执行模块被配置为根据训练集构建变分自编码器深度神经网络。第六执行模块被配置为根据训练集构建变分自编码器深度神经网络的训练函数。第七执行模块被配置为利用训练函数对变分自编码器深度神经网络进行训练。
在一些实施例中,第五执行模块包括:第八执行模块、第九执行模块和第十执行模块。
第八执行模块被配置为对图像数据进行分割,获得图像数据中的感兴趣目标。第九执行模块被配置为对感兴趣目标分配训练参数,形成初始训练模型。第十执行模块对初始训练模型的不同方位分别进行调整,获得多个变形训练模型,以得到训练集。
上述一种图像重建装置中的各个模块可全部或部分通过软件、硬件及其组合来实现。上述各模块可以硬件形式内嵌于或独立于设备中的处理器中,也可以以软件形式存储于处理装置中的存储器中,以便于处理器调用执行以上各个模块对应的操作。需要说明的是,本公开实施例中对模块的划分是示意性的,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。
实施例3
第三方面提供了一种电子设备,包括储存器和处理器,储存器存储有计算机程序,处理器执行计算机程序时实现一种图像重建方法的步骤。
实施例4
第四方面提供了一种存储介质,该存储介质存储的计算机程序,能够被一个或多个处理器执行,计算机程序能够用来实现第一方面中任一项图像重建方法的步骤。
本领域普通技术人员可以理解实现上述实施例方法中的全部或部分流程,是可以通过计算机程序来指令相关的硬件来完成,计算机程序可存储于一非易失性计算机可读取存储介质中,该计算机程序在执行时,可包括如上述各方法的实施例的流程。其中,本公开所提供的各实施例中所使用的对存储器、存储、数据库或其它介质的任何引用,均可包括非易失性和易失性存储器中的至少一种。非易失性存储器可包括只读存储器(Read-Only Memory,ROM)、磁带、软盘、闪存或光存储器等。易失性存储器可包括随机存取存储器(Random Access Memory,RAM)或外部高速缓冲存储器。作为说明而非局限,RAM可以是多种形式,比如静态随机存取存储器(Static Random Access Memory,SRAM)或动态随机存取存储器(Dynamic Random Access Memory,DRAM)等。
应理解,说明书通篇中提到的“一个实施例”或“一些实施例”意味着与实施例有关的特定特征、结构或特性包括在本公开的至少一个实施例中。因此,在整个说明书各处出现的“在一个实施例中”或“在一实施例中”未必一定指相同的实施例。此外,这些特定的特征、结构或特性可以任意适合的方式结合在一个或多个实施例中。应理解,在本公开的各种实施例中,上述各过程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,而不应对本公开实施例的实施过程构成任何限定。上述本公开实施例序号仅仅为了描述,不代表实施例的优劣。
需要说明的是,在本文中,术语“包括”、“包含”或者其任何其他变体意在涵盖非排他性的包含,从而使得包括一系列要素的过程、方法、物品或者装置不仅包括那些要素, 而且还包括没有明确列出的其他要素,或者是还包括为这种过程、方法、物品或者装置所固有的要素。在没有更多限制的情况下,由语句“包括一个……”限定的要素,并不排除在包括该要素的过程、方法、物品或者装置中还存在另外的相同要素。
在本公开所提供的几个实施例中,应该理解到,所揭露的设备和方法,可以通过其它的方式实现。以上所描述的设备实施例仅仅是示意性的,例如,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,如:多个单元或组件可以结合,或可以集成到另一个系统,或一些特征可以忽略,或不执行。另外,所显示或讨论的各组成部分相互之间的耦合、或直接耦合、或通信连接可以是通过一些接口,设备或单元的间接耦合或通信连接,可以是电性的、机械的或其它形式的。
上述作为分离部件说明的单元可以是、或也可以不是物理上分开的,作为单元显示的部件可以是、或也可以不是物理单元;既可以位于一个地方,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全部单元来实现本实施例方案的目的。
另外,在本公开各实施例中的各功能单元可以全部集成在一个处理单元中,也可以是各单元分别单独作为一个单元,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,也可以采用硬件加软件功能单元的形式实现。
本领域普通技术人员可以理解:实现上述方法实施例的全部或部分步骤可以通过程序指令相关的硬件来完成,前述的程序可以存储于计算机可读取存储介质中,该程序在执行时,执行包括上述方法实施例的步骤;而前述的存储介质包括:移动存储设备、只读存储器(ROM,Read Only Memory)、磁碟或者光盘等各种可以存储程序代码的介质。
或者,本公开上述集成的单元如果以软件功能模块的形式实现并作为独立的产品销售或使用时,也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,本公开实施例的技术方案本质上或者说对相关技术做出贡献的部分可以以软件产品的形式体现出来,该计算机软件产品存储在一个存储介质中,包括若干指令用以使得一台控制器执行本公开各个实施例所述方法的全部或部分。而前述的存储介质包括:移动存储设备、ROM、磁碟或者光盘等各种可以存储程序代码的介质。
以上所述,仅为本公开的实施方式,但本公开的保护范围并不局限于此,任何熟悉本技术领域的技术人员在本公开揭露的技术范围内,可轻易想到变化或替换,都应涵盖在本公开的保护范围之内。因此,本公开的保护范围应以所述权利要求的保护范围为准。

Claims (10)

  1. 一种图像重建方法,包括:
    获取目标的实测数据;
    根据所述实测数据构建以变分自编码器深度神经网络的隐空间参数为未知数的实测数据反演目标函数;
    利用所述变分自编码器深度神经网络对所述实测数据反演目标函数进行最小化,得到目标隐空间参数;以及
    利用所述变分自编码器深度神经网络对所述目标隐空间参数进行解码,得到目标重建图像。
  2. 根据权利要求1所述的图像重建方法,其中,所述变分自编码器深度神经网络包括:解码器和编码器;所述利用所述变分自编码器深度神经网络对所述实测数据反演目标函数进行最小化,得到目标隐空间参数,包括:
    根据目标设定初始模型;
    利用所述编码器对所述初始模型进行编码,得到初始模型的编码;
    利用所述解码器对所述编码进行解码并计算,得到仿真数据;
    确定所述仿真数据与所述实测数据之间的差值是否大于第一阈值;
    在所述差值大于第一阈值的情况下,确定所述编码的更新量;
    根据所述更新量对所述编码进行更新,并利用继续执行步骤“利用所述解码器对所述编码进行解码并计算,得到仿真数据”;以及
    在所述差值小于或等于第一阈值的情况下,将所述编码作为所述目标隐空间参数输出。
  3. 根据权利要求2所述的图像重建方法,其中,通过下式确定所述更新量:
    H(vk)·pk=-g(vk);
    其中, J为雅克比矩阵,JH为J的共轭转置;I为单位矩阵。
  4. 根据权利要求1所述的图像重建方法,其中,所述变分自编码器深度神经网络的训练包括:
    获取训练数据,并根据所述训练数据构建训练集;
    根据所述训练集构建变分自编码器深度神经网络;
    根据所述训练集构建所述变分自编码器深度神经网络的训练函数;以及
    利用所述训练函数对所述变分自编码器深度神经网络进行训练。
  5. 根据权利要求4所述的图像重建方法,其中,所述训练数据包括:图像数据,所述 获取训练数据,并根据所述训练数据构建训练集,包括:
    对所述图像数据进行分割,获得图像数据中的感兴趣目标;
    对所述感兴趣目标分配训练参数,形成初始训练模型;以及
    对所述初始训练模型的不同方位分别进行调整,获得多个变形训练模型,以得到训练集。
  6. 根据权利要求4所述的图像重建方法,其中,所述训练函数包括:
    其中,Q为隐空间变量的长度,L为模型中像素点的个数,为编码器输出的方差向量的第q个分量,为编码器输出的均值向量的第q个分量,ml为编码器输入的第l个分量,为解码器输出的第l个分量,α为调整变分自编码器KL散度的正则化系数。
  7. 根据权利要求1所述的图像重建方法,其中,所述实测数据包括以下之一:时序差分电阻抗数据、绝对电阻抗数据和微波数据。
  8. 一种图像重建装置,包括:
    第一获取模块,被配置为获取目标的实测数据;
    第一构建模块,被配置为根据所述实测数据构建以变分自编码器深度神经网络的隐空间参数为未知数的实测数据反演目标函数;
    第一执行模块,被配置为利用所述变分自编码器深度神经网络对所述实测数据反演目标函数进行最小化,得到目标隐空间参数;以及
    第二执行模块,被配置为利用所述变分自编码器深度神经网络对所述目标隐空间参数进行解码,得到重建图像。
  9. 一种电子设备,包括:
    存储器和处理器,所述存储器上存储有计算机程序,该计算机程序被所述处理器执行时,执行如权利要求1至7任意一项所述的图像重建方法。
  10. 一种存储介质,所述存储介质存储有计算机程序,所述计算机程序能够被一个或多个处理器执行,以实现如权利要求1至7中任一项所述的图像重建方法。
PCT/CN2023/079245 2022-04-20 2023-03-02 一种图像重建方法、装置、电子设备及存储介质 Ceased WO2023202231A1 (zh)

Priority Applications (4)

Application Number Priority Date Filing Date Title
CN202380012111.1A CN117425920A (zh) 2022-04-20 2023-03-02 一种图像重建方法、装置、电子设备及存储介质
US18/853,527 US20250225687A1 (en) 2022-04-20 2023-03-02 Image reconstruction method and apparatus, and electronic device and storage medium
EP23790894.2A EP4513436A4 (en) 2022-04-20 2023-03-02 METHOD AND APPARATUS FOR IMAGE RECONSTRUCTION, AND ELECTRONIC DEVICE AND RECORDING MEDIUM
JP2024549760A JP2025513161A (ja) 2022-04-20 2023-03-02 画像再構成方法、装置、電子機器および記憶媒体

Applications Claiming Priority (2)

Application Number Priority Date Filing Date Title
CN202210417793.1 2022-04-20
CN202210417793 2022-04-20

Publications (1)

Publication Number Publication Date
WO2023202231A1 true WO2023202231A1 (zh) 2023-10-26

Family

ID=88419042

Family Applications (1)

Application Number Title Priority Date Filing Date
PCT/CN2023/079245 Ceased WO2023202231A1 (zh) 2022-04-20 2023-03-02 一种图像重建方法、装置、电子设备及存储介质

Country Status (5)

Country Link
US (1) US20250225687A1 (zh)
EP (1) EP4513436A4 (zh)
JP (1) JP2025513161A (zh)
CN (1) CN117425920A (zh)
WO (1) WO2023202231A1 (zh)

Cited By (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN117173543A (zh) * 2023-11-02 2023-12-05 天津大学 一种肺腺癌和肺结核的混合图像重构方法及系统
CN118869430A (zh) * 2024-09-26 2024-10-29 北京航空航天大学杭州创新研究院 一种基于多层次图变分推断的服务网络感知学习方法

Families Citing this family (1)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN118236056B (zh) * 2024-03-22 2024-09-17 济纶医工智能科技(南京)有限公司 基于wovae模型寻求最佳电气特征的方法、存储介质、设备

Citations (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN109146988A (zh) * 2018-06-27 2019-01-04 南京邮电大学 基于vaegan的非完全投影ct图像重建方法
CN111584072A (zh) * 2020-05-12 2020-08-25 苏州脉康医疗科技有限公司 一种适于小样本的神经网络模型训练方法
CN112200306A (zh) * 2020-10-15 2021-01-08 北京航空航天大学 一种基于深度学习的电阻抗成像方法
US20210074036A1 (en) * 2018-03-23 2021-03-11 Memorial Sloan Kettering Cancer Center Deep encoder-decoder models for reconstructing biomedical images

Patent Citations (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US20210074036A1 (en) * 2018-03-23 2021-03-11 Memorial Sloan Kettering Cancer Center Deep encoder-decoder models for reconstructing biomedical images
CN109146988A (zh) * 2018-06-27 2019-01-04 南京邮电大学 基于vaegan的非完全投影ct图像重建方法
CN111584072A (zh) * 2020-05-12 2020-08-25 苏州脉康医疗科技有限公司 一种适于小样本的神经网络模型训练方法
CN112200306A (zh) * 2020-10-15 2021-01-08 北京航空航天大学 一种基于深度学习的电阻抗成像方法

Non-Patent Citations (1)

* Cited by examiner, † Cited by third party
Title
See also references of EP4513436A4 *

Cited By (3)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN117173543A (zh) * 2023-11-02 2023-12-05 天津大学 一种肺腺癌和肺结核的混合图像重构方法及系统
CN117173543B (zh) * 2023-11-02 2024-02-02 天津大学 一种肺腺癌和肺结核的混合图像重构方法及系统
CN118869430A (zh) * 2024-09-26 2024-10-29 北京航空航天大学杭州创新研究院 一种基于多层次图变分推断的服务网络感知学习方法

Also Published As

Publication number Publication date
EP4513436A1 (en) 2025-02-26
US20250225687A1 (en) 2025-07-10
CN117425920A (zh) 2024-01-19
EP4513436A4 (en) 2026-04-01
JP2025513161A (ja) 2025-04-24

Similar Documents

Publication Publication Date Title
WO2023202231A1 (zh) 一种图像重建方法、装置、电子设备及存储介质
Guo et al. Physics embedded deep neural network for solving full-wave inverse scattering problems
CN109754402B (zh) 图像处理方法、图像处理装置以及存储介质
US9965873B2 (en) Systems and methods for data and model-driven image reconstruction and enhancement
CN112258423A (zh) 基于深度学习的去伪影方法、装置、设备和存储介质
CN112884819B (zh) 一种影像配准及神经网络的训练方法、装置和设备
JP2021504830A (ja) 画像内のオブジェクトをセグメント化するためのセグメンテーションシステム
CN116051849A (zh) 一种脑网络数据特征提取方法及装置
Lin et al. Feature-based inversion using variational autoencoder for electrical impedance tomography
CN117083639A (zh) 针对隐式对象表示的机器学习模型
CN117635444A (zh) 基于辐射差和空间距离的深度补全方法、装置和设备
CN114299185B (zh) 磁共振图像生成方法、装置、计算机设备和存储介质
Wang et al. Unsupervised coordinate-based neural network for electrical impedance tomography
Ghimire et al. Improving generalization of deep networks for inverse reconstruction of image sequences
CN116645347B (zh) 基于多视图特征融合的糖尿病足预测方法、装置及设备
CN110415341B (zh) 一种三维人脸模型的生成方法、装置、电子设备及介质
US20160335786A1 (en) Recovery of missing information in diffusion magnetic resonance imaging data
JP7551620B2 (ja) データ拡張
CN112750110A (zh) 基于神经网络对肺部病灶区进行评估的评估系统和相关产品
CN120182469A (zh) 生成图像的方法、装置和训练图像生成模型的方法、装置
CN117541892A (zh) 深度生成模型的训练方法、装置、设备及介质
CN114926582A (zh) 三维模型生成方法和装置、设备、存储介质
CN114913397A (zh) 面向模态缺失的图表示学习方法、系统、装置及存储介质
Aminmansour et al. Learning macroscopic brain connectomes via group-sparse factorization
CN112669450A (zh) 人体模型构建方法和个性化人体模型构建方法

Legal Events

Date Code Title Description
WWE Wipo information: entry into national phase

Ref document number: 202380012111.1

Country of ref document: CN

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

Ref document number: 23790894

Country of ref document: EP

Kind code of ref document: A1

WWE Wipo information: entry into national phase

Ref document number: 2024549760

Country of ref document: JP

WWE Wipo information: entry into national phase

Ref document number: 18853527

Country of ref document: US

WWE Wipo information: entry into national phase

Ref document number: 2023790894

Country of ref document: EP

NENP Non-entry into the national phase

Ref country code: DE

ENP Entry into the national phase

Ref document number: 2023790894

Country of ref document: EP

Effective date: 20241120

WWP Wipo information: published in national office

Ref document number: 18853527

Country of ref document: US