WO2024243733A1 - 一种磁共振定量生理参数图生成方法和装置 - Google Patents

一种磁共振定量生理参数图生成方法和装置 Download PDF

Info

Publication number
WO2024243733A1
WO2024243733A1 PCT/CN2023/096606 CN2023096606W WO2024243733A1 WO 2024243733 A1 WO2024243733 A1 WO 2024243733A1 CN 2023096606 W CN2023096606 W CN 2023096606W WO 2024243733 A1 WO2024243733 A1 WO 2024243733A1
Authority
WO
WIPO (PCT)
Prior art keywords
physiological parameter
dce
parameter map
quantitative physiological
network
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/CN2023/096606
Other languages
English (en)
French (fr)
Inventor
张娜
郑海荣
刘新
胡战利
梁栋
李烨
邹超
曾道辉
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
Priority to PCT/CN2023/096606 priority Critical patent/WO2024243733A1/zh
Publication of WO2024243733A1 publication Critical patent/WO2024243733A1/zh
Anticipated expiration legal-status Critical
Ceased legal-status Critical Current

Links

Classifications

    • G—PHYSICS
    • G16—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR SPECIFIC APPLICATION FIELDS
    • G16H—HEALTHCARE INFORMATICS, i.e. INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR THE HANDLING OR PROCESSING OF MEDICAL OR HEALTHCARE DATA
    • G16H30/00—ICT specially adapted for the handling or processing of medical images

Definitions

  • the embodiments of the present specification relate to the field of imaging technology, and in particular to a method and device for generating a magnetic resonance quantitative physiological parameter map.
  • Quantitative physiological parameters are numerical parameters used to evaluate the physiological function and metabolic state of tissues. They can help doctors diagnose and monitor some diseases, such as tumors and strokes.
  • quantitative physiological parameters are digital data obtained through imaging technology, which can reflect the physiological conditions of blood flow, blood oxygen, metabolism, etc. in tissues. These parameters are usually calculated from imaging data (such as computed tomography (CT), magnetic resonance imaging (MRI), ultrasound imaging (US), and positron emission tomography (PET)). Therefore, different model assumptions and algorithms may lead to different results of the generated parameter graphs.
  • CT computed tomography
  • MRI magnetic resonance imaging
  • US ultrasound imaging
  • PET positron emission tomography
  • AIF Arterial Input Function
  • AIF is one of the important parameters for calculating quantitative physiological parameters. The accuracy and reliability of AIF often depend on the time resolution and sampling rate of the collected data. However, since it is difficult to estimate the AIF of each patient, the group average AIF is usually used for PK analysis. This method has certain limitations and errors.
  • the embodiments of this specification aim to provide a method for generating a magnetic resonance quantitative physiological parameter map.
  • the purpose of the embodiments of this specification is to provide a method for generating a magnetic resonance quantitative physiological parameter map to solve the problems of complicated operation and low accuracy in the prior art when generating a quantitative physiological parameter map.
  • the embodiments of this specification provide a method for generating a magnetic resonance quantitative physiological parameter map, comprising:
  • the groups of DCE-MRI images and the quantitative physiological parameter maps are used as training inputs to train the deep learning model until the loss function value of the deep learning model meets the preset requirements
  • the deep learning model is a generative adversarial network
  • the generative adversarial network includes a generative network and a discriminative network
  • the generative network is used to generate an output image
  • the discriminative network includes a global discriminator and a local discriminator
  • the global discriminator is used to discriminate the similarity between the output image and the quantitative physiological parameter map
  • the local discriminator is used to discriminate the local sense of the output image. similarity between the region of interest and a corresponding local region of interest in the quantitative physiological parameter map;
  • the deep learning model when the loss function value meets the preset requirements is used as a quantitative physiological parameter map generation model, and the DCE-MRI image to be processed is processed to obtain a quantitative physiological parameter map.
  • LOSS is the loss value of the deep learning model
  • loss_G is the loss value of the generator network
  • loss_D is the loss value of the discriminator network
  • loss_D loss_D 1 * ⁇ +loss_D 2 *(1- ⁇ ) ;
  • loss_D 1 is the loss value of the global discriminator
  • loss_D 2 is the loss value of the local discriminator
  • ⁇ is a constant coefficient greater than 0 and less than 1.
  • the generative network is a generative network with group normalization added
  • the discriminative network is a discriminative network with spectral normalization added.
  • a quantitative physiological parameter map corresponding to each group of DCE-MRI images is obtained, including:
  • the Ktrans quantitative physiological parameter map corresponding to the group of DCE-MRI images is obtained.
  • the method further includes:
  • Each group of DCE-MRI images is aligned and/or registered.
  • each group of DCE-MRI images is aligned, including:
  • each group of DCE-MRI images is registered, including:
  • the other images are registered with the reference image according to the determined similarity change model.
  • the embodiments of this specification also provide a magnetic resonance quantitative physiological parameter map generating device, comprising:
  • An acquisition module used for acquiring each group of DCE-MRI images and a quantitative physiological parameter map corresponding to each DCE-MRI image
  • a training module used to use the groups of DCE-MRI images and the quantitative physiological parameter map as training inputs to train a deep learning model until the loss function value of the deep learning model meets a preset requirement
  • the deep learning model is a generative adversarial network
  • the generative adversarial network includes a generative network and a discriminative network
  • the generative network is used to generate an output image
  • the discriminative network includes a global discriminator and a local discriminator
  • the global discriminator is used to discriminate the similarity between the output image and the quantitative physiological parameter map
  • the local discriminator is used to discriminate the similarity between a local region of interest of the output image and a corresponding local region of interest in the quantitative physiological parameter map
  • the processing module is used to use the deep learning model when the loss function value meets the preset requirements as a quantitative physiological parameter map generation model, and process the DCE-MRI image to be processed to obtain a quantitative physiological parameter map.
  • an embodiment of the present specification also provides a computer device, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor implements the method provided by the above technical solution when executing the computer program.
  • an embodiment of the present specification further provides a computer-readable storage medium, wherein the computer-readable storage medium stores a computer program, and when the computer program is executed by a processor, the method provided by the above technical solution is implemented.
  • the method for generating a magnetic resonance quantitative physiological parameter map provided in the embodiment of this specification can learn and identify complex features and patterns from a large amount of training data during the model training stage. There is no need to model based on a preset hemodynamic model, nor is there a need to manually select multiple parameters to calculate parameters, thereby avoiding dependence on assumptions such as vascular continuity and fluid incompressibility, reducing the degree of manual intervention, and helping to improve the accuracy of the calculation results.
  • the requirements for the DCE-MRI images to be processed can be reduced.
  • a relatively accurate Ktrans quantitative physiological parameter map can be generated; and the efficiency of generating quantitative physiological parameter maps can be greatly improved.
  • FIG1 is a schematic diagram showing the steps of a method for generating a magnetic resonance quantitative physiological parameter map provided in an embodiment of this specification
  • FIG2 is a comparison diagram showing a quantitative physiological parameter map obtained by using the magnetic resonance quantitative physiological parameter map generation method provided in an embodiment of this specification and a quantitative physiological parameter map obtained by using a traditional method;
  • FIG3 is a schematic diagram of the steps of obtaining quantitative physiological parameter maps corresponding to each group of DCE-MRI images
  • FIG4 is a schematic diagram showing the steps of aligning each group of DCE-MRI images
  • FIG5 is a schematic diagram showing the steps of performing registration processing on each group of DCE-MRI images
  • FIG6 shows a schematic diagram of the structure of a magnetic resonance quantitative physiological parameter image generating device provided in an embodiment of this specification
  • FIG. 7 shows a schematic diagram of the structure of a computer device provided in an embodiment of this specification.
  • Processing module
  • Figure 1 is a schematic diagram of the steps of a method for generating a magnetic resonance quantitative physiological parameter map provided in an embodiment of this specification.
  • This specification provides method operation steps as described in an embodiment or a flow chart, but based on conventional or non-creative labor, more or fewer operation steps may be included.
  • the order of steps listed in the embodiment is only one way of executing the steps among many steps, and does not represent the only execution order.
  • the system or device product is executed in practice, it can be executed in the order of the method shown in the embodiment or the accompanying drawings or in parallel.
  • the method may include:
  • S110 Acquire each group of DCE-MRI images and a quantitative physiological parameter map corresponding to each group of DCE-MRI images.
  • a set of DCE-MRI images in the embodiments of this specification can be multiple (for example, 45 periods) of image acquisition of the same tissue (for example, breast) of the same object at a certain interval.
  • Contrast agent also called contrast agent
  • the inflow and outflow of contrast agent is a dynamic process, from no contrast agent at the beginning to the amount of contrast agent diffused (flowed) into the tissue reaching a peak value, and then slowly dissipated (flowed out) from the tissue). Repeat the above operation to obtain different Each set of DCE-MRI images of different tissues of the subject.
  • the deep learning model is trained until the loss function value of the deep learning model meets preset requirements, wherein the deep learning model is a generative adversarial network, the generative adversarial network includes a generative network and a discriminative network, the generative network is used to generate an output image, the discriminative network includes a global discriminator and a local discriminator, the global discriminator is used to discriminate the similarity between the output image and the quantitative physiological parameter map, and the local discriminator is used to discriminate the similarity between the local region of interest of the output image and the corresponding local region of interest in the quantitative physiological parameter map.
  • the deep learning model is a generative adversarial network
  • the generative adversarial network includes a generative network and a discriminative network
  • the generative network is used to generate an output image
  • the discriminative network includes a global discriminator and a local discriminator
  • the global discriminator is used to discriminate the similarity between the output image and the quantitative physiological parameter map
  • the local discriminator is used to discriminate the similarity between the
  • the generating network can be resnet, unet and unet++.
  • the generation network is preferably unet.
  • the discriminator network is PatchGAN; the batch size (Batch Size) is set to 4; the total number of training time periods is set to 200 epochs: the learning rate of the first 100 time periods is set to 0.0002, and the learning rate of the remaining 100 time periods gradually decreases to 0.
  • the number of local regions of interest, i.e., local ROIs (Region of regular) in a set of DCE-MRI images can be set to one or more according to actual needs, and the size of the local regions of interest can be set according to actual needs.
  • a tumor area or a suspected tumor area in a DCE-MRI image can be selected as a local region of interest.
  • the deep learning model is a generative adversarial network (GAN), which uses a dual discriminator network, i.e., a global discriminator and a local discriminator, which can improve the accuracy of the judgment of the generation effect of the output image, and is conducive to improving the quality of the quantitative physiological parameter map obtained by the subsequent trained quantitative physiological parameter map generation model.
  • GAN generative adversarial network
  • Figure 2 (a) is a DCE-MRI image to be processed
  • Figure 2 (b) is a quantitative physiological parameter map obtained by using the magnetic resonance quantitative physiological parameter map generation method provided in the embodiment of this specification
  • Figure 2 (c) is a quantitative physiological parameter map obtained by using the traditional method.
  • the quantitative physiological parameter map generation method provided in the embodiment of this specification adopts a deep learning model, and the deep learning model can learn from a large amount of data, and identify and eliminate useless information (such as noise) in the learning process.
  • the quantitative physiological parameter map generation model finally iteratively optimized can remove the interference of noise and improve the generation quality of the quantitative physiological parameter map.
  • the requirements for the DCE-MRI images to be processed can be reduced to a certain extent.
  • the number of DCE-MRI images can be reduced (i.e., less than 45 images).
  • the corresponding quantitative physiological parameter graph is generated, which reduces the workload.
  • the magnetic resonance quantitative physiological parameter map generation method obtained in the embodiments of this specification obtains a trained quantitative physiological parameter map generation model through a generation network and a dual discriminator network, and uses it to generate quantitative physiological parameter maps, which can reduce errors caused by human participation and improve the accuracy of the generated quantitative physiological parameter maps; at the same time, there is no need to perform a large amount of DCE-MRI image acquisition work, modeling work based on hemodynamic models, etc., which greatly reduces the workload and improves the efficiency of quantitative physiological parameter map generation.
  • the quantitative physiological parameter map generation method is to generate corresponding quantitative physiological parameter images for the DCE-MRI images to be processed.
  • imaging data such as computed tomography, ultrasound imaging, and positron emission tomography can also be processed to generate quantitative physiological parameter maps.
  • LOSS is the loss value of the deep learning model
  • loss_G is the loss value of the generator network
  • loss_D is the loss value of the discriminator network.
  • the generator loss and the discriminator loss are independent of each other and constrained by each other.
  • loss_D loss_D 1 * ⁇ +loss_D 2 *(1- ⁇ ) ;
  • loss_D 1 is the loss value of the global discriminator
  • loss_D 2 is the loss value of the local discriminator
  • ⁇ is a constant coefficient greater than 0 and less than 1.
  • the value of ⁇ is 0.5, that is, the weights of the global discriminator and the local discriminator are the same.
  • a separate alternating iterative training method is adopted, that is, the generating network is trained first, and the parameters of the discriminating network remain unchanged during this process, so that the output image generated by the generating network is as similar as possible to the quantitative physiological parameter map in the training data, and the local region of interest of the output image is as similar as possible to the local region of interest corresponding to the quantitative physiological parameter map.
  • the parameters of the generating network are fixed, and the discriminating network is trained, so that the discriminating network's discrimination results of the similarity between the output image and the quantitative physiological parameter map, and the discrimination results of the local region of interest of the output image and the local region of interest corresponding to the quantitative physiological parameter map are more accurate.
  • the generating network and the discriminating network are trained alternately until the training is completed.
  • the generative network is a generative network with group normalization added
  • the discriminative network is a discriminative network with spectral normalization added.
  • the generative network includes convolutional layers, normalization layers, and activation layers.
  • the normalization layer is a group normalization (GN) layer.
  • the input DCE-MRI image is group normalized in the generative network.
  • group normalization divides each channel into several groups, and for each input data, the mean and standard deviation of the feature map in each group are calculated, and the feature map is standardized using them. This allows the feature map of each group to have the same statistical characteristics, thereby reducing the covariate shift of the features within the group, thereby speeding up the training process and improving the performance of the model.
  • the global discriminator and the local discriminator of the discriminant network include a convolution layer, a normalization layer and an activation layer, wherein the normalization layer is a spectral normalization (SN) layer.
  • the spectral normalization method is as follows:
  • the weight matrix W of each layer of the discriminator calculate its left singular vector and right singular vector, u and v respectively; through singular value decomposition, the spectral norm of the weight matrix is obtained as:
  • max
  • u T is the transpose of u.
  • weight matrix W is normalized by dividing it by the largest singular value.
  • the transposed matrix W T of the weight matrix W is repeatedly multiplied by a random vector x, and then normalized to obtain a new vector y until the difference between y and the vector of the previous iteration is less than a threshold. Finally, the value of sigma is set to the inner product of y and x when the power iteration converges.
  • weight matrix W is normalized by dividing it by sigma.
  • the magnetic resonance quantitative physiological parameter map generation method of the embodiment of this specification improves the computational efficiency of the model and enhances the scalability and robustness of the model by adding a group-normalized generation network; and limits the singular values of the weight matrix by adding a spectral-normalized discriminant network to avoid the vanishing or exploding gradient of the weights, thereby improving the stability and generalization ability of the model, which is ultimately beneficial to improving the quality of the quantitative physiological parameter map generated by the quantitative physiological parameter map generation model.
  • obtaining a quantitative physiological parameter map corresponding to each group of DCE-MRI images further includes:
  • the DCE-MRI image may be preprocessed by baseline correction and denoising.
  • S320 Establish a curve of the change of the contrast agent concentration over time at the corresponding pixel points in the group of DCE-MRI images.
  • C(t) is the contrast agent concentration of a certain pixel at time t
  • Ktrans is the transfer constant in the tissue
  • AIF(t') is the AIF curve value at time t'
  • C(t') is the contrast agent concentration of a certain pixel at time t'
  • AIF is the arterial input function (Arterial Input Function, AIF).
  • a certain large blood vessel in the tissue is preferably used as the AIF, or a pre-determined average AIF curve (based on the AIF of the population) is used.
  • nonlinear least squares NLS
  • LLS linear least squares
  • Ktrans represents the rate at which the contrast agent flows from the blood vessels into the interstitial tissue, which is a comprehensive expression of perfusion and penetration. It can be understood as the transfer (rate) constant of the contrast agent flowing from the blood vessels into the interstitial tissue, which can be used to display the vascular permeability of different areas in the tissue, and can provide technical support for subsequent data analysis, tumor diagnosis and treatment, etc.
  • the quantitative physiological parameter map used as the training data of the deep learning model is the Ktrans quantitative physiological parameter map.
  • the quantitative physiological parameter map used as the training data of the deep learning model can also be the Kep quantitative physiological parameter map (Kep is the rate at which the contrast agent flows out of the tissue from the interstitial tissue), the ve quantitative physiological parameter map (ve is the proportion of the interstitial tissue), etc.
  • the processed quantitative physiological parameter map is adapted to the training data.
  • the method before obtaining the quantitative physiological parameter graph corresponding to each group of DCE-MRI images in step S110, the method further includes:
  • Each group of DCE-MRI images is aligned and/or registered.
  • the alignment process of each group of DCE-MRI images may include the following steps:
  • S410 Select any one of the set of DCE-MRI images as a reference image, and obtain deviation information between other images in the set of DCE-MRI images and the reference image.
  • S420 Align other images with the reference image according to the deviation information.
  • other images can be aligned with the reference image through translation, rotation, scaling, and other transformation operations based on the deviation information.
  • the acquisition object may move or the position of the image acquisition location may change due to factors such as breathing, thereby causing the input data to be inaccurate or unable to be analyzed.
  • the accuracy and reliability of the data can be improved by aligning each group of DCE-MRI images.
  • each group of DCE-MRI images is subjected to registration processing, including:
  • S510 Select any one of the set of DCE-MRI images as a reference image, and extract the same feature points and/or the same feature regions of the reference image and other images in the set of DCE-MRI images.
  • Feature points and feature areas may be objects with significance and uniqueness in the DCE-MRI image, and may be, for example, closed boundary areas, edge areas, line intersections, etc. After selecting appropriate feature points and/or feature areas, the feature points and/or feature areas in other images are matched with corresponding feature points and/or feature areas in the reference image, so that the other images are correlated with the reference image.
  • S520 Determine a similarity transformation model between other images and the reference image based on the same feature point and/or the same area.
  • the registration process is performed on the two (or more) images before and after in the same group of DCE-MRI images.
  • the registration process may be performed on two (or more) images in DCE-MRI images belonging to the same tissue of the same object but belonging to different groups. It may also be the registration between different modality images of the same tissue of the same object, for example, the registration process may be performed on CT images and DCE-MRI images.
  • Registration processing can compare images acquired by different acquisition groups or different devices. Through the registration operation, these images can be transformed into a unified coordinate system, thereby achieving more accurate comparison and analysis.
  • alignment processing and registration processing can be used in combination.
  • the positions and directions of the DCE-MRI images and the local areas of interest of the DCE-MRI images are correctly aligned, which is beneficial to improve the accuracy of the calculated Ktrans value and provide a reliable basis for subsequent data analysis and clinical applications.
  • An interpolation operation is performed on the aligned DCE-MRI image and/or the registered DCE-MRI image to generate a new image with the same pixel spacing and direction, that is, to increase the amount of training input data.
  • the magnetic resonance quantitative physiological parameter map generation method can learn and identify complex features and patterns from a large amount of training data during the model training stage, and does not need to be modeled based on a preset hemodynamic model, thereby avoiding dependence on assumptions such as vascular continuity and fluid incompressibility. And because the deep learning model can automatically learn parameters from the data, there is no need to manually select multiple parameters to calculate the perfusion parameters, which reduces the degree of manual intervention and avoids the problem of inaccurate calculation results due to improper parameter selection.
  • a trained deep learning model i.e., a quantitative physiological parameter map generation model
  • a large amount of mathematical calculations and analysis can be avoided, reducing computational complexity while greatly improving data processing efficiency.
  • the trained model can reduce the requirements for the DCE-MRI images to be processed. For a single DCE-MRI image or a small number of DCE-MRI images, the quantitative physiological parameter map generation model can generate accurate results.
  • the embodiment of this specification also provides a magnetic resonance quantitative physiological parameter map generating device, comprising:
  • An acquisition module 61 used for acquiring each group of DCE-MRI images and a quantitative physiological parameter map corresponding to each DCE-MRI image
  • a training module 62 is used to train a deep learning model using the DCE-MRI image and the quantitative physiological parameter map as training inputs until the loss function value of the deep learning model meets a preset requirement, wherein the deep learning model is a generative adversarial network, wherein the generative adversarial network includes a generative network and a discriminative network, wherein the generative network is used to generate an output image, wherein the discriminative network includes a global discriminator and a local discriminator, wherein the global discriminator is used to discriminate the similarity between the output image and the quantitative physiological parameter map, and wherein the local discriminator is used to discriminate the similarity between a local region of interest of the output image and a corresponding local region of interest in the quantitative physiological parameter map.
  • the processing module 63 is used to use the deep learning model when the loss function value meets the preset requirements as a quantitative physiological parameter map generation model, and process the DCE-MRI image to be processed to obtain a quantitative physiological parameter map.
  • the magnetic resonance quantitative physiological parameter image generating device in the present specification may be the computer device in the present embodiment, which executes the magnetic resonance quantitative physiological parameter image generating method provided in the embodiment of the present specification.
  • the computer device 702 may include one or more processors 704, such as one or more central processing units (CPUs), and each processing unit may implement one or more hardware threads.
  • the computer device 702 may also include any memory 706, which is used to store any kind of information such as code, settings, data, etc.
  • the memory 706 may include any one or more combinations of the following: any type of RAM, any type of ROM, flash memory device, hard disk, optical disk, etc.
  • any memory is Any technology can be used to store information. Further, any memory can provide volatile or non-volatile retention of information. Further, any memory can represent a fixed or removable component of the computer device 702. In one case, when the processor 704 executes the associated instructions stored in any memory or combination of memories, the computer device 702 can perform any operation of the associated instructions.
  • the computer device 702 also includes one or more drive mechanisms 708 for interacting with any memory, such as a hard disk drive mechanism, an optical disk drive mechanism, etc.
  • the computer device 702 may also include an input/output module 710 (I/O) for receiving various inputs (via input devices 712) and for providing various outputs (via output devices 714).
  • a specific output mechanism may include a presentation device 716 and an associated graphical user interface (GUI) 718.
  • GUI graphical user interface
  • the input/output module 710 (I/O), the input device 712, and the output device 714 may not be included, and the computer device 702 may be used as a computer device in a network.
  • the computer device 702 may also include one or more network interfaces 720 for exchanging data with other devices via one or more communication links 722.
  • One or more communication buses 724 couple the components described above together.
  • the communication link 722 may be implemented in any manner, for example, through a local area network, a wide area network (e.g., the Internet), a point-to-point connection, etc., or any combination thereof.
  • the communication link 722 may include any combination of hardwired links, wireless links, routers, gateway functions, name servers, etc. governed by any protocol or combination of protocols.
  • the embodiments of this specification further provide a computer-readable storage medium, on which a computer program is stored, and when the computer program is executed by a processor, the steps of the above method are executed.
  • the embodiments of the present specification also provide a computer-readable instruction, wherein when a processor executes the instruction, the program therein causes the processor to execute the method shown in FIG. 1 and FIG. 3 to FIG. 5 .
  • the embodiments of the present specification also provide a computer program product, including at least one instruction or at least one program, wherein the at least one instruction or the at least one program is loaded and executed by a processor to implement the method shown in FIG. 1 and FIG. 3 to FIG. 5 .
  • the disclosed systems, devices and methods can be implemented in other ways.
  • the device embodiments described above are only schematic.
  • the division of the units is only a logical function division. There may be other division methods in actual implementation, such as multiple units or components can be combined or integrated into another system, or some features can be ignored or not executed.
  • the mutual coupling or direct coupling or communication connection shown or discussed can be an indirect coupling or communication connection through some interfaces, devices or units, or it can be an electrical, mechanical or other form of connection.
  • the units described as separate components may or may not be physically separated, and the components shown as units may or may not be physical units, that is, they may be located in one place or distributed on multiple network units. Some or all of the units may be selected according to actual needs to achieve the purpose of the embodiments of this specification.
  • each functional unit in each embodiment of this specification 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 may be implemented in the form of hardware or in the form of software functional units.
  • the integrated unit is implemented in the form of a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium.
  • the technical solution of the embodiment of this specification is essentially or the part that contributes to the prior art, or all or part of the technical solution can be embodied in the form of a software product, and the computer software product is stored in a storage medium, including a number of instructions to enable a computer device (which can be a personal computer, a server, or a network device, etc.) to perform all or part of the steps of the method described in each embodiment of this specification.
  • the aforementioned storage medium includes: U disk, mobile hard disk, read-only memory (ROM, Read-Only Memory), random access memory (RAM, Random Access Memory), disk or optical disk and other media that can store program code.

Landscapes

  • Health & Medical Sciences (AREA)
  • Nuclear Medicine, Radiotherapy & Molecular Imaging (AREA)
  • Radiology & Medical Imaging (AREA)
  • Engineering & Computer Science (AREA)
  • Epidemiology (AREA)
  • General Health & Medical Sciences (AREA)
  • Medical Informatics (AREA)
  • Primary Health Care (AREA)
  • Public Health (AREA)
  • Magnetic Resonance Imaging Apparatus (AREA)

Abstract

提供了一种磁共振定量生理参数图生成方法和装置，包括获取各组DCE-MRI图像和与各组DCE-MRI图像对应的定量生理参数图；将各组DCE-MRI图像和定量生理参数图作为训练输入，对深度学习模型进行训练直至其损失函数值满足预设要求，深度学习模型为包括生成网络和判别网络的生成对抗网络，生成网络用于生成输出图像，判别网络包括全局判别器和局部判别器，将损失函数值满足预设要求时的深度学习模型作为定量生理参数图生成模型，对待处理DCE-MRI图像进行处理得到定量生理参数图，能够提高定量生理参数图生成效率和准确性。

Description

一种磁共振定量生理参数图生成方法和装置 技术领域
本说明书实施例涉及成像技术领域，尤其是一种磁共振定量生理参数图生成方法和装置。
背景技术
定量生理参数是用于评估组织内生理功能和代谢状态的数值化参数，它可以帮助医生诊断和监测一些疾病，例如肿瘤、中风等。在医学影像学中，定量生理参数是通过成像技术获得的数字数据，可以反映组织内血流、血氧、代谢等方面的生理状况。这些参数通常是从成像数据(如计算机断层扫描(CT)、磁共振成像(MRI)、超声成像(US)和正电子发射断层扫描(PET)等成像数据)中计算得出的，因此，不同的模型假设和算法都可能导致生成的参数图结果不同。动脉输入函数(Arterial Input Function，AIF)是计算定量生理参数的重要参数之一，AIF的精度和可靠性往往取决于采集数据的时间分辨率和采样率。然而，由于难以估计每个患者的AIF，因此通常采用群体平均AIF进行PK分析，这种方法存在一定的局限性和误差。此外，传统的定量生理参数生成，还需要花费大量的计算资源和时间。
有鉴于此，本说明书实施例旨在提供一种磁共振定量生理参数图生成方法。
发明内容
针对现有技术的上述问题，本说明书实施例的目的在于，提供一种磁共振定量生理参数图生成方法，以解决现有技术在生成定量生理参数图时操作繁琐、精确度低的问题。
为了解决上述技术问题，本说明书实施例的具体技术方案如下：
第一方面，本说明书实施例提供一种磁共振定量生理参数图生成方法，包括：
获取各组DCE-MRI图像和与各组DCE-MRI图像对应的定量生理参数图；
将所述各组DCE-MRI图像和所述定量生理参数图作为训练输入，对深度学习模型进行训练，直至所述深度学习模型的损失函数值满足预设要求，所述深度学习模型为生成对抗网络，所述生成对抗网络包括生成网络和判别网络，所述生成网络用于生成输出图像，所述判别网络包括全局判别器和局部判别器，所述全局判别器用于判别所述输出图像与所述定量生理参数图的相似性，所述局部判别器用于判别所述输出图像的局部感 兴趣区与所述定量生理参数图中对应的局部感兴趣区的相似性；
将所述损失函数值满足预设要求时的深度学习模型作为定量生理参数图生成模型，对待处理DCE-MRI图像进行处理得到定量生理参数图。
具体地，所述损失函数为：
LOSS＝loss_G+loss_D；
其中，LOSS为所述深度学习模型的损失值，loss_G为生成网络的损失值，loss_D为判别网络的损失值；
loss_D＝loss_D1*α+loss_D2*(1-α)；
其中，loss_D1为全局判别器的损失值，loss_D2为局部判别器的损失值，α为大于0小于1的常系数。
进一步地，所述生成网络为加入组归一化的生成网络，所述判别网络为加入谱归一化的判别网络。
进一步地，获取与各组DCE-MRI图像对应的定量生理参数图，包括：
提取同一组DCE-MRI图像中的任一像素点的时间-信号曲线；
建立该组DCE-MRI图像中的相应像素点的对比剂浓度随时间的变化曲线；
拟合所述时间-信号曲线与所述变化曲线，得到该组DCE-MRI图像中对应像素点的Ktrans值；
重复上述过程，得到同一组DCE-MRI图像中所有像素点的Ktrans值；
基于所有像素点的Ktrans值得到该组DCE-MRI图像对应的Ktrans定量生理参数图。
更进一步地，在获取与各组DCE-MRI图像对应的定量生理参数图之前，所述方法还包括：
对各组DCE-MRI图像进行对齐处理和/或配准处理。
具体地，对各组DCE-MRI图像进行对齐处理，包括：
选择该组DCE-MRI图像中的任意一幅作为参考图像，得到该组DCE-MRI图像中的其他图像与所述参考图像之间的偏差信息；
根据所述偏差信息，将其他图像与所述参考图像对齐。
具体地，对各组DCE-MRI图像进行配准处理，包括：
选择该组DCE-MRI图像中的任意一幅作为参考图像，提取所述参考图像和该组DCE-MRI图像中的其他图像的同一特征点和/或同一特征区域；
根据所述同一特征点和/或同一区域，确定其他图像与所述参考图像之间的相似性变换模型；
根据确定出的相似性变化模型，将其他图像与所述参考图像进行配准。
第二方面，本说明书实施例还提供一种磁共振定量生理参数图生成装置，包括：
获取模块，用于获取各组DCE-MRI图像和与各DCE-MRI图像对应的定量生理参数图；
训练模块，用于将所述各组DCE-MRI图像和所述定量生理参数图作为训练输入，对深度学习模型进行训练，直至所述深度学习模型的损失函数值满足预设要求，所述深度学习模型为生成对抗网络，所述生成对抗网络包括生成网络和判别网络，所述生成网络用于生成输出图像，所述判别网络包括全局判别器和局部判别器，所述全局判别器用于判别所述输出图像与所述定量生理参数图的相似性，所述局部判别器用于判别所述输出图像的局部感兴趣区与所述定量生理参数图中对应的局部感兴趣区的相似性；
处理模块，用于将所述损失函数值满足预设要求时的深度学习模型作为定量生理参数图生成模型，对待处理DCE-MRI图像进行处理得到定量生理参数图。
第三方面，本说明书实施例还提供一种计算机设备，包括存储器、处理器及存储在存储器上并可在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技术方案提供的方法。
第四方面，本说明书实施例还提供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技术方案提供的方法。
采用上述技术方案，本说明书实施例提供的磁共振定量生理参数图生成方法，在模型训练阶段，能够从大量的训练数据中学习和识别复杂的特征和模式，无需基于预设的血流动力学模型进行建模，也无需手动选择多个参数来计算参数，从而避免了对血管连续性和流体不可压缩性等假设的依赖，减少人工干预的程度，有利于提高计算结果的准确性。
在模型使用阶段，可降低对待处理的DCE-MRI图像的要求，对于单张DCE-MRI图像或少量DCE-MRI图像，均可生成较为准确的Ktrans定量生理参数图；且能够极大地提高定量生理参数图生成效率。
为让本说明书实施例的上述和其他目的、特征和优点能更明显易懂，下文特举较佳实施例，并配合所附图式，作详细说明如下。
附图说明
为了更清楚地说明本文实施例或现有技术中的技术方案，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文的一些实施例，对于本领域普通技术人员来讲，在不付出创造性劳动的前提下，还可以根据这些附图获得其他的附图。
图1示出了本说明书实施例提供的一种磁共振定量生理参数图生成方法的步骤示意图；
图2示出了采用本说明书实施例提供的磁共振定量生理参数图生成方法得到的定量生理参数图和采用传统方法得到定量生理参数图的对比图；
图3获取与各组DCE-MRI图像对应的定量生理参数图的步骤示意图；
图4示出了对各组DCE-MRI图像进行对齐处理的步骤示意图；
图5示出了对各组DCE-MRI图像进行配准处理的步骤示意图；
图6示出了本说明书实施例提供的一种磁共振定量生理参数图生成装置的结构示意图；
图7示出了本说明书实施例提供的一种计算机设备的结构示意图。
附图符号说明：
61、获取模块；
62、训练模块；
63、处理模块；
702、计算机设备；
704、处理器；
706、存储器；
708、驱动机构；
710、输入/输出模块；
712、输入设备；
714、输出设备；
716、呈现设备；
718、图形用户接口；
720、网络接口；
722、通信链路；
724、通信总线。
具体实施方式
下面将结合本文实施例中的附图，对本文实施例中的技术方案进行清楚、完整地描述，显然，所描述的实施例仅仅是本文一部分实施例，而不是全部的实施例。基于本文中的实施例，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他实施例，都属于本文保护的范围。
需要说明的是，本文的说明书和权利要求书及上述附图中的术语“第一”、“第二”等是用于区别类似的对象，而不必用于描述特定的顺序或先后次序。应该理解这样使用的数据在适当情况下可以互换，以便这里描述的本文的实施例能够以除了在这里图示或描述的那些以外的顺序实施。此外，术语“包括”和“具有”以及他们的任何变形，意图在于覆盖不排他的包含，例如，包含了一系列步骤或单元的过程、方法、装置、产品或设备不必限于清楚地列出的那些步骤或单元，而是可包括没有清楚地列出的或对于这些过程、方法、产品或设备固有的其它步骤或单元。
本说明书实施例提供了磁共振定量生理参数图生成方法，能够减少人工干预降低误差、提高处理效率。图1是本说明书实施例提供的一种磁共振定量生理参数图生成方法的步骤示意图，本说明书提供了如实施例或流程图所述的方法操作步骤，但基于常规或者无创造性的劳动可以包括更多或者更少的操作步骤。实施例中列举的步骤顺序仅仅为众多步骤执行顺序中的一种方式，不代表唯一的执行顺序。在实际中的系统或装置产品执行时，可以按照实施例或者附图所示的方法顺序执行或者并行执行。具体的如图1所示，所述方法可以包括：
S110：获取各组DCE-MRI图像和与各组DCE-MRI图像对应的定量生理参数图。
示例性的，本说明书实施例中的一组DCE-MRI图像可以是按一定的间隔时间，对同一对象的同一处组织(例如，乳腺)进行的多次(例如，45期)的图像采集。采集前需要注入造影剂(也称为对比剂)，通过图像采集观测到造影剂在组织内的聚集情况(造影剂的流入流出是一个动态的过程，从刚开始没有，到造影剂扩散(流入)到组织内的量达到一个峰值，再慢慢从组织内消散(流出)出去)。重复上述操作，即可获得不同 对象不同组织的各组DCE-MRI图像。
S120：将所述各组DCE-MRI图像和所述定量生理参数图作为训练输入，对深度学习模型进行训练，直至所述深度学习模型的损失函数值满足预设要求，所述深度学习模型为生成对抗网络，所述生成对抗网络包括生成网络和判别网络，所述生成网络用于生成输出图像，所述判别网络包括全局判别器和局部判别器，所述全局判别器用于判别所述输出图像与所述定量生理参数图的相似性，所述局部判别器用于判别所述输出图像的局部感兴趣区与所述定量生理参数图中对应的局部感兴趣区的相似性。
所述生成网络可以是resnet、unet和unet++。
本说明书实施例中，所述生成网络优选为unet。
所述判别网络为PatchGAN；批量大小(Batch Size)设置为4；训练时间段的总数被设置为200epochs：前100个时间段的学习率被设置为0.0002，而剩余100个时间的学习率逐渐下降到0。
本说明书实施例中，一组DCE-MRI图像中的局部感兴趣区即局部ROI(Region of interes)的数量可根据实际需要设置有一个或多个，且局部感兴趣的大小可以根据实际需要进行设置。一般的，可选择DCE-MRI图像中的肿瘤区域或疑似肿瘤区域作为局部感兴趣区。进一步地，所述深度学习模型为生成对抗网络(Generative Adversarial Networks，GAN)，采用双判别器网络，即全局判别器和局部判别器，可提高对输出图像的生成效果判别的准确性，有利于提高后续用训练好的定量生理参数图生成模型得到的定量生理参数图的质量。
S130：将所述损失函数值满足预设要求时的深度学习模型作为定量生理参数图生成模型，对待处理DCE-MRI图像进行处理得到定量生理参数图。
如图2所示，图2的(a)为待处理的DCE-MRI图像；图2的(b)为采用本说明书实施例提供的磁共振定量生理参数图生成方法得到的定量生理参数图；图2的(c)为采用传统方法得到的定量生理参数图。对比图2的(b)和图2的(c)可以看出，传统的定量生理参数图生成方法容易受到噪声(运动伪影)的干扰。本说明书实施例提供的定量生理参数图生成方法，采用了深度学习模型，而深度学习模型能够从大量的数据中学习，并在学习的过程中识别并剔除无用的信息(比如噪声)，最终迭代优化得到的定量生理参数图生成模型能够去除噪声的干扰，提高定量生理参数图的生成质量。
并且，当定量生理参数图生成模型训练完成后，在一定程度上可降低对待处理DCE-MRI图像的要求，例如，DCE-MRI图像的数量可进行减少(即在少于45期图像的 基础上生成对应的定量生理参数图)，减轻了工作量。
本说明书实施例提供的磁共振定量生理参数图生成方法，通过生成网络和双判别器网络得到训练好的定量生理参数图生成模型，并用于生成定量生理参数图，能够减少由于人工参与带来的误差，提高生成的定量生理参数图的精度；同时无需进行大量的DCE-MRI图像的采集工作、基于血流动力学模型进行建模工作等，极大地降低了工作量，提高了定量生理参数图生成效率。
需要说明的是，本说明书实施例提供的定量生理参数图生成方法，是对待处理的DCE-MRI图像生成对应的定量生理参数图像，在一些其他可行的实施例中，还可以是对计算机断层扫描、超声成像、正电子发射断层扫描等成像数据进行处理，以生成定量生理参数图。
进一步地，本说明书实施例中，所述损失函数为：
LOSS＝loss_G+loss_D；
其中，LOSS为所述深度学习模型的损失值，loss_G为生成网络的损失值，loss_D为判别网络的损失值。生成器损失和判别器损失相互独立又相互制约。
具体地，
loss_D＝loss_D1*α+loss_D2*(1-α)；
其中，loss_D1为全局判别器的损失值，loss_D2为局部判别器的损失值，α为大于0小于1的常系数。
在一些可选的实施例中，α的值为0.5，即全局判别器和局部判别器的比重相同。
需要说明的是，在对本说明书实施例中的深度学习模型进行训练时，采用的是单独交替迭代训练方法，即先训练生成网络，此过程中判别网络的参数不变，使得生成网络生成的输出图像与训练数据中的定量生理参数图的相似程度尽可能高，以及使得输出图像的局部感兴趣区与定量生理参数图对应的局部感兴趣的相似程度尽可能高。迭代训练多次后，再将生成网络的参数固定，对判别网络进行训练，使得判别网络对输出图像与定量生理参数图间相似度的判别结果、输出图像局部感兴趣区与定量生理参数图对应的局部感兴趣间相似度的判别结果更加准确。最后，交替训练生成网络和判别网络，直至训练结束。
进一步地，本说明书实施例中，所述生成网络为加入组归一化的生成网络，所述判别网络为加入谱归一化的判别网络。
生成网络包括卷积层、归一化层和激活层，其中，归一化层为组归一化(Group Normalization，GN)层，输入的DCE-MRI图像在生成网络中进行组归一化处理。在训练时，组归一化将每个通道分成若干个组，并对于每个输入数据计算出其每个组内特征图的均值和标准差，并利用它们对特征图进行标准化。这样可以使每个组的特征图具有相同的统计特性，从而减少了组内特征的协变量偏移(Covariate Shift)现象，进而加快训练进程和提高了模型的性能。
与之相似的，判别网络的全局判别器和局部判别器包括卷积层、归一化层和激活层，其中，归一化层为谱归一化(Spectral Normalization，SN)层，本说明书实施例中，谱归一化方法如下：
对于判别器每一层的权重矩阵W，计算其左奇异向量和右奇异向量，分别为u和v；通过奇异值分解，得到权重矩阵的谱范数为：
||W||＝max|uTWv||；
其中，uT为u的转置。
最后，将权重矩阵W除以最大奇异值进行归一化。
除上述奇异值法外，对于每个权重矩阵W，还可以通过幂迭代算法来求解计算其谱范数的上界sigma，具体步骤为：
将权重矩阵W的转置矩阵WT与一个随机向量x重复相乘，然后归一化得到一个新的向量y，直至y与上一次迭代的向量差距小于一个阈值。最终，sigma的值被设置为幂迭代收敛时y与x的内积。
最后，将权重矩阵W除以sigma进行归一化。
本说明书实施例磁共振定量生理参数图生成方法，通过加入组归一化的生成网络来提高模型的计算效率、增强模型的扩展性和稳健性；通过加入谱归一化的判别网络来限制权重矩阵的奇异值，避免权重的梯度消失或爆炸，提高了模型的稳定性和泛化能力，最终有利于提高定量生理参数图生成模型所生成的定量生理参数图的质量。
如图3所示，步骤S110中的，获取与各组DCE-MRI图像对应的定量生理参数图，进一步包括：
S310：提取同一组DCE-MRI图像中的任一像素点的时间-信号曲线。
可选的，在进行时间-信号曲线提取之前，还可以对DCE-MRI图像进行基线校正和去噪等预处理。
S320：建立该组DCE-MRI图像中的相应像素点的对比剂浓度随时间的变化曲线。
具体地，可根据单室药代动力学模型，建立组织中对比剂浓度与时间之间的关系：
C(t)＝Ktrans∫[AIF(t’)-C(t’)]dt’
其中，C(t)为某个像素点在t时刻的对比剂浓度，Ktrans为组织中的转移常数，AIF(t’)为在t’时刻AIF曲线值，C(t’)为在t’时刻某个像素点的对比剂浓度，AIF为动脉输入函数(Arterial Input Function，AIF)。本说明书实施例中，优选组织中某个大血管作为AIF，或使用预先测定的平均AIF曲线(基于人群的AIF)。
S330：拟合所述时间-信号曲线与所述变化曲线，得到该组DCE-MRI图像中对应像素点的Ktrans值。
具体地，可选用非线性最小二乘法(Nonlinear Least Squares，NLS)或线性最小二乘法(Linear Least Squares，LLS)等方法，对每个像素点的时间-信号曲线和变化曲线进行拟合。
S340：重复上述过程，得到同一组DCE-MRI图像中所有像素点的Ktrans值。
S350：基于所有像素点的Ktrans值得到该组DCE-MRI图像对应的Ktrans定量生理参数图。
Ktrans表示的是造影剂从血管流入组织间质的速率，是灌注和渗透的综合表现，可以理解为造影剂从血管流入组织间质的转移(速率)常数，可用于显示组织中不同区域的血管通透性，能够为后续数据的分析、肿瘤的诊断和治疗等提供技术支持。需要说明的是，本说明书实施例中，作为深度学习模型训练数据的定量生理参数图为Ktrans定量生理参数图，在一些其他可行的实施例中，作为深度学习模型训练数据的定量生理参数图还可以是Kep定量生理参数图(Kep为造影剂从组织间质流出组织的速率)、ve定量生理参数图(ve为组织间质占比)等，则模型训练完成后，处理得到的定量生理参数图与训练数据相适配。
优选地，本说明书实施例中，在步骤S110中的，获取与各组DCE-MRI图像对应的定量生理参数图之前，所述方法还包括：
对各组DCE-MRI图像进行对齐处理和/或配准处理。
如图4所示，对各组DCE-MRI图像进行对齐处理，可以包括如下步骤：
S410：选择该组DCE-MRI图像中的任意一幅作为参考图像，得到该组DCE-MRI图像中的其他图像与所述参考图像之间的偏差信息。
S420：根据所述偏差信息，将其他图像与所述参考图像对齐。
例如，可根据偏差信息，通过平移、旋转或缩放等变换操作将其他图像与参考图像对齐。由于DCE-MRI图像采集中，采集对象可能发生移动或由于呼吸等因素导致图像采集处的位置发生变化，从而造成输入数据不准确或者无法分析的问题，本说明书实施例中通过对各组DCE-MRI图像进行对齐处理进行齐处理可提高数据的准确性和可靠性。
如图5所示，对各组DCE-MRI图像进行配准处理，包括：
S510：选择该组DCE-MRI图像中的任意一幅作为参考图像，提取所述参考图像和该组DCE-MRI图像中的其他图像的同一特征点和/或同一特征区域。
特征点、特征区可以是DCE-MRI图像中具有显著性和独特性的对象，示例性的，可以是封闭边界区域、边缘区域、线交叉点等。选择合适的特征点和/或特征区域后，将其他图像中的特征点和/或特征区域与参考图像中的对应的特征点和/或特征区域相匹配，使其他图像与参考图像建立相关性。
S520：根据所述同一特征点和/或同一区域，确定其他图像与所述参考图像之间的相似性变换模型。
即确定出变换的范围和方式。
S530：根据确定出的相似性变化模型，将其他图像与所述参考图像进行配准。
需要说明的是，本说明书实施例中是对同一组DCE-MRI图像中的前后两幅(或多幅)图像进行配准处理。在一些其他的实施例中，还可以是对属于同一对象同一处组织但属于不同组的DCE-MRI图像中的两幅(或多幅)图像进行配准。也可以是同一对象同一处组织的不同模态图像间的配准等，例如可以将CT图像与DCE-MRI图像进行配准。
配准处理可以将不同采集组或不同设备采集的图像进行对比，通过配准操作，可以将这些图像进行统一的坐标系转换，从而实现更准确的比较和分析。
在实际操作中，对齐处理和配准处理可以是结合使用的，本说明书实施例中，通过对采集的DCE-MRI图像进行对齐和配准处理操作，使得DCE-MRI图像之间以及DCE-MRI图像的局部感兴趣区之间的位置和方向均正确对齐，从而，有利于提高计算出Ktrans值的准确性，有利于为后续的数据分析和临床应用提供可靠的依据。
在一些可行的实施例中，在对各组DCE-MRI图像进行对齐处理和/或配准处理之后，还可以包括：
对对齐处理后的DCE-MRI图像和/或配准后的DCE-MRI图像进行插值操作，以生成与其具有相同像素间距和方向的新图像，即增加训练输入数据的数量。
综上所述，本说明书实施例提供的磁共振定量生理参数图生成方法，在模型训练阶段能够从大量的训练数据中学习和识别复杂的特征和模式，不需要基于预设的血流动力学模型进行建模，从而避免了对血管连续性和流体不可压缩性等假设的依赖。且由于深度学习模型能够自动从数据中学习参数，因此，不需要手动选择多个参数来计算灌注参数，起到了减少人工干预的程度的作用，避免由于参数选择不当导致的计算结果不准确的问题。
当使用训练好的深度学习模型，即定量生理参数图生成模型来生成定量生理参数图时，能够避免大量的数学计算和分析，减少计算复杂性的同时，极大地提高了数据处理效率。并且训练完成后的模型，可降低对待处理的DCE-MRI图像的要求，对于单张DCE-MRI图像或少量DCE-MRI图像，定量生理参数图生成模型均可生成准确的结果。
如图6所示，本说明书实施例还提供一种磁共振定量生理参数图生成装置，包括：
获取模块61，用于获取各组DCE-MRI图像和与各DCE-MRI图像对应的定量生理参数图；
训练模块62，用于将所述DCE-MRI图像和所述定量生理参数图作为训练输入，对深度学习模型进行训练，直至所述深度学习模型的损失函数值满足预设要求，所述深度学习模型为生成对抗网络，所述生成对抗网络包括生成网络和判别网络，所述生成网络用于生成输出图像，所述判别网络包括全局判别器和局部判别器，所述全局判别器用于判别所述输出图像与所述定量生理参数图的相似性，所述局部判别器用于判别所述输出图像的局部感兴趣区与所述定量生理参数图中对应的局部感兴趣区的相似性
处理模块63，用于将所述损失函数值满足预设要求时的深度学习模型作定量生理参数图生成模型，对待处理DCE-MRI图像进行处理得到定量生理参数图。
通过本说明书实施例提供的装置所取得的有益效果和上述方法所取得的有益效果相一致，此处不再赘述。
如图7所示，为本说明书实施例提供的一种计算机设备，本说明书中磁共振定量生理参数图生成装置可以为本实施例中的计算机设备，执行本说明书实施例提供的磁共振定量生理参数图生成方法。所述计算机设备702可以包括一个或多个处理器704，诸如一个或多个中央处理单元(CPU)，每个处理单元可以实现一个或多个硬件线程。计算机设备702还可以包括任何存储器706，其用于存储诸如代码、设置、数据等之类的任何种类的信息。非限制性的，比如，存储器706可以包括以下任一项或多种组合：任何类型的RAM，任何类型的ROM，闪存设备，硬盘，光盘等。更一般地，任何存储器都 可以使用任何技术来存储信息。进一步地，任何存储器可以提供信息的易失性或非易失性保留。进一步地，任何存储器可以表示计算机设备702的固定或可移除部件。在一种情况下，当处理器704执行被存储在任何存储器或存储器的组合中的相关联的指令时，计算机设备702可以执行相关联指令的任一操作。计算机设备702还包括用于与任何存储器交互的一个或多个驱动机构708，诸如硬盘驱动机构、光盘驱动机构等。
计算机设备702还可以包括输入/输出模块710(I/O)，其用于接收各种输入(经由输入设备712)和用于提供各种输出(经由输出设备714)。一个具体输出机构可以包括呈现设备716和相关联的图形用户接口(GUI)718。在其他实施例中，还可以不包括输入/输出模块710(I/O)、输入设备712以及输出设备714，仅作为网络中的一台计算机设备。计算机设备702还可以包括一个或多个网络接口720，其用于经由一个或多个通信链路722与其他设备交换数据。一个或多个通信总线724将上文所描述的部件耦合在一起。
通信链路722可以以任何方式实现，例如，通过局域网、广域网(例如，因特网)、点对点连接等、或其任何组合。通信链路722可以包括由任何协议或协议组合支配的硬连线链路、无线链路、路由器、网关功能、名称服务器等的任何组合。
对应于图1、图3至图5中的方法，本说明书实施例还提供了一种计算机可读存储介质，该计算机可读存储介质上存储有计算机程序，该计算机程序被处理器运行时执行上述方法的步骤。
本说明书实施例还提供一种计算机可读指令，其中当处理器执行所述指令时，其中的程序使得处理器执行如图1、图3至图5所示的方法。
本说明书实施例还提供一种计算机程序产品，包括至少一条指令或至少一段程序，所述至少一条指令或所述至少一段程序由处理器加载并执行以实现如图1、图3至图5所示的方法。
应理解，在本说明书的各种实施例中，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，而不应对本说明书实施例的实施过程构成任何限定。
还应理解，在本说明书实施例中，术语“和/或”仅仅是一种描述关联对象的关联关系，表示可以存在三种关系。例如，A和/或B，可以表示：单独存在A，同时存在A和B，单独存在B这三种情况。另外，本说明书中字符“/”，一般表示前后关联对象是一种“或”的关系。
本领域普通技术人员可以意识到，结合本说明书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、计算机软件或者二者的结合来实现，为了清楚地说明硬件和软件的可互换性，在上述说明中已经按照功能一般性地描述了各示例的组成及步骤。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说明书实施例的范围。
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，上述描述的系统、装置和单元的具体工作过程，可以参考前述方法实施例中的对应过程，在此不再赘述。
在本说明书实施例所提供的几个实施例中，应该理解到，所揭露的系统、装置和方法，可以通过其它的方式实现。例如，以上所描述的装置实施例仅仅是示意性的，例如，所述单元的划分，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。另外，所显示或讨论的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、装置或单元的间接耦合或通信连接，也可以是电的，机械的或其它的形式连接。
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说明书实施例方案的目的。
另外，在本说明书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，也可以是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，本说明书实施例的技术方案本质上或者说对现有技术做出贡献的部分，或者该技术方案的全部或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计算机设备(可以是个人计算机，服务器，或者网络设备等)执行本说明书各个实施例所述方法的全部或部分步骤。而前述的存储介质包括：U盘、移动硬盘、只读存储器(ROM，Read-Only Memory)、随机存取存储器(RAM，Random Access Memory)、磁碟或者光盘等各种可以存储程序代码的介质。
本说明书中应用了具体实施例对本说明书的原理及实施方式进行了阐述，以上实施例的说明只是用于帮助理解本说明书的方法及其核心思想；同时，对于本领域的一般技术人员，依据本说明书的思想，在具体实施方式及应用范围上均会有改变之处，综上所述，本说明书内容不应理解为对本说明书实施例的限制。

Claims (10)

  1. 一种磁共振定量生理参数图生成方法，其特征在于，包括：
    获取各组DCE-MRI图像和与各组DCE-MRI图像对应的定量生理参数图；
    将所述各组DCE-MRI图像和所述定量生理参数图作为训练输入，对深度学习模型进行训练，直至所述深度学习模型的损失函数值满足预设要求，所述深度学习模型为生成对抗网络，所述生成对抗网络包括生成网络和判别网络，所述生成网络用于生成输出图像，所述判别网络包括全局判别器和局部判别器，所述全局判别器用于判别所述输出图像与所述定量生理参数图的相似性，所述局部判别器用于判别所述输出图像的局部感兴趣区与所述定量生理参数图中对应的局部感兴趣区的相似性；
    将所述损失函数值满足预设要求时的深度学习模型作为定量生理参数图生成模型，对待处理DCE-MRI图像进行处理得到定量生理参数图。
  2. 根据权利要求1所述的磁共振定量生理参数图生成方法，其特征在于，所述损失函数为：
    LOSS＝loss_G+loss_D；
    其中，LOSS为所述深度学习模型的损失值，loss_G为生成网络的损失值，loss_D为判别网络的损失值；
    loss_D＝loss_D1*α+loss_D2*(1-α)；
    其中，loss_D1为全局判别器的损失值，loss_D2为局部判别器的损失值，α为大于0小于1的常系数。
  3. 根据权利要求2所述的磁共振定量生理参数图生成方法，其特征在于，所述生成网络为加入组归一化的生成网络，所述判别网络为加入谱归一化的判别网络。
  4. 根据权利要求1所述的磁共振定量生理参数图生成方法，其特征在于，获取与各组DCE-MRI图像对应的定量生理参数图，包括：
    提取同一组DCE-MRI图像中的任一像素点的时间-信号曲线；
    建立该组DCE-MRI图像中的相应像素点的对比剂浓度随时间的变化曲线；
    拟合所述时间-信号曲线与所述变化曲线，得到该组DCE-MRI图像中对应像素点的 Ktrans值；
    重复上述过程，得到同一组DCE-MRI图像中所有像素点的Ktrans值；
    基于所有像素点的Ktrans值得到该组DCE-MRI图像对应的Ktrans定量生理参数图。
  5. 根据权利要求1所述的磁共振定量生理参数图生成方法，其特征在于，在获取与各组DCE-MRI图像对应的定量生理参数图之前，所述方法还包括：
    对各组DCE-MRI图像进行对齐处理和/或配准处理。
  6. 根据权利要求5所述的磁共振定量生理参数图生成方法，其特征在于，对各组DCE-MRI图像进行对齐处理，包括：
    选择该组DCE-MRI图像中的任意一幅作为参考图像，得到该组DCE-MRI图像中的其他图像与所述参考图像之间的偏差信息；
    根据所述偏差信息，将其他图像与所述参考图像对齐。
  7. 根据权利要求5所述的磁共振定量生理参数图生成方法，其特征在于，对各组DCE-MRI图像进行配准处理，包括：
    选择该组DCE-MRI图像中的任意一幅作为参考图像，提取所述参考图像和该组DCE-MRI图像中的其他图像的同一特征点和/或同一特征区域；
    根据所述同一特征点和/或同一区域，确定其他图像与所述参考图像之间的相似性变换模型；
    根据确定出的相似性变化模型，将其他图像与所述参考图像进行配准。
  8. 一种磁共振定量生理参数图生成装置，其特征在于，包括：
    获取模块，用于获取各组DCE-MRI图像和与各DCE-MRI图像对应的定量生理参数图；
    训练模块，用于将所述各组DCE-MRI图像和所述定量生理参数图作为训练输入，对深度学习模型进行训练，直至所述深度学习模型的损失函数值满足预设要求，所述深度学习模型为生成对抗网络，所述生成对抗网络包括生成网络和判别网络，所述生成网络用于生成输出图像，所述判别网络包括全局判别器和局部判别器，所述全局判别器用 于判别所述输出图像与所述定量生理参数图的相似性，所述局部判别器用于判别所述输出图像的局部感兴趣区与所述定量生理参数图中对应的局部感兴趣区的相似性；
    处理模块，用于将所述损失函数值满足预设要求时的深度学习模型作为定量生理参数图生成模型，对待处理DCE-MRI图像进行处理得到定量生理参数图。
  9. 一种计算机设备，包括存储器、处理器及存储在存储器上并可在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7任意一项所述的方法。
  10. 一种计算机可读存储介质，其特征在于，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权利要求1至7任意一项所述的方法。
PCT/CN2023/096606 2023-05-26 2023-05-26 一种磁共振定量生理参数图生成方法和装置 Ceased WO2024243733A1 (zh)

Priority Applications (1)

Application Number Priority Date Filing Date Title
PCT/CN2023/096606 WO2024243733A1 (zh) 2023-05-26 2023-05-26 一种磁共振定量生理参数图生成方法和装置

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
PCT/CN2023/096606 WO2024243733A1 (zh) 2023-05-26 2023-05-26 一种磁共振定量生理参数图生成方法和装置

Publications (1)

Publication Number Publication Date
WO2024243733A1 true WO2024243733A1 (zh) 2024-12-05

Family

ID=93656467

Family Applications (1)

Application Number Title Priority Date Filing Date
PCT/CN2023/096606 Ceased WO2024243733A1 (zh) 2023-05-26 2023-05-26 一种磁共振定量生理参数图生成方法和装置

Country Status (1)

Country Link
WO (1) WO2024243733A1 (zh)

Citations (6)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN110298897A (zh) * 2019-04-03 2019-10-01 上海联影医疗科技有限公司 用于正电子发射断层成像图像重建的系统和方法
CN113052937A (zh) * 2021-04-19 2021-06-29 厦门大学 一种动态对比增强核磁共振图像的重建方法及重建系统
EP3863022A1 (en) * 2020-02-06 2021-08-11 Siemens Healthcare GmbH Method and system for automatically characterizing liver tissue of a patient, computer program and electronically readable storage medium
CN113763442A (zh) * 2021-09-07 2021-12-07 南昌航空大学 一种可变形医学图像配准方法及系统
CN114424079A (zh) * 2019-09-20 2022-04-29 皇家飞利浦有限公司 使用mri的经运动校正的示踪剂动力学映射
WO2022217157A1 (en) * 2021-04-09 2022-10-13 The Regents Of The University Of California System and method for quantitative magnetic resonance imaging using a deep learning network

Patent Citations (6)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN110298897A (zh) * 2019-04-03 2019-10-01 上海联影医疗科技有限公司 用于正电子发射断层成像图像重建的系统和方法
CN114424079A (zh) * 2019-09-20 2022-04-29 皇家飞利浦有限公司 使用mri的经运动校正的示踪剂动力学映射
EP3863022A1 (en) * 2020-02-06 2021-08-11 Siemens Healthcare GmbH Method and system for automatically characterizing liver tissue of a patient, computer program and electronically readable storage medium
WO2022217157A1 (en) * 2021-04-09 2022-10-13 The Regents Of The University Of California System and method for quantitative magnetic resonance imaging using a deep learning network
CN113052937A (zh) * 2021-04-19 2021-06-29 厦门大学 一种动态对比增强核磁共振图像的重建方法及重建系统
CN113763442A (zh) * 2021-09-07 2021-12-07 南昌航空大学 一种可变形医学图像配准方法及系统

Similar Documents

Publication Publication Date Title
CN112885453B (zh) 用于标识后续医学图像中的病理变化的方法和系统
Jafari et al. Automatic biplane left ventricular ejection fraction estimation with mobile point-of-care ultrasound using multi-task learning and adversarial training
US10076299B2 (en) Systems and methods for determining hepatic function from liver scans
JP7205034B2 (ja) 磁気共鳴イメージで正中矢状平面を決定するための方法、イメージ処理デバイス及び格納媒体
US8068650B2 (en) Lesion quantification and tracking using multiple modalities
US20160283689A1 (en) Systems and methods for determination of blood flow characteristics and pathologies through modeling of myocardial blood supply
US8682051B2 (en) Smoothing of dynamic data sets
CN113989231B (zh) 动力学参数的确定方法、装置、计算机设备和存储介质
CN104346821A (zh) 用于医学成像的自动规划
CN117116485A (zh) 数据处理方法、装置、计算设备及存储介质
CN111476791A (zh) 图像处理方法、图像处理装置和非暂时性计算机可读介质
CN106709920B (zh) 血管提取方法及其装置
JP2021052999A (ja) 評価システム、評価方法、学習方法、学習済みモデル、プログラム
Gao et al. RETRACTED ARTICLE: Multimodal deep learning for cephalometric landmark detection and treatment prediction
CN116703842B (zh) 一种磁共振定量生理参数图生成方法和装置
JP2012085833A (ja) ３次元医用画像データの画像処理システム、その画像処理方法及びプログラム
KR102723565B1 (ko) 혈액 유속 결정 방법, 장치, 컴퓨팅 디바이스 및 저장 매체
CN113850794A (zh) 一种图像处理方法及装置
WO2024243733A1 (zh) 一种磁共振定量生理参数图生成方法和装置
JP5081224B2 (ja) 動的分子イメージング手順のため、運動パラメータ推定を区別する方法
CN114266759B (zh) 图像分析方法、装置及计算机设备
CN116779135A (zh) 用于计算血液储备分数的方法、装置、计算设备及介质
CN118196011B (zh) 血液流场信息确定方法、装置、计算设备及存储介质
CN119379556B (zh) 一种基于生成式人工智能技术的图像去噪方法与装置
CN114862981B (zh) 校正算法自动调用方法、装置、电子设备及存储介质

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: 23938732

Country of ref document: EP

Kind code of ref document: A1

NENP Non-entry into the national phase

Ref country code: DE