WO2020253138A1 - 一种分类方法、装置、设备和存储介质 - Google Patents

一种分类方法、装置、设备和存储介质 Download PDF

Info

Publication number
WO2020253138A1
WO2020253138A1 PCT/CN2019/123061 CN2019123061W WO2020253138A1 WO 2020253138 A1 WO2020253138 A1 WO 2020253138A1 CN 2019123061 W CN2019123061 W CN 2019123061W WO 2020253138 A1 WO2020253138 A1 WO 2020253138A1
Authority
WO
WIPO (PCT)
Prior art keywords
magnetic resonance
classification
classification model
target
space data
Prior art date
Application number
PCT/CN2019/123061
Other languages
English (en)
French (fr)
Inventor
梁栋
朱燕杰
程静
刘新
郑海荣
Original Assignee
深圳先进技术研究院
Priority date (The priority date is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the date listed.)
Filing date
Publication date
Application filed by 深圳先进技术研究院 filed Critical 深圳先进技术研究院
Publication of WO2020253138A1 publication Critical patent/WO2020253138A1/zh

Links

Images

Classifications

    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F18/00—Pattern recognition
    • G06F18/20—Analysing
    • G06F18/24—Classification techniques
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F18/00—Pattern recognition
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F18/00—Pattern recognition
    • G06F18/20—Analysing
    • G06F18/21—Design or setup of recognition systems or techniques; Extraction of features in feature space; Blind source separation
    • G06F18/214—Generating training patterns; Bootstrap methods, e.g. bagging or boosting
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
    • G06V2201/00—Indexing scheme relating to image or video recognition or understanding
    • G06V2201/03—Recognition of patterns in medical or anatomical images

Definitions

  • the embodiments of the present invention relate to the technical field of data processing, for example, to a classification method, device, device, and storage medium.
  • the process of using magnetic resonance data to classify target parts is: image preprocessing; region of interest segmentation; feature extraction, selection and classification.
  • preprocessing refers to correcting the image distortion caused by noise or motion artifacts, normalizing the image, and enhancing the display quality of the image such as denoising and increasing the contrast; after the preprocessing, the image is segmented and the feeling
  • the region of interest is separated from the background or surrounding tissues to reduce the interference of peripheral tissues or background to the detection of the region of interest, and reduce the amount of calculation; use algorithms to calculate various features of the region of interest, such as shape features, visual features, and density features, etc.
  • the feature dimension is large, the feature should be optimized; finally the target part is classified.
  • the method of acquiring the magnetic resonance image is a fast imaging method
  • the image needs to be reconstructed before the image preprocessing.
  • the above-mentioned reconstructed image has the problem of blurring, resulting in the loss of part of the detailed information, and this missing information is very important for the classification of the later target parts.
  • the segmentation of the region of interest on the image often leads to segmentation errors, and ultimately affects the classification of the target part.
  • the present application provides a classification method, device, equipment and storage medium, which improve the accuracy of classifying target parts.
  • an embodiment of the present invention provides a classification method, which includes:
  • an embodiment of the present invention also provides a classification device, which includes:
  • the magnetic resonance K-space data acquisition module is configured to acquire magnetic resonance K-space data corresponding to the target part of the current subject
  • the classification module is configured to input the magnetic resonance K-space data into the trained target classification model to obtain a first classification result corresponding to the target part, wherein the target classification model includes a model based on supervised learning The classification model trained in the same way.
  • an embodiment of the present invention also provides a classification device, the device including one or more processors;
  • Storage device for storing one or more programs
  • the one or more processors When the one or more programs are executed by the one or more processors, the one or more processors implement the classification method according to any embodiment of the present application.
  • an embodiment of the present invention also provides a computer-readable storage medium on which a computer program is stored, and when the computer program is executed by a processor, the classification method as described in any embodiment of the present application is implemented.
  • the embodiment of the present invention provides a classification method, device, equipment, and storage medium to obtain magnetic resonance k-space data corresponding to the target part of the current subject, and input the magnetic resonance k-space data into the training target
  • the classification model a first classification result corresponding to the target part is obtained
  • the target classification model includes a classification model trained based on a supervised learning method.
  • the above method uses the magnetic resonance K-space data as the target data and uses the trained classification model to classify the target part corresponding to the target data, which overcomes the inaccurate classification caused by the classification of the target part based on the magnetic resonance image in the related technology. The problem of improving the accuracy of classifying the target part.
  • Figure 1 is a flowchart of a classification method in Embodiment 1 of the present invention.
  • Figure 3 is a schematic structural diagram of a classification device in the third embodiment of the present invention.
  • Fig. 4 is a schematic structural diagram of a sorting device in the fourth embodiment of the present invention.
  • Fig. 1 is a flow chart of a classification method provided in the first embodiment of the present invention. This embodiment can be applied to the case of classifying target parts based on magnetic resonance technology. The method can be executed by a classification device or by a classification device. In software and/or hardware. As shown in Figure 1, the method of this embodiment may include:
  • the basic principle of magnetic resonance imaging is: put the scanned object in a magnetic field, and generate an excitation magnetic field by applying radio frequency pulses, so that the hydrogen protons in the target part of the scanned object resonate and produce energy level transitions. After the application of radio frequency pulses is stopped, The resonant hydrogen protons will release energy and return to the original energy level to generate a free induction signal.
  • the signal is encoded by an external gradient field, and the signal is received by the receiving coil. In actual situations, the signal is collected from a large number of spatial locations and consists of many complex frequencies. Using mathematical analysis methods, such as Fourier transform, the resonance frequency of each location can be obtained.
  • the object to be scanned is the subject in this embodiment, and the space of the collection point is the magnetic resonance k-space in this embodiment (magnetic resonance k-space is the dual space of the ordinary space under the Fourier transform).
  • the resonance frequency of the point is the magnetic resonance K-space data in this embodiment.
  • the subject may be a human body
  • the target part may be a human body part of interest, for example, a human body organ such as a liver, a heart, etc.
  • a human body organ may have a tumor, and correspondingly, the magnetic resonance K-space data may It is the magnetic resonance K-space data corresponding to the tumor.
  • the target classification model can be a classification model trained based on supervised learning, or a classification model trained based on unsupervised learning.
  • the target classification model can be a deep learning network model.
  • it can be a convolutional neural network model.
  • it can be a fully convolutional neural network model, U-Net model, V-Net model, ResNet model, DenseNet model, etc. .
  • the first classification result may be whether the target part is a part of interest.
  • the part of interest may be a liver part.
  • the first classification result may be that the part of interest is a liver part or a part of interest. Not the liver.
  • the first classification result can also be that the target site is a target site with a certain attribute.
  • the target site can be a tumor.
  • the first classification result can be that the target site is a benign tumor or the target site is a malignant tumor. Tumor etc.
  • the classification method provided in this embodiment obtains the magnetic resonance k-space data corresponding to the target part of the current subject, and inputs the magnetic resonance k-space data into the trained target classification model to obtain the corresponding target part
  • the above method uses the magnetic resonance K-space data as the target data and uses the trained classification model to classify the target part corresponding to the target data, which overcomes the inaccurate classification caused by the classification of the target part based on the magnetic resonance image in the related technology. The problem of improving the accuracy of classifying the target part.
  • acquiring magnetic resonance k-space data corresponding to the target part of the current subject includes:
  • the fast sampling method is used to obtain the magnetic resonance K-space data corresponding to the target part of the current subject, wherein the fast sampling method includes a variable density sampling method or a uniform sampling method.
  • the fast sampling method can be used to obtain the magnetic resonance K-space data corresponding to the target part of the current subject.
  • the fast sampling method can include a variable density sampling method or a uniform sampling method, wherein the variable density sampling method can include a variable density Random sampling method.
  • FIG. 2 is a flowchart of a classification method provided in Embodiment 2 of the present invention.
  • this embodiment may optionally further include: using zero padding or data retention to correspond to the magnetic resonance k-space data before inputting the magnetic resonance k-space data into the trained target classification model
  • the matrix size of the first matrix of is adjusted to a preset size to obtain a second matrix that can be processed by the classification model, and the data corresponding to the second matrix is used as the magnetic resonance k-space data of the input target classification model.
  • the optional data retention method is to start from the center data of the first matrix and sequentially retain the data in the rows and columns from the inside to the outside, where if the rows and columns of the first matrix are both odd numbers, then The center data of the first matrix is one. If the rows and/or columns of the first matrix are even numbers, the center data of the first matrix is even numbers.
  • the method of this embodiment may include:
  • the matrix size of the matrix corresponding to the magnetic resonance k-space data corresponding to different subjects and/or different parts can be adjusted, so that the magnetic resonance k-space data can be Target classification model processing.
  • the matrix size of the first matrix corresponding to the magnetic resonance k-space data can be adjusted to the preset size by means of zero padding. If the matrix size of the first matrix is greater than the preset size, the matrix size of the first matrix corresponding to the magnetic resonance K-space data can be adjusted to the preset size by data retention.
  • the data retention method may start from the center data of the first matrix and sequentially retain the data in the rows and columns from the inside to the outside, where if the rows and columns of the first matrix are both odd numbers, then the first matrix There is one center data of a matrix. If the rows and/or columns of the first matrix are even numbers, the center data of the first matrix are even numbers.
  • the classification method provided in this embodiment obtains the magnetic resonance K-space data corresponding to the target part of the current subject, and uses the zero padding method or the data retention method to calculate the matrix size of the first matrix corresponding to the magnetic resonance K-space data Adjust to the preset size to obtain the second matrix that can be processed by the classification model, and use the data corresponding to the second matrix as the magnetic resonance K-space data input to the target classification model, and input the magnetic resonance K-space data into the trained target classification
  • a first classification result corresponding to the target part is obtained, where the target classification model includes a classification model trained based on a supervised learning method.
  • the magnetic resonance K-space data is used as the target data, and the matrix size corresponding to the target data is adjusted to the matrix size that the target classification model can handle, and the trained classification model is used to classify the target part corresponding to the target data.
  • the problem of inaccurate classification caused by the classification of the target part based on the magnetic resonance image in the related technology is overcome, and the accuracy of the classification of the target part is improved.
  • the method before inputting the magnetic resonance K-space data into the trained target classification model, the method further includes:
  • the parameters of the classification model to be trained are adjusted based on the second classification result and the class label value to obtain the target classification model.
  • the historical subject is multiple patients with known benign or malignant tumors and the same tumor name.
  • the historical subject is a liver cancer patient, and the target site of the historical subject is the liver.
  • the historical magnetic resonance k-space data can be multi-channel magnetic resonance k-space data or single-channel magnetic resonance k-space data, which may also include the corresponding part Contrast parameters such as T1 weighting, T2 weighting, dispersion weighting and dynamic enhancement.
  • the multiple historical magnetic resonance K-space data can be marked for benign and malignant by using the category label value, wherein the benign corresponding
  • the category label value can be 0, and the category label value of the malignant correspondence can be 1, and the above-mentioned labeled data can be used to adjust the parameters of the classification model to be trained.
  • part of the data can also be selected from the labeled data as verification data to verify the accuracy of the model to be trained.
  • the pre-established classification model to be trained in this embodiment may be a convolutional neural network model, which is used to perform high-level feature extraction on input data.
  • the convolutional neural network model may include an input layer, multiple hidden layers, and an output layer.
  • the multiple hidden layers may include a convolutional layer, a nonlinear activation function layer, a pooling layer, a fully connected layer, and the like.
  • the size of the convolution kernel of the convolution layer can be set to 3*3.
  • the output layer will output an output result x.
  • the activation function can be used to perform nonlinear adjustments on the output result, and finally the second classification result is obtained.
  • the activation function may adopt any one of a ReLU function (Rectified Linear Unit, linear rectification function), a sigmoid function, and a tanh function.
  • the activation function may be a sigmoid function, which is defined as follows:
  • x is the output result of the output layer
  • f(x) is the second classification result
  • the parameter adjustment of the classification model to be trained based on the second classification result and the class label value to obtain the target classification model includes:
  • the stochastic gradient descent method is used to adjust the parameters of the classification model to be trained, and the parameter that minimizes the value of the cross-entropy loss function is used as the parameter corresponding to the target classification model to obtain the target classification model.
  • the loss function can be used to adjust the parameters of the classification model to be trained.
  • the parameter that minimizes the loss function is the best parameter of the classification model to be trained.
  • the loss function may be a cross-entropy loss function.
  • the cross-entropy loss function is defined as follows:
  • a stochastic gradient descent algorithm may be used to perform network training on the classification model to be trained to finally obtain the target classification model.
  • the fast sampling method can be used to obtain the liver magnetic resonance K-space data of the target liver cancer patient, and the zero-padding method or the data retention method can be used to adjust the matrix size of the matrix corresponding to the data. So that the data can be processed by the target classification model.
  • the adjusted data is used as the input of the target classification model and input into the target classification model.
  • the target classification model will output a value between 0 and 1 (including 0 or 1). If the output value is less than 0.5, then the tumor is benign, if the output value is greater than 0.5, then the tumor is malignant.
  • the method before inputting historical magnetic resonance K-space data into the pre-established classification model to be trained, the method further includes:
  • the matrix size of the matrix corresponding to the acquired historical magnetic resonance K-space data can be optionally adjusted.
  • the adjustment method may include a zero padding method or a data retention method.
  • Fig. 3 is a schematic structural diagram of a classification device in the third embodiment of the present invention. As shown in Figure 3, the device of this embodiment includes:
  • the magnetic resonance K-space data acquisition module 310 is configured to acquire magnetic resonance K-space data corresponding to the target part of the current subject
  • the classification module 320 is configured to input the magnetic resonance K-space data into the trained target classification model to obtain a first classification result corresponding to the target part, wherein the target classification model includes a classification trained based on supervised learning model.
  • the classification device uses the magnetic resonance K-space data acquisition module to acquire the magnetic resonance K-space data corresponding to the target part of the current subject, and uses the classification module to input the magnetic resonance K-space data to the training completion In the target classification model of, the first classification result corresponding to the target part is obtained, where the target classification model includes a classification model trained based on supervised learning.
  • the above device uses the magnetic resonance K-space data as the target data, and uses the trained classification model to classify the target part corresponding to the target data, which overcomes the inaccurate classification caused by the classification of the target part based on the magnetic resonance image in the related technology. The problem of improving the accuracy of classifying the target part.
  • the magnetic resonance K-space data acquisition module 310 can be used to:
  • the fast sampling method is used to obtain the magnetic resonance K-space data corresponding to the target part of the current subject, wherein the fast sampling method includes a variable density sampling method or a uniform sampling method.
  • the classification device may further include a first matrix size adjustment module, which may be used for:
  • the matrix size of the first matrix corresponding to the magnetic resonance k-space data Before inputting the magnetic resonance k-space data into the trained target classification model, adjust the matrix size of the first matrix corresponding to the magnetic resonance k-space data to the preset size by zero-padding or data retention to obtain the target
  • the second matrix processed by the classification model, and the data corresponding to the second matrix is used as the magnetic resonance K-space data of the input target classification model.
  • the data retention method is to start from the center data of the first matrix and sequentially retain the data in the rows and columns from the inside to the outside. Wherein, if the rows and columns of the first matrix are both odd numbers, the first The center data of the matrix is one. If the rows and/or columns of the first matrix are even numbers, the center data of the first matrix is even numbers.
  • the classification device may further include:
  • the historical data acquisition and labeling module is used to obtain historical magnetic resonance K-space data corresponding to the target part of the historical subject before inputting the magnetic resonance K-space data into the trained target classification model, and use the category label value Perform category labeling on historical magnetic resonance K-space data;
  • the second classification result determination module is used to input historical magnetic resonance K-space data into the pre-established classification model to be trained to obtain the second classification result;
  • the target classification model determination module is used to adjust the parameters of the classification model to be trained based on the second classification result and the class label value to obtain the target classification model.
  • the target classification model determination module can be used to:
  • the stochastic gradient descent method is used to adjust the parameters of the classification model to be trained, and the parameter that minimizes the value of the cross-entropy loss function is used as the parameter corresponding to the target classification model to obtain the target classification model.
  • the classification device may further include a second matrix size adjustment module, which may be used for:
  • the matrix size of the third matrix corresponding to the historical magnetic resonance K-space data is adjusted to the preset size by means of zero padding or data retention, to obtain the first Four matrices, and the data corresponding to the fourth matrix is used as the historical magnetic resonance K-space data input to the classification model to be trained.
  • the classification device provided in the embodiment of the present invention can execute the classification method provided in any embodiment of the present application, and has the functional modules and beneficial effects corresponding to the execution method.
  • Fig. 4 is a schematic structural diagram of a classification device provided in Embodiment 4 of the present invention.
  • Fig. 4 shows a block diagram of an exemplary classification device 412 suitable for implementing embodiments of the present application.
  • the classification device 412 shown in FIG. 4 is only an example.
  • the classification device 412 is in the form of a general-purpose computing device.
  • the components of the classification device 412 may include: one or more processors 416, a memory 428, and a bus 418 connecting different system components (including the memory 428 and the processor 416).
  • the bus 418 represents one or more of several types of bus structures, including a memory bus or a memory controller, a peripheral bus, a graphics acceleration port, a processor, or a local bus using any bus structure among multiple bus structures.
  • these architectures can include industry standard architecture (ISA) bus, microchannel architecture (MAC) bus, enhanced ISA bus, Video Electronics Standards Association (VESA) local bus, and peripheral component interconnect (PCI) bus.
  • ISA industry standard architecture
  • MAC microchannel architecture
  • VESA Video Electronics Standards Association
  • PCI peripheral component interconnect
  • the classification device 412 typically includes a variety of computer system readable media. These media can be any available media that can be accessed by the classification device 412, including volatile and non-volatile media, removable and non-removable media.
  • the memory 428 may include a computer system readable medium in the form of a volatile memory, such as random access memory (RAM) 430 and/or cache memory 432.
  • the classification device 412 may include other removable/non-removable, volatile/non-volatile computer system storage media.
  • the storage device 434 may be used to read and write non-removable, nonvolatile magnetic media (commonly referred to as "hard drives").
  • a disk drive for reading and writing to removable non-volatile disks such as "floppy disks”
  • a removable non-volatile disk such as CD-ROM, DVD-ROM or other optical media
  • each drive may be connected to the bus 418 through one or more data media interfaces.
  • the memory 428 may include at least one program product, and the program product has a set of (for example, at least one) program modules, which are configured to perform the functions of the embodiments of the present application.
  • a program/utility tool 440 having a set of (at least one) program module 442 may be stored in, for example, the memory 428.
  • Such program module 442 may include an operating system, one or more application programs, other program modules, and program data. Each of the examples or some combination may include the realization of a network environment.
  • the program module 442 generally executes the functions and/or methods in the embodiments described in this application.
  • the classification device 412 can also communicate with one or more external devices 414 (such as a keyboard, a pointing device, a display 424, etc., where the display 424 can be configured according to actual needs), and can also communicate with one or more so that the user can communicate with the
  • the classification device 412 communicates with the device that it interacts with, and/or communicates with any device (such as a network card, modem, etc.) that enables the classification device 412 to communicate with one or more other computing devices. This communication can be performed through an input/output (I/O) interface 422.
  • I/O input/output
  • the classification device 412 may also communicate with one or more networks (such as a local area network (LAN), a wide area network (WAN), and/or a public network, such as the Internet) through the network adapter 420. As shown in the figure, the network adapter 420 communicates with other modules of the classification device 412 through the bus 418. It should be understood that other hardware and/or software modules may be used in conjunction with the classification device 412, which may include: microcode, device drivers, redundant processing units, external disk drive arrays, RAID systems, tape drives, and data backup storage devices.
  • the processor 416 executes various functional applications and data processing by running programs stored in the memory 428, for example, implements the classification method provided by the embodiment of the present invention.
  • the fifth embodiment of the present invention provides a computer-readable storage medium on which a computer program is stored.
  • the program is executed by a processor, the classification method as provided in the embodiment of the present invention is implemented, including:
  • the magnetic resonance K-space data is input into the trained target classification model to obtain a first classification result corresponding to the target part.
  • the target classification model includes a classification model trained based on a supervised learning method.
  • the computer-readable storage medium provided in the embodiment of the present invention and the computer program stored thereon can also perform related operations in the classification method based on the classification device provided in any embodiment of the present application.
  • the computer storage medium of the embodiment of the present invention may adopt any combination of one or more computer-readable media.
  • the computer-readable medium may be a computer-readable signal medium or a computer-readable storage medium.
  • the computer-readable storage medium may be, for example, an electrical, magnetic, optical, electromagnetic, infrared, or semiconductor system, apparatus, or device, or a combination of any of the above.
  • Examples of computer-readable storage media may include: electrical connections with one or more wires, portable computer disks, hard disks, random access memory (RAM), read-only memory (ROM), erasable Type programmable read only memory (EPROM or flash memory), optical fiber, portable compact disk read only memory (CD-ROM), optical storage device, magnetic storage device, or any suitable combination of the above.
  • the computer-readable storage medium can be any tangible medium that contains or stores a program, and the program can be used by or in combination with an instruction execution system, apparatus, or device.
  • the computer-readable signal medium may include a data signal propagated in baseband or as a part of a carrier wave, and computer-readable program code is carried therein. This propagated data signal can take many forms, including electromagnetic signals, optical signals, or any suitable combination of the foregoing.
  • the computer-readable signal medium may also be any computer-readable medium other than the computer-readable storage medium.
  • the computer-readable medium may send, propagate, or transmit the program for use by or in combination with the instruction execution system, apparatus, or device .
  • the program code contained on the computer-readable medium can be transmitted by any suitable medium, which can include wireless, wire, optical cable, RF, etc., or any suitable combination of the above.
  • the computer program code used to perform the operations of the present application can be written in one or more programming languages or a combination thereof.
  • the programming languages include object-oriented programming languages-such as Java, Smalltalk, C++, and also conventional Procedural programming language-such as "C" language or similar programming language.
  • the program code can be executed entirely on the user's computer, partly on the user's computer, executed as an independent software package, partly on the user's computer and partly executed on a remote computer, or entirely executed on the remote computer or server.
  • the remote computer can be connected to the user’s computer through any kind of network, including a local area network (LAN) or a wide area network (WAN), or it can be connected to an external computer (for example, using an Internet service provider to pass Internet connection).
  • LAN local area network
  • WAN wide area network
  • Internet service provider for example, using an Internet service provider to pass Internet connection.

Landscapes

  • Engineering & Computer Science (AREA)
  • Theoretical Computer Science (AREA)
  • Data Mining & Analysis (AREA)
  • Computer Vision & Pattern Recognition (AREA)
  • Bioinformatics & Cheminformatics (AREA)
  • Bioinformatics & Computational Biology (AREA)
  • Artificial Intelligence (AREA)
  • Evolutionary Biology (AREA)
  • Evolutionary Computation (AREA)
  • Physics & Mathematics (AREA)
  • General Engineering & Computer Science (AREA)
  • General Physics & Mathematics (AREA)
  • Life Sciences & Earth Sciences (AREA)
  • Magnetic Resonance Imaging Apparatus (AREA)

Abstract

一种分类方法、装置、设备和存储介质，其中，分类方法包括：获取与当前被检体的目标部位相对应的磁共振K空间数据（S110）；将磁共振K空间数据输入到训练完成的目标分类模型中，得到与目标部位相对应的第一分类结果，其中，目标分类模型包括基于有监督学习的方式训练的分类模型（S120）。此方法提高了对目标部位分类的准确度。

Description

一种分类方法、装置、设备和存储介质
本公开要求在2019年06月18日提交中国专利局、申请号为201910527343.6的中国专利申请的优先权，以上申请的全部内容通过引用结合在本公开中。
技术领域
本发明实施例涉及数据处理技术领域，例如涉及一种分类方法、装置、设备和存储介质。
背景技术
医学磁共振由于能提供多种对比度信息，良好的软组织成像能力以及无电离辐射等特性，越来越多的应用于临床检查中。
相关技术中，利用磁共振数据对目标部位进行分类的流程为：图像预处理；感兴趣区域分割；特征提取、选择与分类。其中，预处理是指校正由于噪声或运动伪影而导致的图像失真，对图像做归一化处理，以及去噪和增加对比度等增强图像的显示质量；预处理之后，进行图像分割，把感兴趣区域从背景或周围组织中分离出来以减少外围组织或背景对感兴趣区域检测的干扰，减少计算量；利用算法计算感兴趣区域的各种特征，如形状特征、视觉特征和密度特征等，当特征维数较多时要对特征做优化选择；最终对目标部位进行分类。上述如果获取磁共振图像的方式为快速成像方法，则在图像预处理之前，还需要对图像进行重建。上述重建图像存在模糊的问题，导致部分细节信息丢失，而这些丢失的信息对后期目标部位的分类非常重要。此外，对图像进行感兴趣区域分割往往也会导致分割误差，最终也会对目标部位的分类造成影响。
发明内容
本申请提供一种分类方法、装置、设备和存储介质，提高了对目标部位分类的准确度。
第一方面，本发明实施例提供了一种分类方法，所述方法包括：
获取与当前被检体的目标部位相对应的磁共振K空间数据；
将所述磁共振K空间数据输入到训练完成的目标分类模型中，得到与所述目标部位相对应的第一分类结果，其中，所述目标分类模型包括基于有监督学习的方式训练的分类模型。
第二方面，本发明实施例还提供了一种分类装置，所述装置包括：
磁共振K空间数据获取模块，被配置为获取与当前被检体的目标部位相对应的磁共振K空间数据；
分类模块，被配置为将所述磁共振K空间数据输入到训练完成的目标分类模型中，得到与所述目标部位相对应的第一分类结果，其中，所述目标分类模型包括基于有监督学习的方式训练的分类模型。
第三方面，本发明实施例还提供了一种分类设备，所述设备包括一个或多个处理器；
存储装置，用于存储一个或多个程序，
当所述一个或多个程序被所述一个或多个处理器执行，使得所述一个或多个处理器实现如本申请任一实施例所述的分类方法。
第四方面，本发明实施例还提供了一种计算机可读存储介质，其上存储有计算机程序，所述计算机程序被处理器执行时实现如本申请任一实施例所述的分类方法。
本发明实施例提供了一种分类方法、装置、设备和存储介质，获取与当前被检体的目标部位相对应的磁共振K空间数据，将所述磁共振K空间数据输入到训练完成的目标分类模型中，得到与所述目标部位相对应的第一分类结果，其中，所述目标分类模型包括基于有监督学习的方式训练的分类模型。上述方法通过将磁共振K空间数据作为目标数据，利用训练完成的分类模型对目标数据对应的目标部位进行分类，克服了相关技术中基于磁共振图像对目标部位进行分类，所造成的分类不准确的问题，提高了对目标部位分类的准确度。
附图说明
下面将对实施例或相关技术描述中所需要使用的附图做一简单地介绍，显而易见地，下面描述中的附图是本申请的一些实施例，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是本发明实施例一中的一种分类方法的流程图；
图2是本发明实施例二中的一种分类方法的流程图；
图3是本发明实施例三中的一种分类装置的结构示意图；
图4是本发明实施例四中的一种分类设备的结构示意图。
具体实施方式
下面结合附图和实施例对本申请作进一步的详细说明。可以理解的是，此处所描述的具体实施例仅仅用于解释本申请，而非对本申请的限定。另外还需要说明的是，为了便于描述，附图中仅示出了与本申请相关的部分而非全部结构。
实施例一
图1为本发明实施例一提供的一种分类方法的流程图，本实施例可适用于基于磁共振技术，对目标部位进行分类的情况，该方法可以由分类装置来执行，可以通过分类装置中的软件和/或硬件来实施。如图1所示，本实施例的方法可以包括：
S110、获取与当前被检体的目标部位相对应的磁共振K空间数据。
磁共振成像的基本原理为：将被扫描物体放入磁场中，通过施加射频脉冲产生激励磁场，使得被扫描物体的目标部位内的氢质子发生共振，产生能级跃迁，停止施加射频脉冲后，共振的氢质子会释放能量，回到原来的能级，从而产生自由感应信号，通过外加梯度场对信号进行编码，并用接收线圈接收信号。实际情况下，信号是从大量空间位置点收集的，由许多复合频率组成，利用数学分析方法，如傅里叶变换，可求出各位置点的共振频率。上述被扫描物体为本实施例中的被检体，上述收集位置点的空间为本实施例中的磁共振K空间(磁共振k空间是寻常空间在傅利叶转换下的对偶空间)，上述各位置点的共振频率为本实施例中的磁共振K空间数据。
示例性的，被检体可以是人体，目标部位可以是感兴趣的人体部位，例如可以是人体内的肝脏、心脏等人体器官，人体器官可能病变有肿瘤，相应的，磁共振K空间数据可以是肿瘤对应的磁共振K空间数据。
S120、将磁共振K空间数据输入到训练完成的目标分类模型中，得到与目标部位相对应的第一分类结果，其中，目标分类模型包括基于有监督学习的方式训练的分类模型。
目标分类模型可以是基于有监督学习的方式训练的分类模型，也可以是基于无监督学习的方式训练的分类模型。目标分类模型可以是深度学习网络模型，在一些实施例中，可以是卷积神经网络模型，例如可以是全卷积神经网络模型、U-Net模型、V-Net模型、ResNet模型和DenseNet模型等。
其中，第一分类结果可以是目标部位是否是感兴趣部位，示例性的，感兴 趣部位可以是肝脏部位，相应的，第一分类结果可以是感兴趣部位是肝脏部位，也可以是感兴趣部位不是肝脏部位。第一分类结果还可以是目标部位是具有某一属性的目标部位，示例性的，目标部位可以是肿瘤，相应的，第一分类结果可以是目标部位是良性肿瘤，也可以是目标部位是恶性肿瘤等。
本实施例提供的一种分类方法，获取与当前被检体的目标部位相对应的磁共振K空间数据，将磁共振K空间数据输入到训练完成的目标分类模型中，得到与目标部位相对应的第一分类结果，其中，目标分类模型包括基于有监督学习的方式训练的分类模型。上述方法通过将磁共振K空间数据作为目标数据，利用训练完成的分类模型对目标数据对应的目标部位进行分类，克服了相关技术中基于磁共振图像对目标部位进行分类，所造成的分类不准确的问题，提高了对目标部位分类的准确度。
在上述各实施例的基础上，在一些实施例中，获取与当前被检体的目标部位相对应的磁共振K空间数据，包括：
利用快速采样方法获取与当前被检体的目标部位相对应的磁共振K空间数据，其中，快速采样方法包括变密度采样方法或均匀采样方法。
在磁共振数据采集的过程中，由于受奈奎斯特采样定理的限制，磁共振采样速度非常慢，极大地降低了采样效率；此外，被检体在较长扫描时间段内可能发生轻微移动，造成采集的数据中不准确，进而导致整个扫描过程失败。因此，可以利用快速采样方法获取与当前被检体的目标部位相对应的磁共振K空间数据，其中，快速采样方法可以包括变密度采样方法或均匀采样方法，其中变密度采样方法可以包括变密度随机采样方法。
实施例二
图2为本发明实施例二提供的一种分类方法的流程图。本实施例在上述各实施例的基础上，可选在将磁共振K空间数据输入到训练完成的目标分类模型中之前，还包括：利用补零方式或数据保留方式将磁共振K空间数据对应的第一矩阵的矩阵大小调整为预设大小，得到能够被分类模型处理的第二矩阵，并将第二矩阵对应的数据作为输入目标分类模型的磁共振K空间数据。在一些实施例中，可选数据保留方式为从第一矩阵的中心数据开始，从内向外依次对行和列中的数据进行保留，其中，若第一矩阵的行和列均为奇数，则第一矩阵的中心数据为一个，若第一矩阵的行和/或列为偶数，则第一矩阵的中心数据为偶 数个。如图2所示，本实施例的方法可以包括：
S210、获取与当前被检体的目标部位相对应的磁共振K空间数据。
S220、利用补零方式或数据保留方式将磁共振K空间数据对应的第一矩阵的矩阵大小调整为预设大小，得到能够被目标分类模型处理的第二矩阵，并将第二矩阵对应的数据作为输入目标分类模型的磁共振K空间数据。
本实施例中，为了保证分类结果的准确性，可以对不同被检体和/或不同部位对应的磁共振K空间数据所对应的矩阵的矩阵大小进行调整，以使磁共振K空间数据能够被目标分类模型处理。
在一些实施例中，如果第一矩阵的矩阵大小小于预设大小，则可以通过补零方式将磁共振K空间数据对应的第一矩阵的矩阵大小调整为预设大小。如果第一矩阵的矩阵大小大于预设大小，则可以通过数据保留方式将磁共振K空间数据对应的第一矩阵的矩阵大小调整为预设大小。
在一些实施例中，数据保留方式可以是从第一矩阵的中心数据开始，从内向外依次对行和列中的数据进行保留，其中，若第一矩阵的行和列均为奇数，则第一矩阵的中心数据为一个，若第一矩阵的行和/或列为偶数，则第一矩阵的中心数据为偶数个。
S230、将磁共振K空间数据输入到训练完成的目标分类模型中，得到与目标部位相对应的第一分类结果，其中，目标分类模型包括基于有监督学习的方式训练的分类模型。
本实施例提供的一种分类方法，获取与当前被检体的目标部位相对应的磁共振K空间数据，利用补零方式或数据保留方式将磁共振K空间数据对应的第一矩阵的矩阵大小调整为预设大小，得到能够被分类模型处理的第二矩阵，并将第二矩阵对应的数据作为输入目标分类模型的磁共振K空间数据，将磁共振K空间数据输入到训练完成的目标分类模型中，得到与目标部位相对应的第一分类结果，其中，目标分类模型包括基于有监督学习的方式训练的分类模型。上述方法通过将磁共振K空间数据作为目标数据，同时将目标数据对应的矩阵大小调整为目标分类模型所能处理的矩阵大小，并利用训练完成的分类模型对目标数据对应的目标部位进行分类，克服了相关技术中基于磁共振图像对目标部位进行分类，所造成的分类不准确的问题，提高了对目标部位分类的准确度。
在上述各实施例的基础上，在一些实施例中，在将磁共振K空间数据输入到训练完成的目标分类模型中之前，还包括：
获取历史被检体的目标部位相对应的历史磁共振K空间数据，并利用类别标记值对历史磁共振K空间数据进行类别标记；
将历史磁共振K空间数据输入预先建立的待训练分类模型中，得到第二分类结果；
基于第二分类结果和类别标记值对待训练分类模型进行参数调整，得到目标分类模型。
示例性的，历史被检体为多名已知肿瘤良恶性，且肿瘤名称相同的患者，例如历史被检体为肝癌患者，历史被检体的目标部位为肝部，相应的，获取多个历史被检体的肝部对应的历史磁共振K空间数据，该历史磁共振K空间数据可以是多通道磁共振K空间数据，也可以单通道磁共振K空间数据，其还可以包括该部位对应的T1加权、T2加权、弥散加权和动态增强等多个对比度参数。
在获取多个历史被检体的肿瘤部位对应的历史磁共振K空间数据之后，可选的，可以利用类别标记值对该多个历史磁共振K空间数据进行良恶性标记，其中，良性对应的类别标记值可以为0，恶性对应的类别标记值可以为1，上述标记后的数据可以用于调整待训练分类模型的参数。此外，也可以从上述标记后的数据中选取部分数据作为验证数据，用于验证待训练模型的准确性。
本实施例中的预先建立的待训练分类模型可以是卷积神经网络模型，该模型用于对输入数据进行高层次的特征提取。该卷积神经网络模型可以包括输入层、多个隐含层和输出层，其中，多个隐含层可以包括卷积层、非线性激活函数层、池化层以及全连接层等。可选的，卷积层的卷积核的大小可以设置为3*3。
可选的，在将历史磁共振K空间数据输入预先建立的待训练分类模型中之后，输出层会输出一个输出结果x。可以利用激活函数对该输出结果进行非线性调整，最终得到第二分类结果。其中，激活函数可以采用ReLU函数(Rectified Linear Unit，线性整流函数)、sigmoid函数和tanh函数中的任一种。示例性的，激活函数可以是sigmoid函数，其定义如下：
Figure PCTCN2019123061-appb-000001
其中，x为输出层的输出结果，f(x)为第二分类结果。
在一些实施例中，基于第二分类结果和类别标记值对待训练分类模型进行参数调整，得到目标分类模型，包括：
将第二分类结果和类别标记值代入交叉熵损失函数中；
利用随机梯度下降法调节待训练分类模型的参数，将使交叉熵损失函数的值最小的参数作为目标分类模型所对应的参数，得到目标分类模型。
可选的，可以利用损失函数调节待训练分类模型的参数，在一些实施例中，使损失函数最小的参数为待训练分类模型的最佳参数。示例性的，损失函数可以是交叉熵损失函数。其中，交叉熵损失函数的定义如下：
Figure PCTCN2019123061-appb-000002
其中，
Figure PCTCN2019123061-appb-000003
为分类标记值，y i为第二分类结果。
本实施例中，可选的，可以采用随机梯度下降算法对待训练分类模型进行网络训练，最终得到目标分类模型。示例性的，在得到目标分类模型之后，可以利用快速采样方法获取目标肝癌患者的肝磁共振K空间数据，并利用补零方式或数据保留方式，对该数据对应的矩阵的矩阵大小进行调整，以使该数据能够被目标分类模型处理。将调整后的数据作为目标分类模型的输入，输入至目标分类模型中，目标分类模型会输出一个0到1之间(包含0或者1)的值。如果输出的值小于0.5，则该肿瘤为良性，如果输出的值大于0.5，则该肿瘤为恶性肿瘤。
在一些实施例中，在将历史磁共振K空间数据输入预先建立的待训练分类模型中之前，还包括：
利用补零方式或数据保留方式将历史磁共振K空间数据对应的第三矩阵的矩阵大小调整为预设大小，得到第四矩阵，并将第四矩阵对应的数据作为输入待训练分类模型的历史磁共振K空间数据。
为了提高待训练分类模型的训练效率和准确性，可选可以对获取到的历史磁共振K空间数据所对应的矩阵的矩阵大小进行调整。调整方式可以包括补零方式或数据保留方式，上述两种方式在模型训练中的执行过程与利用训练好的模型确定目标部位的分类结果中的执行过程相同，在此不再赘述。
实施例三
图3是本发明实施例三中的一种分类装置的结构示意图。如图3所示，本实施例的装置包括：
磁共振K空间数据获取模块310，被配置为获取与当前被检体的目标部位相对应的磁共振K空间数据；
分类模块320，被配置为将磁共振K空间数据输入到训练完成的目标分类模型中，得到与目标部位相对应的第一分类结果，其中，目标分类模型包括基于有监督学习的方式训练的分类模型。
本实施例提供的一种分类装置，利用磁共振K空间数据获取模块获取与当前被检体的目标部位相对应的磁共振K空间数据，并利用分类模块将磁共振K空间数据输入到训练完成的目标分类模型中，得到与目标部位相对应的第一分类结果，其中，目标分类模型包括基于有监督学习的方式训练的分类模型。上述装置通过将磁共振K空间数据作为目标数据，利用训练完成的分类模型对目标数据对应的目标部位进行分类，克服了相关技术中基于磁共振图像对目标部位进行分类，所造成的分类不准确的问题，提高了对目标部位分类的准确度。
在上述各实施例的基础上，在一些实施例中，磁共振K空间数据获取模块310可用于：
利用快速采样方法获取与当前被检体的目标部位相对应的磁共振K空间数据，其中，快速采样方法包括变密度采样方法或均匀采样方法。
在一些实施例中，该分类装置还可以包括第一矩阵大小调整模块，该模块可用于：
在将磁共振K空间数据输入到训练完成的目标分类模型中之前，利用补零方式或数据保留方式将磁共振K空间数据对应的第一矩阵的矩阵大小调整为预设大小，得到能够被目标分类模型处理的第二矩阵，并将第二矩阵对应的数据作为输入目标分类模型的磁共振K空间数据。
在一些实施例中，数据保留方式为从第一矩阵的中心数据开始，从内向外依次对行和列中的数据进行保留，其中，若第一矩阵的行和列均为奇数，则第一矩阵的中心数据为一个，若第一矩阵的行和/或列为偶数，则第一矩阵的中心数据为偶数个。
在一些实施例中，该分类装置还可以包括：
历史数据获取及标记模块，用于在将磁共振K空间数据输入到训练完成的 目标分类模型中之前，获取历史被检体的目标部位相对应的历史磁共振K空间数据，并利用类别标记值对历史磁共振K空间数据进行类别标记；
第二分类结果确定模块，用于将历史磁共振K空间数据输入预先建立的待训练分类模型中，得到第二分类结果；
目标分类模型确定模块，用于基于第二分类结果和类别标记值对待训练分类模型进行参数调整，得到目标分类模型。
在一些实施例中，目标分类模型确定模块可用于：
将第二分类结果和类别标记值代入交叉熵损失函数中；
利用随机梯度下降法调节待训练分类模型的参数，将使交叉熵损失函数的值最小的参数作为目标分类模型所对应的参数，得到目标分类模型。
在一些实施例中，该分类装置还可以包括第二矩阵大小调整模块，该模块可用于：
在将历史磁共振K空间数据输入预先建立的待训练分类模型中之前，利用补零方式或数据保留方式将历史磁共振K空间数据对应的第三矩阵的矩阵大小调整为预设大小，得到第四矩阵，并将第四矩阵对应的数据作为输入待训练分类模型的历史磁共振K空间数据。
本发明实施例所提供的分类装置可执行本申请任意实施例所提供的分类方法，具备执行方法相应的功能模块和有益效果。
实施例四
图4为本发明实施例四提供的分类设备的结构示意图。图4示出了适于用来实现本申请实施方式的示例性分类设备412的框图。图4显示的分类设备412仅仅是一个示例。
如图4所示，分类设备412以通用计算设备的形式表现。分类设备412的组件可以包括：一个或者多个处理器416，存储器428，连接不同系统组件(包括存储器428和处理器416)的总线418。
总线418表示几类总线结构中的一种或多种，包括存储器总线或者存储器控制器，外围总线，图形加速端口，处理器或者使用多种总线结构中的任意总线结构的局域总线。举例来说，这些体系结构可以包括工业标准体系结构(ISA)总线，微通道体系结构(MAC)总线，增强型ISA总线、视频电子标准协会(VESA)局域总线以及外围组件互连(PCI)总线。
分类设备412典型地包括多种计算机系统可读介质。这些介质可以是任何能够被分类设备412访问的可用介质，包括易失性和非易失性介质，可移动的和不可移动的介质。
存储器428可以包括易失性存储器形式的计算机系统可读介质，例如随机存取存储器(RAM)430和/或高速缓存存储器432。分类设备412可以包括其它可移动/不可移动的、易失性/非易失性计算机系统存储介质。仅作为举例，存储装置434可以用于读写不可移动的、非易失性磁介质(通常称为“硬盘驱动器”)。在实现中，可以提供用于对可移动非易失性磁盘(例如“软盘”)读写的磁盘驱动器，以及对可移动非易失性光盘(例如CD-ROM,DVD-ROM或者其它光介质)读写的光盘驱动器。在这些情况下，每个驱动器可以通过一个或者多个数据介质接口与总线418相连。存储器428可以包括至少一个程序产品，该程序产品具有一组(例如至少一个)程序模块，这些程序模块被配置以执行本申请各实施例的功能。
具有一组(至少一个)程序模块442的程序/实用工具440，可以存储在例如存储器428中，这样的程序模块442可以包括操作系统、一个或者多个应用程序、其它程序模块以及程序数据，这些示例中的每一个或某种组合中可能包括网络环境的实现。程序模块442通常执行本申请所描述的实施例中的功能和/或方法。
分类设备412也可以与一个或多个外部设备414(例如键盘、指向设备、显示器424等，其中，显示器424可根据实际需要决定是否配置)通信，还可与一个或者多个使得用户能与该分类设备412交互的设备通信，和/或与使得该分类设备412能与一个或多个其它计算设备进行通信的任何设备(例如网卡，调制解调器等等)通信。这种通信可以通过输入/输出(I/O)接口422进行。并且，分类设备412还可以通过网络适配器420与一个或者多个网络(例如局域网(LAN)，广域网(WAN)和/或公共网络，例如因特网)通信。如图所示，网络适配器420通过总线418与分类设备412的其它模块通信。应当明白，可以结合分类设备412使用其它硬件和/或软件模块，可以包括：微代码、设备驱动器、冗余处理单元、外部磁盘驱动阵列、RAID系统、磁带驱动器以及数据备份存储装置等。
处理器416通过运行存储在存储器428中的程序，从而执行各种功能应用以及数据处理，例如实现本发明实施例所提供的分类方法。
实施例五
本发明实施例五提供了一种计算机可读存储介质，其上存储有计算机程序，该程序被处理器执行时实现如本发明实施例所提供的分类方法，包括：
获取与当前被检体的目标部位相对应的磁共振K空间数据；
将磁共振K空间数据输入到训练完成的目标分类模型中，得到与目标部位相对应的第一分类结果，其中，目标分类模型包括基于有监督学习的方式训练的分类模型。
当然，本发明实施例所提供的计算机可读存储介质，其上存储的计算机程序，还可以执行本申请任意实施例所提供的基于分类设备的分类方法中的相关操作。
本发明实施例的计算机存储介质，可以采用一个或多个计算机可读的介质的任意组合。计算机可读介质可以是计算机可读信号介质或者计算机可读存储介质。计算机可读存储介质例如可以是电、磁、光、电磁、红外线、或半导体的系统、装置或器件，或者任意以上的组合。计算机可读存储介质的例子(非穷举的列表)可以包括：具有一个或多个导线的电连接、便携式计算机磁盘、硬盘、随机存取存储器(RAM)、只读存储器(ROM)、可擦式可编程只读存储器(EPROM或闪存)、光纤、便携式紧凑磁盘只读存储器(CD-ROM)、光存储器件、磁存储器件、或者上述的任意合适的组合。在本文件中，计算机可读存储介质可以是任何包含或存储程序的有形介质，该程序可以被指令执行系统、装置或者器件使用或者与其结合使用。
计算机可读的信号介质可以包括在基带中或者作为载波一部分传播的数据信号，其中承载了计算机可读的程序代码。这种传播的数据信号可以采用多种形式，可以包括电磁信号、光信号或上述的任意合适的组合。计算机可读的信号介质还可以是计算机可读存储介质以外的任何计算机可读介质，该计算机可读介质可以发送、传播或者传输用于由指令执行系统、装置或者器件使用或者与其结合使用的程序。
计算机可读介质上包含的程序代码可以用任何适当的介质传输，可以包括无线、电线、光缆、RF等等，或者上述的任意合适的组合。
可以以一种或多种程序设计语言或其组合来编写用于执行本申请操作的计算机程序代码，所述程序设计语言包括面向对象的程序设计语言—诸如Java、 Smalltalk、C++，还包括常规的过程式程序设计语言—诸如”C”语言或类似的程序设计语言。程序代码可以完全地在用户计算机上执行、部分地在用户计算机上执行、作为一个独立的软件包执行、部分在用户计算机上部分在远程计算机上执行、或者完全在远程计算机或服务器上执行。在涉及远程计算机的情形中，远程计算机可以通过任意种类的网络——包括局域网(LAN)或广域网(WAN)—连接到用户计算机，或者，可以连接到外部计算机(例如利用因特网服务提供商来通过因特网连接)。

Claims (10)

  1. 一种分类方法，包括：
    获取与当前被检体的目标部位相对应的磁共振K空间数据；
    将所述磁共振K空间数据输入到训练完成的目标分类模型中，得到与所述目标部位相对应的第一分类结果，其中，所述目标分类模型包括基于有监督学习的方式训练的分类模型。
  2. 根据权利要求1所述的方法，其中，所述获取与当前被检体的目标部位相对应的磁共振K空间数据，包括：
    利用快速采样方法获取与当前被检体的目标部位相对应的磁共振K空间数据，其中，所述快速采样方法包括变密度采样方法或均匀采样方法。
  3. 根据权利要求1或2所述的方法，在所述将所述磁共振K空间数据输入到训练完成的目标分类模型中之前，所述方法还包括：
    利用补零方式或数据保留方式将所述磁共振K空间数据对应的第一矩阵的矩阵大小调整为预设大小，得到能够被所述目标分类模型处理的第二矩阵，并将所述第二矩阵对应的数据作为输入所述目标分类模型的所述磁共振K空间数据。
  4. 根据权利要求3所述的方法，其中，所述数据保留方式为从所述第一矩阵的中心数据开始，从内向外依次对行和列中的数据进行保留，其中，若所述第一矩阵的行和列均为奇数，则所述第一矩阵的中心数据为一个，若所述第一矩阵的行和列中至少之一为偶数，则所述第一矩阵的中心数据为偶数个。
  5. 根据权利要求3所述的方法，在所述将所述磁共振K空间数据输入到训练完成的目标分类模型中之前，所述方法还包括：
    获取历史被检体的目标部位相对应的历史磁共振K空间数据，并利用类别标记值对所述历史磁共振K空间数据进行类别标记；
    将所述历史磁共振K空间数据输入预先建立的待训练分类模型中，得到第二分类结果；
    基于所述第二分类结果和所述类别标记值对所述待训练分类模型进行参数调整，得到目标分类模型。
  6. 根据权利要求5所述的方法，所述基于第二分类结果和所述类别标记值对所述待训练分类模型进行参数调整，得到目标分类模型，包括：
    将所述第二分类结果和所述类别标记值代入交叉熵损失函数中；
    利用随机梯度下降法调节所述待训练分类模型的参数，将使所述交叉熵损 失函数的值最小的参数作为目标分类模型所对应的参数，得到目标分类模型。
  7. 根据权利要求5所述的方法，在所述将所述历史磁共振K空间数据输入预先建立的待训练分类模型中之前，所述方法还包括：
    利用所述补零方式或所述数据保留方式将所述历史磁共振K空间数据对应的第三矩阵的矩阵大小调整为所述预设大小，得到第四矩阵，并将所述第四矩阵对应的数据作为输入所述待训练分类模型的所述历史磁共振K空间数据。
  8. 一种分类装置，包括：
    磁共振K空间数据获取模块，被配置为获取与当前被检体的目标部位相对应的磁共振K空间数据；
    分类模块，被配置为将所述磁共振K空间数据输入到训练完成的目标分类模型中，得到与所述目标部位相对应的第一分类结果，其中，所述目标分类模型包括基于有监督学习的方式训练的分类模型。
  9. 一种分类设备，所述设备包括：
    一个或多个处理器；
    存储装置，用于存储一个或多个程序，
    当所述一个或多个程序被所述一个或多个处理器执行，使得所述一个或多个处理器实现如权利要求1-7中任一所述的分类方法。
  10. 一种计算机可读存储介质，其上存储有计算机程序，所述计算机程序被处理器执行时实现如权利要求1-7中任一所述的分类方法。
PCT/CN2019/123061 2019-06-18 2019-12-04 一种分类方法、装置、设备和存储介质 WO2020253138A1 (zh)

Applications Claiming Priority (2)

Application Number Priority Date Filing Date Title
CN201910527343.6 2019-06-18
CN201910527343.6A CN112101396A (zh) 2019-06-18 2019-06-18 一种分类方法、装置、设备和存储介质

Publications (1)

Publication Number Publication Date
WO2020253138A1 true WO2020253138A1 (zh) 2020-12-24

Family

ID=73749118

Family Applications (1)

Application Number Title Priority Date Filing Date
PCT/CN2019/123061 WO2020253138A1 (zh) 2019-06-18 2019-12-04 一种分类方法、装置、设备和存储介质

Country Status (2)

Country Link
CN (1) CN112101396A (zh)
WO (1) WO2020253138A1 (zh)

Cited By (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN113392974A (zh) * 2021-07-07 2021-09-14 北京百度网讯科技有限公司 模型训练方法、装置、电子设备以及存储介质
CN113407180A (zh) * 2021-05-28 2021-09-17 济南浪潮数据技术有限公司 一种配置页面生成方法、系统、设备以及介质

Citations (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN104573745A (zh) * 2015-01-21 2015-04-29 中国计量学院 基于磁共振成像的实蝇分类方法
CN106096632A (zh) * 2016-06-02 2016-11-09 哈尔滨工业大学 基于深度学习和mri图像的心室功能指标预测方法
CN106526511A (zh) * 2016-10-21 2017-03-22 杭州电子科技大学 基于k空间中心鬼影定位的SPEED磁共振成像方法
US20180008187A1 (en) * 2016-07-11 2018-01-11 Sony Corporation System of joint brain tumor and cortex reconstruction

Family Cites Families (1)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN103646410B (zh) * 2013-11-27 2016-06-08 中国科学院深圳先进技术研究院 磁共振快速参数成像方法和系统

Patent Citations (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN104573745A (zh) * 2015-01-21 2015-04-29 中国计量学院 基于磁共振成像的实蝇分类方法
CN106096632A (zh) * 2016-06-02 2016-11-09 哈尔滨工业大学 基于深度学习和mri图像的心室功能指标预测方法
US20180008187A1 (en) * 2016-07-11 2018-01-11 Sony Corporation System of joint brain tumor and cortex reconstruction
CN106526511A (zh) * 2016-10-21 2017-03-22 杭州电子科技大学 基于k空间中心鬼影定位的SPEED磁共振成像方法

Cited By (3)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN113407180A (zh) * 2021-05-28 2021-09-17 济南浪潮数据技术有限公司 一种配置页面生成方法、系统、设备以及介质
CN113392974A (zh) * 2021-07-07 2021-09-14 北京百度网讯科技有限公司 模型训练方法、装置、电子设备以及存储介质
CN113392974B (zh) * 2021-07-07 2024-04-09 北京百度网讯科技有限公司 模型训练方法、装置、电子设备以及存储介质

Also Published As

Publication number Publication date
CN112101396A (zh) 2020-12-18

Similar Documents

Publication Publication Date Title
CN109961491B (zh) 多模态图像截断补偿方法、装置、计算机设备和介质
CN107123095B (zh) 一种pet图像重建方法、成像系统
US9076197B2 (en) Probability density function estimation
US20200214619A1 (en) Automated co-registration of prostate mri data
CN113689342B (zh) 一种图像质量优化的方法及系统
RU2667879C1 (ru) Обработка и анализ данных на изображениях компьютерной томографии
US20080247618A1 (en) Interactive diagnostic display system
CN112465834B (zh) 一种血管分割方法及装置
CN111369562B (zh) 图像处理方法、装置、电子设备及存储介质
CN110728673A (zh) 一种目标部位分析方法、装置、计算机设备及存储介质
CN112102276B (zh) 基于迁移学习图像增强的低场强mr胃部分割方法
CN109741312A (zh) 一种肺结节鉴别方法、装置、设备及介质
CN112767505B (zh) 图像处理方法、训练方法、装置、电子终端及存储介质
CN112365980A (zh) 脑肿瘤多靶点辅助诊断与前瞻性治疗演化可视化方法及系统
WO2020253138A1 (zh) 一种分类方法、装置、设备和存储介质
CN112070785A (zh) 一种基于计算机视觉的医学图像分析方法
CN108742679B (zh) 结节检测装置和方法
Sanabria et al. Comparative study of raw ultrasound data representations in deep learning to classify hepatic steatosis
EP2498222A2 (en) Method and system for regression-based 4D mitral valve segmentation from 2D+T magnetic resonance imaging slices
WO2024051018A1 (zh) 一种pet参数图像的增强方法、装置、设备及存储介质
Jaffar et al. Tumor detection from enhanced magnetic resonance imaging using fuzzy curvelet
CN116109610A (zh) 一种超声检查报告图像中乳腺肿瘤分割方法和系统
WO2022193378A1 (zh) 图像重建模型生成方法及装置、图像重建方法及装置、设备、介质
CN115239655A (zh) 一种甲状腺超声影像肿瘤分割分类方法及装置
CN113052930A (zh) 一种胸部dr双能量数字减影图像生成方法

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: 19933302

Country of ref document: EP

Kind code of ref document: A1

NENP Non-entry into the national phase

Ref country code: DE

122 Ep: pct application non-entry in european phase

Ref document number: 19933302

Country of ref document: EP

Kind code of ref document: A1

122 Ep: pct application non-entry in european phase

Ref document number: 19933302

Country of ref document: EP

Kind code of ref document: A1

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

Free format text: NOTING OF LOSS OF RIGHTS PURSUANT TO RULE 112(1) EPC (EPO FORM 1205A DATED 10.08.2022)

122 Ep: pct application non-entry in european phase

Ref document number: 19933302

Country of ref document: EP

Kind code of ref document: A1