WO2022178995A1 - Ct图像去噪处理方法、装置、计算机设备及介质 - Google Patents

Ct图像去噪处理方法、装置、计算机设备及介质 Download PDF

Info

Publication number
WO2022178995A1
WO2022178995A1 PCT/CN2021/096638 CN2021096638W WO2022178995A1 WO 2022178995 A1 WO2022178995 A1 WO 2022178995A1 CN 2021096638 W CN2021096638 W CN 2021096638W WO 2022178995 A1 WO2022178995 A1 WO 2022178995A1
Authority
WO
WIPO (PCT)
Prior art keywords
image
model
noise
target
preset
Prior art date
Application number
PCT/CN2021/096638
Other languages
English (en)
French (fr)
Inventor
李泽远
王健宗
Original Assignee
平安科技(深圳)有限公司
Priority date (The priority date is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the date listed.)
Filing date
Publication date
Application filed by 平安科技(深圳)有限公司 filed Critical 平安科技(深圳)有限公司
Publication of WO2022178995A1 publication Critical patent/WO2022178995A1/zh

Links

Images

Classifications

    • GPHYSICS
    • G06COMPUTING; CALCULATING OR COUNTING
    • G06TIMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T5/00Image enhancement or restoration
    • G06T5/70Denoising; Smoothing
    • GPHYSICS
    • G06COMPUTING; CALCULATING OR COUNTING
    • G06FELECTRIC DIGITAL DATA PROCESSING
    • G06F18/00Pattern recognition
    • G06F18/20Analysing
    • G06F18/21Design or setup of recognition systems or techniques; Extraction of features in feature space; Blind source separation
    • G06F18/214Generating training patterns; Bootstrap methods, e.g. bagging or boosting
    • GPHYSICS
    • G06COMPUTING; CALCULATING OR COUNTING
    • G06NCOMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00Computing arrangements based on biological models
    • G06N3/02Neural networks
    • G06N3/04Architecture, e.g. interconnection topology
    • GPHYSICS
    • G06COMPUTING; CALCULATING OR COUNTING
    • G06NCOMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00Computing arrangements based on biological models
    • G06N3/02Neural networks
    • G06N3/08Learning methods
    • GPHYSICS
    • G06COMPUTING; CALCULATING OR COUNTING
    • G06TIMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T2207/00Indexing scheme for image analysis or image enhancement
    • G06T2207/10Image acquisition modality
    • G06T2207/10072Tomographic images
    • G06T2207/10081Computed x-ray tomography [CT]

Definitions

  • the present application relates to the technical field of data processing, and in particular, to a CT image denoising processing method, apparatus, computer equipment, and computer-readable storage medium.
  • CT imaging technology has been more and more widely used in the diagnosis and treatment of different parts of the human body, such as the mouth, teeth, lungs, and head.
  • the noise sources of CT images mainly include quantum noise, noise introduced by the inherent limitations of CT hardware systems, and noise introduced in the image generation process. Since the noise of CT images has a great influence on the diagnosis results, denoising of CT images is a hot research topic in the field of image processing.
  • CT image denoising in the process of implementing the present application, the inventor found that the prior art has at least the following problems: most of the CT image denoising methods based on deep learning use training a single neural network to realize the denoising task.
  • the noise distribution of CT images in different parts is not the same, and the number of CT images in the same part is small, resulting in a scarce data set for model training, and the accuracy of CT image denoising cannot be guaranteed.
  • a first aspect of the embodiments of the present application provides a CT image denoising processing method, where the CT image denoising processing method includes:
  • the target CT image denoising model is invoked to process the second CT real image to obtain a target CT image.
  • a second aspect of the embodiments of the present application further provides a CT image denoising processing device, where the CT image denoising processing device includes:
  • an image noise adding module used for adding noise to process the first CT real image set to obtain the first CT noise image set
  • a picture classification module configured to classify the first CT noise image set according to a preset position, obtain a first CT noise image subset, and configure a task set according to the first CT noise image subset;
  • a parameter acquisition module configured to perform meta-migration learning based on the task set to obtain target meta-model parameters of the preset meta-model
  • a model determination module configured to determine a preset initial CT image denoising model corresponding to the target part, and update the initial CT image denoising model according to the target metamodel parameters to obtain a target CT image denoising model
  • a denoising processing module configured to acquire a second CT noise image to be denoised, and call the target CT image denoising model to process the second CT noise image to obtain a second CT denoising image
  • a loss detection module configured to calculate the loss value of the second CT real image corresponding to the target part and the second CT denoised image, and detect whether the loss value is less than a preset loss threshold requirement
  • An image determination module configured to call the target CT image denoising model to process the second CT real image to obtain a target CT image when the detection result is that the loss value is less than a preset loss threshold requirement.
  • a third aspect of the embodiments of the present application further provides a computer device, where the computer device includes a processor, and the processor is configured to execute computer-readable instructions stored in a memory to implement the following steps:
  • the target CT image denoising model is invoked to process the second CT real image to obtain a target CT image.
  • a fourth aspect of the embodiments of the present application further provides a computer-readable storage medium, where computer-readable instructions are stored on the computer-readable storage medium, wherein when the computer-readable instructions are executed by a processor, the following steps are implemented:
  • the target CT image denoising model is invoked to process the second CT real image to obtain a target CT image.
  • the above-mentioned CT image denoising processing method, CT image denoising processing device, computer equipment, and computer-readable storage medium add noise to a first CT real image set to obtain a first CT noise image set.
  • the first CT noise image set is trained by means of transfer learning, and target metamodel parameters of a preset metamodel conforming to the real noise distribution can be obtained;
  • the target CT image denoising corresponding to the target part is determined according to the target metamodel parameters model, and call the second CT noise image to be denoised to fine-tune the model parameters of the target CT image denoising model, so that the second CT real image corresponding to the target part and the second CT noise image are different.
  • the loss value is smaller than the preset loss threshold, and finally the real noise of the second CT real image is denoised to obtain the target CT image.
  • the target CT image denoising model of the target part is determined by the target metamodel parameters, which can enhance the rapid convergence ability of CT images under the condition of few sample data, and improve the training efficiency of the target CT image denoising model and the CT image.
  • the present application can be applied to various functional modules of smart cities such as smart government affairs, smart transportation, and smart medical care, such as the CT image denoising processing module of smart medical care, etc., which can promote the rapid development of smart medical care.
  • FIG. 1 is a flowchart of a CT image denoising processing method provided in Embodiment 1 of the present application.
  • FIG. 2 is a structural diagram of a CT image denoising processing apparatus provided in Embodiment 2 of the present application.
  • FIG. 3 is a schematic structural diagram of a computer device provided in Embodiment 3 of the present application.
  • Meta-learning also known as learning to learn, is an important research direction in the field of machine learning, which solves the problem of learning how to learn.
  • the traditional model of machine learning research is to acquire task-specific datasets and use those datasets to train a model from scratch each time.
  • humans can quickly learn the same type of tasks or common tasks by acquiring past experience, because humans know how to learn. If feature extraction is regarded as the process of machine learning on the data set, then the meta-learner is to evaluate this learning process, that is, the process of letting the machine learn to learn, that is, to gain learning experience through learning, and then use these experiences to go to the final target tasks for evaluation.
  • the CT image denoising processing method provided in the embodiment of the present application is executed by a computer device, and correspondingly, the CT image denoising processing method runs in the computer device.
  • FIG. 1 is a flowchart of a CT image denoising processing method according to the first embodiment of the present application.
  • the CT image denoising processing method is used to perform denoising processing on the CT image of the target part.
  • the CT image denoising processing method may include the following steps. According to different requirements, the steps in the flowchart The order can be changed and some can be omitted.
  • the first CT real image set is processed by adding noise to obtain a first CT noise image set.
  • the first CT real image set refers to a CT image set containing real noise
  • the real noise may include quantum noise, noise introduced by inherent limitations of CT hardware systems, and noise in the image generation process. introduced noise.
  • the quantum noise obeys Poisson distribution, which is a kind of readout noise in experimental observation. When the limited number of X-ray photons in the observation is small enough to cause an observable statistical fluctuation in the data readout at the detector, the statistical fluctuation in the readout is called quantum noise, which is also low The main noise faced in dose CT images.
  • the noise introduced by the inherent limitations of the CT hardware system mainly includes the electronic noise in the detector photodiode, the point noise introduced in the data acquisition system, and the noise introduced by X-ray scattering. Such noise is unavoidable and cannot be avoided. subject to human control. This type of noise can be mitigated by improving the hardware system, but it cannot be completely eliminated.
  • the noise introduced in the image generation process refers to the noise generated under the influence of factors such as the selection of the reconstruction algorithm and the selection of parameters.
  • the first CT real image set includes first CT real images of several preset parts, and the preset parts are non-target parts.
  • the preset part is a preset part, for example, the preset part may be a human body part such as a lung, a head, and a tooth.
  • the preset part may be any other body part except the lung.
  • the process of adding noise to the first CT real image set to obtain the first CT noise image set includes:
  • the first Gaussian noise is added to the cropped first CT real image set to obtain a first CT noise image set.
  • the preset size is preset and is used to unify the size of the first CT real image set. By unifying the size of the first CT real image set, the efficiency of adding Gaussian noise can be improved.
  • the preset intensity refers to the preset intensity of Gaussian noise, and for each first CT real image in the first CT real image set, Gaussian noise with the same preset intensity may be added, or may be added in batches.
  • the Gaussian noise with different preset intensities is not limited here.
  • the method of adding Gaussian noise with a preset intensity to an image is in the prior art, and details are not described here.
  • S12 Classify the first CT noise image set according to a preset position to obtain a first CT noise image subset, and configure a task set according to the first CT noise image subset.
  • each first CT noise image in the first CT noise image set is provided with a part code, and the part code is used to identify the part to which the first CT noise image belongs.
  • the part encoding can determine part information of the first CT noise image.
  • the part coding may be numerical coding, letter coding or color coding, which is not limited herein.
  • classifying the first CT noise image set according to a preset position to obtain a first CT noise image subset, and configuring a task set according to the first CT noise image subset includes:
  • the tasks are combined to obtain a task set.
  • first CT noise images corresponding to each of the part codes are used as a first CT noise image subset, and each of the first CT noise image subsets is determined as a task, and the first CT noise image subset is determined as a task. How many different part codes exist in the set, the corresponding number of tasks exist in the task set. For example, if there are 20 different part codes in the first CT noise image set, there are 20 tasks in the task set.
  • S13 Perform meta-transfer learning based on the task set to obtain target meta-model parameters of the preset meta-model.
  • the preset meta-model refers to a meta-model for performing meta-transfer learning
  • the preset meta-model may be a preset neural network structure
  • the deep neural network structure is a It is an algorithm mathematical model that imitates the behavioral characteristics of animal neural network and performs distributed parallel information processing. This kind of network structure relies on the complexity of the system, and achieves the purpose of processing information by adjusting the interconnected relationship between a large number of internal nodes.
  • the preset meta-model has good ductility in denoising CT images of different parts.
  • performing meta-transfer learning based on the task set to obtain target meta-model parameters of the preset meta-model includes:
  • the network layers of the preset metamodel are traversed, and the preset parameters of each network layer are obtained as target metamodel parameters.
  • the task set is split according to a preset split ratio to obtain a training task set and a verification task set.
  • the preset split ratio can be 8:1 or 9:1, and the splitting principle can be random splitting For example, if the task set contains 20 training tasks, 18 tasks are randomly divided as the training task set, and the remaining 2 tasks are used as the verification task set, which is not limited here.
  • the amount of data in the training task set is greater than the amount of data in the verification task set, that is, the training task set is rich sample data, and the verification task set is small sample data.
  • each network layer of the preset meta-model is traversed to obtain the preset parameters of each network layer as the target meta-model parameters, and the traversal order can be To traverse a network layer with a shallow network depth to a network layer with a deeper network depth, or to traverse a network layer with a deeper network depth to a network layer with a shallow network depth.
  • the preset parameter refers to a preset parameter type, and the preset parameter may include, but is not limited to, the weight value and gradient value of the model.
  • the training task set is used to perform meta-training on the preset meta-model.
  • the task set is denoted as D meta data
  • the training task set is denoted as D tr data
  • the verification task is denoted as D tr data.
  • the set is denoted as D te data
  • each training task in the training task set is denoted as T i
  • the meta-model parameters of each training task T i corresponding to the preset meta-model are denoted as ⁇ i .
  • the initial meta-model parameter ⁇ i can be:
  • is the learning rate for task training, refers to the loss function of the preset meta-model corresponding to the training task T i .
  • meta-transfer learning is first performed on a training task with a relatively large amount of CT image data, which can generate meta-model parameters of a preset meta-model with better denoising effect, and the preset meta-model can pass a small amount of to generate generalization performance by iterating the gradient of , i.e., train a pre-set meta-model that is easy to fine-tune through the above steps.
  • S14 Determine a preset initial CT image denoising model corresponding to the target part, and update the initial CT image denoising model according to the target metamodel parameters to obtain a target CT image denoising model.
  • the initial CT image denoising model refers to a preset deep neural network structure used for denoising the CT image of the target part, and the initial CT image denoising model includes Initial model parameters.
  • the network structure of the initial CT image denoising model is the same as the network structure of the preset meta model.
  • the initial model parameters also include weight values and gradient values of each network layer.
  • the step of updating the initial CT image denoising model according to the target metamodel parameters to obtain the target CT image denoising model includes:
  • the initial model parameters are adjusted according to the difference model parameters to update the initial CT image denoising model to obtain a target CT image denoising model.
  • the difference model parameter refers to a parameter that is different between the initial model parameter and the target meta-model parameter.
  • the comparing the initial model parameters and the target meta-model parameters to obtain the difference model parameters may include: arranging the initial model parameters and the target meta-model parameters in a predetermined order; comparing the predetermined order. whether the initial model parameters and the target meta-model parameters at each same position in the sequence are the same; when the comparison result is that the initial model parameters at the same position in the predetermined order are different from the target meta-model parameters , and determine the difference between the initial model parameter and the target meta-model parameter as a difference model parameter.
  • the difference model parameter may also be the target metamodel parameter different from the initial model parameter, and replacing the initial model parameter according to the target metamodel parameter may update all the target metamodel parameters.
  • the initial CT image denoising model is described, and the target CT image denoising model is obtained.
  • updating the initial CT image denoising model according to the target metamodel parameters to obtain the target CT image denoising model further includes:
  • the initial model parameters are replaced according to the target meta-model parameters to update the initial CT image denoising model to obtain a target CT image denoising model.
  • S15 Acquire a second CT noise image to be denoised, and call the target CT image denoising model to process the second CT noise image to obtain a second CT denoising image.
  • the second CT noise image to be denoised refers to an image in which Gaussian noise is added to the second real CT image, and the target CT image denoising model is called to process the first CT image to be denoised.
  • Two CT noise images are obtained to obtain a second CT denoised image after removing Gaussian noise.
  • the acquiring the second CT noise image to be denoised comprises:
  • the second Gaussian noise is added to the second CT real image to obtain a second CT noise image to be denoised.
  • the predetermined intensity refers to a preset intensity for identifying the second Gaussian noise.
  • the present application adopts the method of first adding the second Gaussian noise of preset intensity to the second CT real image, and then calling the target CT image denoising model to process the second CT noise image to remove the second Gaussian noise.
  • the target CT image denoising model processes the second CT real image to remove real noise, which can quickly analyze the denoising processing effect of the target CT image denoising model, reduce the training difficulty of the target CT image denoising model, and further improve the target CT image denoising model. Training efficiency of CT image denoising models.
  • step S16 calculate the loss value of the second CT real image corresponding to the target part and the second CT denoised image, and detect whether the loss value is less than a preset loss threshold requirement, when the detection result is that the loss value is less than When the preset loss threshold is required, step S17 is executed.
  • the preset loss threshold requirement is a preset threshold value used for evaluating whether the loss function converges.
  • the method further includes:
  • the second CT real image refers to a CT image containing real noise
  • the target CT image refers to a clean CT image obtained by performing real noise denoising on the second CT real image CT images.
  • the above CT images can be stored in the target node of the blockchain.
  • a first CT real image set is processed by adding noise to obtain a first CT noise image set, and the first CT noise image set is trained by means of meta-transfer learning,
  • the target meta-model parameters of the preset meta-model that conform to the real noise distribution can be obtained;
  • the target CT image denoising model corresponding to the target part is determined according to the target meta-model parameters, and the CT noise image to be denoised is called for the target CT image
  • the model parameters of the denoising model are fine-tuned, so that the loss values of the second CT real image corresponding to the target part and the second CT noise image meet the preset loss threshold requirements, and finally realize the real image of the second CT image.
  • the noise is denoised to obtain the target CT image.
  • the target CT image denoising model of the target part is determined by the target metamodel parameters, which can enhance the rapid convergence ability of CT images under the condition of few sample data, and improve the training efficiency of the target CT image denoising model and the CT image.
  • the accuracy of the denoising process can be applied to various functional modules of smart cities such as smart government affairs, smart transportation, and smart medical care, such as the CT image denoising processing module of smart medical care, etc., which can promote the rapid development of smart medical care.
  • FIG. 2 is a structural diagram of a CT image denoising processing apparatus provided in Embodiment 2 of the present application.
  • the CT image denoising processing apparatus 20 may include a plurality of functional modules composed of computer program segments.
  • the computer program of each program segment in the CT image denoising processing device 20 can be stored in the memory of the computer device and executed by at least one processor to perform the function of image denoising processing (see Fig. 1 for details). .
  • the CT image denoising processing apparatus 20 may be divided into a plurality of functional modules according to the functions performed by the CT image denoising processing apparatus 20 .
  • the functional modules may include: an image noise addition module 201 , an image classification module 202 , a parameter acquisition module 203 , a model determination module 204 , a denoising processing module 205 , a loss detection module 206 and an image determination module 207 .
  • a module referred to in this application refers to a series of computer program segments that can be executed by at least one processor and can perform fixed functions, and are stored in a memory. In this embodiment, the functions of each module will be described in detail in subsequent embodiments.
  • the image noise adding module 201 can be used to add noise to process the first CT real image set to obtain the first CT noise image set;
  • the picture classification module 202 can be configured to classify the first CT noise image set according to a preset position, obtain a first CT noise image subset, and configure a task set according to the first CT noise image subset;
  • the parameter acquisition module 203 can be used to perform meta-transfer learning based on the task set, and obtain target meta-model parameters of the preset meta-model;
  • the model determination module 204 can be used to determine a preset initial CT image denoising model corresponding to the target part, and update the initial CT image denoising model according to the target metamodel parameters to obtain a target CT image denoising model ;
  • the denoising processing module 205 can acquire a second CT noise image to be denoised, and call the target CT image denoising model to process the second CT noise image to obtain a second CT denoising image;
  • the loss detection module 206 can be configured to calculate the loss value of the second CT real image corresponding to the target part and the second CT denoised image, and detect whether the loss value is less than a preset loss threshold requirement;
  • the image determination module 207 may be configured to call the target CT image denoising model to process the second CT real image to obtain a target CT image when the detection result is that the loss value is less than the preset loss threshold requirement.
  • the computer device 3 includes a memory 31 , at least one processor 32 , at least one communication bus 33 and a transceiver 34 .
  • FIG. 3 does not constitute a limitation of the embodiments of the present application, and may be a bus-type structure or a star-shaped structure. more or less other hardware or software, or a different arrangement of components is shown.
  • the computer device 3 is a device that can automatically perform numerical calculation and/or information processing according to pre-set or stored instructions, and its hardware includes but is not limited to microprocessors, application-specific integrated circuits, Programmable gate arrays, digital processors and embedded devices, etc.
  • the computer equipment 3 may also include client equipment, including but not limited to any electronic product that can interact with the client through a keyboard, a mouse, a remote control, a touchpad or a voice-activated device, etc., for example, Personal computers, tablets, smartphones, digital cameras, etc.
  • a computer program is stored in the memory 31, and when the computer program is executed by the at least one processor 32, all or part of the steps in the above-mentioned big data-based information processing method are implemented.
  • the computer program may be divided into one or more modules/units, and the one or more modules/units may be a series of computer-readable instruction segments capable of performing specific functions, and the instruction segments are used to describe The execution process of the computer program in the computer device.
  • each module described in FIG. 2 is a computer program stored in the memory 31 and executed by the at least one processor 32, thereby realizing the functions of the various modules to achieve information processing based on big data the goal of.
  • Described memory 31 comprises read-only memory (Read-Only Memory, ROM), programmable read-only memory (Programmable Read-Only Memory, PROM), erasable programmable read-only memory (Erasable Programmable Read-Only Memory, EPROM) , One-time Programmable Read-Only Memory (OTPROM), Electronically-Erasable Programmable Read-Only Memory (EEPROM), Compact Disc Read- Only Memory, CD-ROM) or other optical disk storage, magnetic disk storage, magnetic tape storage, or any other computer-readable medium that can be used to carry or store data.
  • Read-Only Memory Read-Only Memory
  • PROM Programmable Read-Only Memory
  • EPROM Erasable Programmable Read-Only Memory
  • OTPROM One-time Programmable Read-Only Memory
  • EEPROM Electronically-Erasable Programmable Read-Only Memory
  • CD-ROM Compact Disc Read- Only Memory
  • CD-ROM Compact Disc Read- Only Memory
  • the computer-readable storage medium may mainly include a storage program area and a storage data area, wherein the storage program area may store an operating system, an application program required by at least one function, and the like; The data created by the use of the node, etc.
  • the computer-readable storage medium may be non-volatile or volatile.
  • the blockchain referred to in this application is a new application mode of computer technologies such as distributed data storage, point-to-point transmission, consensus mechanism, and encryption algorithm.
  • Blockchain essentially a decentralized database, is a series of data blocks associated with cryptographic methods. Each data block contains a batch of network transaction information to verify its Validity of information (anti-counterfeiting) and generation of the next block.
  • the blockchain can include the underlying platform of the blockchain, the platform product service layer, and the application service layer.
  • the at least one processor 32 is a control core (Control Unit) of the computer device 3, using various interfaces and lines to connect various components of the entire computer device 3, and by running or executing storage in the computer device 3
  • the programs or modules in the memory 31 and the data stored in the memory 31 are called to perform various functions of the computer device 3 and process data.
  • the at least one processor 32 executes the computer program stored in the memory, all or part of the steps of the CT image denoising processing method described in the embodiments of the present application are implemented; or all the steps of the CT image denoising processing device are implemented. or some functions.
  • the at least one processor 32 may be composed of integrated circuits, for example, may be composed of a single packaged integrated circuit, or may be composed of multiple integrated circuits packaged with the same function or different functions, including one or more central processing units. (Central Processing unit, CPU), microprocessor, digital processing chip, graphics processor and combination of various control chips, etc.
  • CPU Central Processing unit
  • microprocessor digital processing chip
  • graphics processor and combination of various control chips, etc.
  • the at least one communication bus 33 is configured to enable connection communication between the memory 31 and the at least one processor 32 and the like.
  • the computer device 3 may also include a power source (such as a battery) for supplying power to various components.
  • the power source may be logically connected to the at least one processor 32 through a power management device, so as to be implemented by the power management device Manage charging, discharging, and power management functions.
  • the power source may also include one or more DC or AC power sources, recharging devices, power failure detection circuits, power converters or inverters, power status indicators, and any other components.
  • the computer device 3 may also include a variety of sensors, Bluetooth modules, Wi-Fi modules, etc., which will not be repeated here.
  • the above-mentioned integrated units implemented in the form of software functional modules may be stored in a computer-readable storage medium.
  • the above-mentioned software function modules are stored in a storage medium, and include several instructions to make a computer device (which may be a personal computer, a computer device, or a network device, etc.) or a processor (processor) to execute the methods described in the various embodiments of the present application. part.
  • modules described as separate components may or may not be physically separated, and the components shown as modules may or may not be physical units, and may be located in one place or distributed to multiple network units. Some or all of the modules may be selected according to actual needs to achieve the purpose of the solution in this embodiment.
  • each functional module in each embodiment of the present application may be integrated into one processing unit, or each unit may exist physically alone, or two or more units may be integrated into one unit.
  • the above-mentioned integrated units can be implemented in the form of hardware, or can be implemented in the form of hardware plus software function modules.

Landscapes

  • Engineering & Computer Science (AREA)
  • Theoretical Computer Science (AREA)
  • Physics & Mathematics (AREA)
  • Data Mining & Analysis (AREA)
  • General Physics & Mathematics (AREA)
  • Life Sciences & Earth Sciences (AREA)
  • Artificial Intelligence (AREA)
  • General Engineering & Computer Science (AREA)
  • Evolutionary Computation (AREA)
  • Molecular Biology (AREA)
  • Computational Linguistics (AREA)
  • Software Systems (AREA)
  • Mathematical Physics (AREA)
  • Health & Medical Sciences (AREA)
  • Biomedical Technology (AREA)
  • Biophysics (AREA)
  • Computing Systems (AREA)
  • General Health & Medical Sciences (AREA)
  • Bioinformatics & Cheminformatics (AREA)
  • Evolutionary Biology (AREA)
  • Bioinformatics & Computational Biology (AREA)
  • Computer Vision & Pattern Recognition (AREA)
  • Apparatus For Radiation Diagnosis (AREA)
  • Image Processing (AREA)

Abstract

本申请涉及数据处理技术,提供一种CT图像去噪处理方法、装置、计算机设备与存储介质,包括:加噪处理第一CT真实图像集,得到第一CT噪声图像集;分类第一CT噪声图像集,得到第一CT噪声图像子集,根据第一CT噪声图像子集配置任务集;基于任务集进行元迁移学习,得到目标元模型参数;更新初始CT图像去噪模型,得到目标CT图像去噪模型;获取待去噪的第二CT噪声图像,去噪处理第二CT噪声图像,得到第二CT去噪图像;计算第二CT真实图像与第二CT去噪图像的损失值,检测损失值是否小于预设损失阈值要求;当检测结果为是时,去噪处理第二CT真实图像,得到目标CT图像。本申请能够提高CT图像去躁处理的准确性,促进智慧医疗及智慧城市的建设。

Description

CT图像去噪处理方法、装置、计算机设备及介质
本申请要求于2021年02月26日提交中国专利局,申请号为202110220871.4发明名称为“CT图像去噪处理方法、装置、计算机设备及介质”的中国专利申请的优先权,其全部内容通过引用结合在本申请中。
技术领域
本申请涉及数据处理技术领域,尤其涉及一种CT图像去噪处理方法、装置、计算机设备及计算机可读存储介质。
背景技术
近年来,CT成像技术越来越广泛地应用在口腔、牙齿、肺部以及头部等不同人体部位的诊断与治疗过程中,通过获取各个人体部位的CT图像,能够辅助相关疾病的诊断。在CT成像过程中,CT图像的噪声来源主要包括量子噪声、CT硬件系统固有的限制所引入的噪声以及图像生成过程中引入的噪声。由于CT图像的噪声对诊断结果影响较大,因此对CT图像进行去噪处理,是图像处理领域研究的热点。
针对CT图像去噪的情况,在实现本申请的过程中,发明人发现现有技术至少存在如下问题:大部分基于深度学习的CT图像去噪方法是采用训练单神经网络来实现去噪任务的,然而不同部位的CT图像的噪声分布不尽相同,相同部位的CT图像的数量较少,导致模型训练的数据集稀少,无法保证CT图像去噪的准确性。
因此,有必要提供一种CT图像去噪处理方法,能够提高CT图像去噪处理的准确性。
发明内容
鉴于以上内容,有必要提出一种CT图像去噪处理方法、CT图像去噪处理装置、计算机设备及计算机可读存储介质,能够提高CT图像去噪处理的准确性。
本申请实施例第一方面提供一种CT图像去噪处理方法,所述CT图像去噪处理方法包括:
加噪处理第一CT真实图像集,得到第一CT噪声图像集;
根据预设部位分类所述第一CT噪声图像集,得到第一CT噪声图像子集,并根据所述第一CT噪声图像子集配置任务集;
基于所述任务集进行元迁移学习,得到预设元模型的目标元模型参数;
确定预设的对应目标部位的初始CT图像去噪模型,并根据所述目标元模型参数更新所述初始CT图像去噪模型,得到目标CT图像去噪模型;
获取待去噪的第二CT噪声图像,并调用所述目标CT图像去噪模型处理所述第二CT噪声 图像,得到第二CT去噪图像;
计算对应所述目标部位的第二CT真实图像与所述第二CT去噪图像的损失值,并检测所述损失值是否小于预设损失阈值要求;
当检测结果为所述损失值小于预设损失阈值要求时,调用所述目标CT图像去噪模型处理所述第二CT真实图像,得到目标CT图像。
本申请实施例第二方面还提供一种CT图像去噪处理装置,所述CT图像去噪处理装置包括:
图像加噪模块,用于加噪处理第一CT真实图像集,得到第一CT噪声图像集;
图片分类模块,用于根据预设部位分类所述第一CT噪声图像集,得到第一CT噪声图像子集,并根据所述第一CT噪声图像子集配置任务集;
参数获取模块,用于基于所述任务集进行元迁移学习,得到预设元模型的目标元模型参数;
模型确定模块,用于确定预设的对应目标部位的初始CT图像去噪模型,并根据所述目标元模型参数更新所述初始CT图像去噪模型,得到目标CT图像去噪模型;
去噪处理模块,用于获取待去噪的第二CT噪声图像,并调用所述目标CT图像去噪模型处理所述第二CT噪声图像,得到第二CT去噪图像;
损失检测模块,用于计算对应所述目标部位的第二CT真实图像与所述第二CT去噪图像的损失值,并检测所述损失值是否小于预设损失阈值要求;
图像确定模块,用于当检测结果为所述损失值小于预设损失阈值要求时,调用所述目标CT图像去噪模型处理所述第二CT真实图像,得到目标CT图像。
本申请实施例第三方面还提供一种计算机设备,所述计算机设备包括处理器,所述处理器用于执行存储器中存储的计算机可读指令以实现以下步骤:
加噪处理第一CT真实图像集,得到第一CT噪声图像集;
根据预设部位分类所述第一CT噪声图像集,得到第一CT噪声图像子集,并根据所述第一CT噪声图像子集配置任务集;
基于所述任务集进行元迁移学习,得到预设元模型的目标元模型参数;
确定预设的对应目标部位的初始CT图像去噪模型,并根据所述目标元模型参数更新所述初始CT图像去噪模型,得到目标CT图像去噪模型;
获取待去噪的第二CT噪声图像,并调用所述目标CT图像去噪模型处理所述第二CT噪声图像,得到第二CT去噪图像;
计算对应所述目标部位的第二CT真实图像与所述第二CT去噪图像的损失值,并检测所述损失值是否小于预设损失阈值要求;
当检测结果为所述损失值小于预设损失阈值要求时,调用所述目标CT图像去噪模型处理所述第二CT真实图像,得到目标CT图像。
本申请实施例第四方面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上存储有计算机可读指令,其中,所述计算机可读指令被处理器执行时实现以下步骤:
加噪处理第一CT真实图像集,得到第一CT噪声图像集;
根据预设部位分类所述第一CT噪声图像集,得到第一CT噪声图像子集,并根据所述第一CT噪声图像子集配置任务集;
基于所述任务集进行元迁移学习,得到预设元模型的目标元模型参数;
确定预设的对应目标部位的初始CT图像去噪模型,并根据所述目标元模型参数更新所述初始CT图像去噪模型,得到目标CT图像去噪模型;
获取待去噪的第二CT噪声图像,并调用所述目标CT图像去噪模型处理所述第二CT噪声图像,得到第二CT去噪图像;
计算对应所述目标部位的第二CT真实图像与所述第二CT去噪图像的损失值,并检测所述损失值是否小于预设损失阈值要求;
当检测结果为所述损失值小于预设损失阈值要求时,调用所述目标CT图像去噪模型处理所述第二CT真实图像,得到目标CT图像。
本申请实施例提供的上述CT图像去噪处理方法、CT图像去噪处理装置、计算机设备以及计算机可读存储介质,加噪处理第一CT真实图像集,得到第一CT噪声图像集,采用元迁移学习的方式对所述第一CT噪声图像集进行训练,能够得到符合真实噪声分布的预设元模型的目标元模型参数;根据所述目标元模型参数确定对应目标部位的目标CT图像去噪模型,并调用待去噪的第二CT噪声图像对所述目标CT图像去噪模型的模型参数进行微调,以使得对应所述目标部位的第二CT真实图像与所述第二CT噪声图像的损失值小于预设损失阈值,最终实现对第二CT真实图像的真实噪声进行去噪处理,得到目标CT图像。本申请通过所述目标元模型参数确定目标部位的目标CT图像去噪模型,能够增强CT图像在少样本数据条件下的快速收敛能力,提高所述目标CT图像去噪模型的训练效率以及CT图像去噪处理的准确性。本申请可应用于智慧政务、智慧交通、智慧医疗等智慧城市的各个功能模块中,比如智慧医疗的CT图像去噪处理模块等,能够促进智慧医疗的快速发展。
附图说明
图1是本申请实施例一提供的CT图像去噪处理方法的流程图。
图2是本申请实施例二提供的CT图像去噪处理装置的结构图。
图3是本申请实施例三提供的计算机设备的结构示意图。
如下具体实施方式将结合上述附图进一步说明本申请。
具体实施方式
为了能够更清楚地理解本申请的上述目的、特征和优点,下面结合附图和具体实施 例对本申请进行详细描述。需要说明的是,在不冲突的情况下,本申请的实施例及实施例中的特征可以相互组合。
在下面的描述中阐述了很多具体细节以便于充分理解本申请,所描述的实施例是本申请一部分实施例,而不是全部的实施例。基于本申请中的实施例,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,都属于本申请保护的范围。
除非另有定义,本文所使用的所有的技术和科学术语与属于本申请的技术领域的技术人员通常理解的含义相同。本文中在本申请的说明书中所使用的术语只是为了描述具体的实施例的目的,不是旨在于限制本申请。
元学习又叫做学会学习,是机器学习领域一个重要的研究方向,它解决的是学会如何学习的问题。传统的机器学习研究模式是:获取特定任务的数据集,每次再利用这些数据集从头开始训练模型。然而,人类可以通过获取以往的经验,对同类型的任务或有共性的任务进行快速学习,这是因为人类懂得如何学习。如果把特征提取视为机器在数据集上学习的过程,那么元学习器就是要评估这个学习过程,也就是让机器学习学习的过程,即通过学习获得学习经验,利用这些经验再去对最终的目标任务进行评估。
本申请实施例提供的CT图像去噪处理方法由计算机设备执行,相应地,所述CT图像去噪处理方法运行于计算机设备中。
图1是本申请第一实施方式的CT图像去噪处理方法的流程图。所述CT图像去噪处理方法用于对目标部位的CT图像进行去噪处理,如图1所示,所述CT图像去噪处理方法可以包括如下步骤,根据不同的需求,该流程图中步骤的顺序可以改变,某些可以省略。
S11,加噪处理第一CT真实图像集,得到第一CT噪声图像集。
在本申请的至少一实施例中,第一CT真实图像集是指包含真实噪声的CT图像集,所述真实噪声可以包括量子噪声、CT硬件系统固有的限制所引入的噪声以及图像生成过程中引入的噪声。其中,所述量子噪声服从泊松分布,是一种实验观测中的读出噪声。当观测中数量有限的X射线光子数量少到能够引发探测器上数据读出中出现可观测到的统计涨落时,这种读出的统计涨落被称作量子噪声,这种噪声也是低剂量CT图像中面临的主要噪声。所述CT硬件系统固有的限制所引入的噪声主要包含探测器光电二极管中的电子噪声、数据采集系统中引入的点子噪声以及X射线散射等引入的噪声,这类噪声是无法避免的,也不受人为控制影响。可以通过改善硬件系统来减轻这类噪声,但是无法完全消除。所述图像生成过程中引入的噪声是指受重建算法的选择、参数选取等因素的影响而产生的噪声。
所述第一CT真实图像集包含若干个预设部位的第一CT真实图像,所述预设部位为非目标部位。所述预设部位为预先设置的部位,例如,所述预设部位可以为肺部、头部、牙齿部等人体部位。示例性地,当所述目标部位为肺部时,所述预设部位可以是除肺部 外的其他任意人体部位。
可选地,所述加噪处理第一CT真实图像集,得到第一CT噪声图像集包括:
裁剪处理所述第一CT真实图像集,得到预设尺寸的第一CT真实图像集;
获取预设强度的第一高斯噪声;
添加所述第一高斯噪声至裁剪后的所述第一CT真实图像集中,得到第一CT噪声图像集。
其中,所述预设尺寸为预先设置的,用于统一第一CT真实图像集的尺寸,通过统一所述第一CT真实图像集的尺寸,能够提高高斯噪声添加的效率。所述预设强度是指预先设置的高斯噪声的强度,对于所述第一CT真实图像集中的每一第一CT真实图像,可以添加所述预设强度相同的高斯噪声,也可以分批添加所述预设强度不相同的高斯噪声,在此不做限制。对图像添加预设强度的高斯噪声的方式为现有技术,在此不做赘述。
S12,根据预设部位分类所述第一CT噪声图像集,得到第一CT噪声图像子集,并根据所述第一CT噪声图像子集配置任务集。
在本申请的至少一实施例中,对所述第一CT噪声图像集中每一张第一CT噪声图像均设有部位编码,所述部位编码用于标识第一CT噪声图像所属部位,通过所述部位编码能够确定所述第一CT噪声图像的部位信息。所述部位编码可以为数字编码、字母编码或颜色编码,在此不做限制。
可选地,所述根据预设部位分类所述第一CT噪声图像集,得到第一CT噪声图像子集,并根据所述第一CT噪声图像子集配置任务集包括:
获取所述第一CT噪声图像集中每一第一CT噪声图像的部位编码;
选取所述部位编码一致的所述第一CT噪声图像为第一CT噪声图像子集;
确定每一所述第一CT噪声图像子集为一个任务;
组合所述任务,得到任务集。
其中,将每一所述部位编码对应的若干个第一CT噪声图像作为第一CT噪声图像子集,确定每一所述第一CT噪声图像子集为一个任务,所述第一CT噪声图像集中存在多少个不相同的部位编码,所述任务集中就存在相应数量的任务。例如,所述第一CT噪声图像集中存在20个不相同的部位编码,所述任务集中就存在20个任务。
S13,基于所述任务集进行元迁移学习,得到预设元模型的目标元模型参数。
在本申请的至少一实施例中,所述预设元模型是指进行元迁移学习的元模型,所述预设元模型可以是预设的神经网络结构,所述深度神经网络结构是一种模仿动物神经网络行为特征、进行分布式并行信息处理的算法数学模型。这种网络结构依靠系统的复杂程度,通过调整内部大量节点之间相互连接的关系,从而达到处理信息的目的。所述预设元模型在针对不同部位的CT图像进行去噪具有较好的延展性。
可选地,所述基于所述任务集进行元迁移学习,得到预设元模型的目标元模型参数包括:
拆分所述任务集,得到训练任务集与验证任务集;
计算所述训练任务集中每一训练任务对应所述预设元模型的初始元模型参数,得到初始元模型参数集;
基于所述初始元模型参数集调用所述验证任务集计算所述预设元模型的总损失函数;
采用随机梯度下降算法优化所述总损失函数,并检测优化后的所述总损失函数是否收敛;
当检测结果为优化后的所述总损失函数收敛时,遍历所述预设元模型的网络层,得到各个网络层的预设参数作为目标元模型参数。
其中,按照预设拆分比例拆分所述任务集,得到训练任务集与验证任务集,所述预设拆分比例可以为8:1或者9:1,拆分原则可以为随机拆分的方式,例如,对于任务集中包含20个训练任务来说,随机拆分18个任务作为训练任务集,剩余2个任务作为验证任务集,在此不做限制。所述训练任务集中的数据量大于所述验证任务集中的数据量,即所述训练任务集为富样本数据,所述验证任务集为少样本数据。
其中,当检测结果为优化后的所述总损失函数收敛时,对所述预设元模型的各个网络层进行遍历,以获取各个网络层的预设参数作为目标元模型参数,遍历的顺序可以为对网络深度较浅的网络层至网络深度较深的网络层进行遍历,或者为对网络深度较深的网络层至网络深度较浅的网络层进行遍历。所述预设参数是指预先设置的参数类型,所述预设参数可以包括但不限于:模型的权重值与梯度值。
所述训练任务集用于对所述预设元模型进行元训练,示例性地,将所述任务集记作D meta数据,将所述训练任务集记作D tr数据,将所述验证任务集记作D te数据,将所述训练任务集中的每一训练任务记作T i,将每一训练任务T i对应所述预设元模型的元模型参数记作θ i
依据元模型参数的初始值θ的一次或多次梯度更新来适应一个新的训练任务Τ i,对于一次梯度更新,所述初始元模型参数θ i可以为:
Figure PCTCN2021096638-appb-000001
其中,α是任务训练的学习率,
Figure PCTCN2021096638-appb-000002
是指所述预设元模型对应训练任务T i的损失函数。
基于所述初始元模型参数集调用所述验证任务集D te数据计算所述预设元模型的总损失函数,优化元模型参数的参数值θ,使任务集D meta数据相对于元模型参数θ i的测试误差最小,具体的目标为:
Figure PCTCN2021096638-appb-000003
其中,
Figure PCTCN2021096638-appb-000004
是指基于所述验证任务集D te数据得到的预设元模型的总损失函数。
采用随机梯度下降使用上述公式2对元迁移学习的过程进行优化,使得优化后的总损失 函数达到收敛状态,此时,参数更新规则表示为:
Figure PCTCN2021096638-appb-000005
其中,β是学习率。
本申请首先在CT图像数据量相对较大的训练任务上进行元迁移学习,能够生成去噪效果较好的预设元模型的元模型参数,所述预设元模型能够在测试任务上通过少量的梯度迭代来产生泛化的性能,即通过上述步骤训练了一个易于微调的预设元模型。
S14,确定预设的对应目标部位的初始CT图像去噪模型,并根据所述目标元模型参数更新所述初始CT图像去噪模型,得到目标CT图像去噪模型。
在本申请的至少一实施例中,所述初始CT图像去噪模型是指预先设置的用于对目标部位的CT图像进行去噪处理的深度神经网络结构,所述初始CT图像去噪模型包含初始模型参数。在一实施例中,所述初始CT图像去噪模型的网络结构与所述预设元模型的网络结构相同。所述初始模型参数也包含各个网络层的权重值与梯度值。
可选地,所述根据所述目标元模型参数更新所述初始CT图像去噪模型,得到目标CT图像去噪模型包括:
获取所述初始CT图像去噪模型的初始模型参数;
对比所述初始模型参数与所述目标元模型参数,得到差异模型参数;
根据所述差异模型参数调整所述初始模型参数以更新所述初始CT图像去噪模型,得到目标CT图像去噪模型。
其中,所述差异模型参数是指所述初始模型参数与所述目标元模型参数相比,不同的参数。示例性地,所述对比所述初始模型参数与所述目标元模型参数,得到差异模型参数可以包括:按照预定顺序分别排列所述初始模型参数与所述目标元模型参数;对比所述预定顺序中每一相同位置处的所述初始模型参数与所述目标元模型参数是否相同;当对比结果为所述预定顺序中相同位置处的所述初始模型参数与所述目标元模型参数不相同时,确定所述初始模型参数与所述目标元模型参数的差值作为差异模型参数。可以理解的是,在其他实施例中,所述差异模型参数还可以为与所述初始模型参数不同的所述目标元模型参数,根据所述目标元模型参数替换所述初始模型参数可以更新所述初始CT图像去噪模型,得到目标CT图像去噪模型。
在其他实施例中,可选地,所述根据所述目标元模型参数更新所述初始CT图像去噪模型,得到目标CT图像去噪模型还包括:
获取所述初始CT图像去噪模型的初始模型参数;
根据所述目标元模型参数替换所述初始模型参数以更新所述初始CT图像去噪模型,得到目标CT图像去噪模型。
S15,获取待去噪的第二CT噪声图像,并调用所述目标CT图像去噪模型处理所述第二CT噪声图像,得到第二CT去噪图像。
在本申请的至少一实施例中,所述待去噪的第二CT噪声图像是指对第二CT真实图像添加高斯噪声的图像,调用所述目标CT图像去噪模型处理待去噪的第二CT噪声图像,得到去除高斯噪声后的第二CT去噪图像。
可选地,所述获取待去噪的第二CT噪声图像包括:
获取对应所述目标部位的第二CT真实图像;
确定预定强度的第二高斯噪声;
添加所述第二高斯噪声至所述第二CT真实图像中,得到待去噪的第二CT噪声图像。
其中,所述预定强度是指预先设置的标识所述第二高斯噪声的强度。
本申请采用先对第二CT真实图像添加预设强度的第二高斯噪声,再调用目标CT图像去噪模型处理所述第二CT噪声图像以去除第二高斯噪声的方式,相较于直接调用目标CT图像去噪模型处理所述第二CT真实图像以去除真实噪声的方式,能够快速分析目标CT图像去噪模型的去噪处理效果,降低目标CT图像去噪模型的训练难度,进而提高目标CT图像去噪模型的训练效率。
S16,计算对应所述目标部位的第二CT真实图像与所述第二CT去噪图像的损失值,并检测所述损失值是否小于预设损失阈值要求,当检测结果为所述损失值小于预设损失阈值要求时,执行步骤S17。
在本申请的至少一实施例中,所述预设损失阈值要求为预先设置的,用于评估损失函数是否收敛的阈值。
在一实施例中,所述方法还包括:
当检测结果为所述损失值未满足预设损失阈值要求时,根据所述第二CT真实图像与所述第二CT去噪图像获取所述目标CT图像去噪模型的模型损失函数;
采用梯度下降算法优化所述模型损失函数,直至所述模型损失函数收敛;
获取所述模型损失函数收敛时的模型参数,并根据所述模型参数建立目标CT图像去噪模型。
S17,调用所述目标CT图像去噪模型处理所述第二CT真实图像,得到目标CT图像。
在本申请的至少一实施例中,所述第二CT真实图像是指包含真实噪声的CT图像,所述目标CT图像是指对所述第二CT真实图像进行真实噪声去噪处理后的干净CT图像。
需要强调的是,为进一步保证上述CT图像的私密性和安全性,上述CT图像可存储于区块链的目标节点中。
本申请实施例提供的上述CT图像去噪处理方法,加噪处理第一CT真实图像集,得到第一CT噪声图像集,采用元迁移学习的方式对所述第一CT噪声图像集进行训练,能够得到符合真实噪声分布的预设元模型的目标元模型参数;根据所述目标元模型参数确定对应目标部位的目标CT图像去噪模型,并调用待去噪CT噪声图像对所述目标CT图像去噪模型的模型参数进行微调,以使得对应所述目标部位的第二CT真实图像与所述第二 CT噪声图像的损失值满足预设损失阈值要求,最终实现对第二CT真实图像的真实噪声进行去噪处理,得到目标CT图像。本申请通过所述目标元模型参数确定目标部位的目标CT图像去噪模型,能够增强CT图像在少样本数据条件下的快速收敛能力,提高所述目标CT图像去噪模型的训练效率以及CT图像去噪处理的准确性。本申请可应用于智慧政务、智慧交通、智慧医疗等智慧城市的各个功能模块中,比如智慧医疗的CT图像去噪处理模块等,能够促进智慧医疗的快速发展。
图2是本申请实施例二提供的CT图像去噪处理装置的结构图。
在一些实施例中,所述CT图像去噪处理装置20可以包括多个由计算机程序段所组成的功能模块。所述CT图像去噪处理装置20中的各个程序段的计算机程序可以存储于计算机设备的存储器中,并由至少一个处理器所执行,以执行(详见图1描述)图像去躁处理的功能。
本实施例中,所述CT图像去噪处理装置20根据其所执行的功能,可以被划分为多个功能模块。所述功能模块可以包括:图像加噪模块201、图片分类模块202、参数获取模块203、模型确定模块204、去噪处理模块205、损失检测模块206以及图像确定模块207。本申请所称的模块是指一种能够被至少一个处理器所执行并且能够完成固定功能的一系列计算机程序段,其存储在存储器中。在本实施例中,关于各模块的功能将在后续的实施例中详述。
所述图像加噪模块201,可以用于加噪处理第一CT真实图像集,得到第一CT噪声图像集;
所述图片分类模块202,可以用于根据预设部位分类所述第一CT噪声图像集,得到第一CT噪声图像子集,并根据所述第一CT噪声图像子集配置任务集;
所述参数获取模块203,可以用于基于所述任务集进行元迁移学习,得到预设元模型的目标元模型参数;
所述模型确定模块204,可以用于确定预设的对应目标部位的初始CT图像去噪模型,并根据所述目标元模型参数更新所述初始CT图像去噪模型,得到目标CT图像去噪模型;
所述去噪处理模块205,可以获取待去噪的第二CT噪声图像,并调用所述目标CT图像去噪模型处理所述第二CT噪声图像,得到第二CT去噪图像;
所述损失检测模块206,可以用于计算对应所述目标部位的第二CT真实图像与所述第二CT去噪图像的损失值,并检测所述损失值是否小于预设损失阈值要求;
所述图像确定模块207,可以用于当检测结果为所述损失值小于预设损失阈值要求时,调用所述目标CT图像去噪模型处理所述第二CT真实图像,得到目标CT图像。
参阅图3所示,为本申请实施例三提供的计算机设备的结构示意图。在本申请较佳实施例中,所述计算机设备3包括存储器31、至少一个处理器32、至少一条通信总线33及收发器34。
本领域技术人员应该了解,图3示出的计算机设备的结构并不构成本申请实施例的限定,既可以是总线型结构,也可以是星形结构,所述计算机设备3还可以包括比图示更多或更少的其他硬件或者软件,或者不同的部件布置。
在一些实施例中,所述计算机设备3是一种能够按照事先设定或存储的指令,自动进行数值计算和/或信息处理的设备,其硬件包括但不限于微处理器、专用集成电路、可编程门阵列、数字处理器及嵌入式设备等。所述计算机设备3还可包括客户设备,所述客户设备包括但不限于任何一种可与客户通过键盘、鼠标、遥控器、触摸板或声控设备等方式进行人机交互的电子产品,例如,个人计算机、平板电脑、智能手机、数码相机等。
需要说明的是,所述计算机设备3仅为举例,其他现有的或今后可能出现的电子产品如可适应于本申请,也应包含在本申请的保护范围以内,并以引用方式包含于此。
在一些实施例中,所述存储器31中存储有计算机程序,所述计算机程序被所述至少一个处理器32执行时实现如所述的基于大数据的信息处理方法中的全部或者部分步骤。示例性的,所述计算机程序可以被分割成一个或多个模块/单元,所述一个或多个模块/单元可以是能够完成特定功能的一系列计算机可读指令段,该指令段用于描述所述计算机程序在所述计算机设备中的执行过程。例如,图2中所述的各个模块是存储在所述存储器31中的计算机程序,并由所述至少一个处理器32所执行,从而实现所述各个模块的功能以达到基于大数据的信息处理的目的。所述存储器31包括只读存储器(Read-Only Memory,ROM)、可编程只读存储器(Programmable Read-Only Memory,PROM)、可擦除可编程只读存储器(Erasable Programmable Read-Only Memory,EPROM)、一次可编程只读存储器(One-time Programmable Read-Only Memory,OTPROM)、电子擦除式可复写只读存储器(Electrically-Erasable Programmable Read-Only Memory,EEPROM)、只读光盘(Compact Disc Read-Only Memory,CD-ROM)或其他光盘存储器、磁盘存储器、磁带存储器、或者能够用于携带或存储数据的计算机可读的任何其他介质。
进一步地,所述计算机可读存储介质可主要包括存储程序区和存储数据区,其中,存储程序区可存储操作系统、至少一个功能所需的应用程序等;存储数据区可存储根据区块链节点的使用所创建的数据等。所述计算机可读存储介质可以是非易失性,也可以是易失性。
本申请所指区块链是分布式数据存储、点对点传输、共识机制、加密算法等计算机技术的新型应用模式。区块链(Blockchain),本质上是一个去中心化的数据库,是一串使用密码学方法相关联产生的数据块,每一个数据块中包含了一批次网络交易的信息,用于验证其信息的有效性(防伪)和生成下一个区块。区块链可以包括区块链底层平台、平台产品服务层以及应用服务层等。
在一些实施例中,所述至少一个处理器32是所述计算机设备3的控制核心(Control Unit),利用各种接口和线路连接整个计算机设备3的各个部件,通过运行或执行存储在所述存储器 31内的程序或者模块,以及调用存储在所述存储器31内的数据,以执行计算机设备3的各种功能和处理数据。例如,所述至少一个处理器32执行所述存储器中存储的计算机程序时实现本申请实施例中所述的CT图像去噪处理方法的全部或者部分步骤;或者实现CT图像去噪处理装置的全部或者部分功能。所述至少一个处理器32可以由集成电路组成,例如可以由单个封装的集成电路所组成,也可以是由多个相同功能或不同功能封装的集成电路所组成,包括一个或者多个中央处理器(Central Processing unit,CPU)、微处理器、数字处理芯片、图形处理器及各种控制芯片的组合等。
在一些实施例中,所述至少一条通信总线33被设置为实现所述存储器31以及所述至少一个处理器32等之间的连接通信。
尽管未示出,所述计算机设备3还可以包括给各个部件供电的电源(比如电池),优选的,电源可以通过电源管理装置与所述至少一个处理器32逻辑相连,从而通过电源管理装置实现管理充电、放电、以及功耗管理等功能。电源还可以包括一个或一个以上的直流或交流电源、再充电装置、电源故障检测电路、电源转换器或者逆变器、电源状态指示器等任意组件。所述计算机设备3还可以包括多种传感器、蓝牙模块、Wi-Fi模块等,在此不再赘述。
上述以软件功能模块的形式实现的集成的单元,可以存储在一个计算机可读取存储介质中。上述软件功能模块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,计算机设备,或者网络设备等)或处理器(processor)执行本申请各个实施例所述方法的部分。
在本申请所提供的几个实施例中,应该理解到,所揭露的装置和方法,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,所述模块的划分,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。
所述作为分离部件说明的模块可以是或者也可以不是物理上分开的,作为模块显示的部件可以是或者也可以不是物理单元,既可以位于一个地方,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模块来实现本实施例方案的目的。
另外,在本申请各个实施例中的各功能模块可以集成在一个处理单元中,也可以是各个单元单独物理存在,也可以两个或两个以上单元集成在一个单元中。上述集成的单元既可以采用硬件的形式实现,也可以采用硬件加软件功能模块的形式实现。
对于本领域技术人员而言,显然本申请不限于上述示范性实施例的细节,而且在不背离本申请的精神或基本特征的情况下,能够以其他的具体形式实现本申请。因此,无论从哪一点来看,均应将实施例看作是示范性的,而且是非限制性的,本申请的范围由所附权利要求而不是上述说明限定,因此旨在将落在权利要求的等同要件的含义和范围内的所有变化涵括在本申请内。不应将权利要求中的任何附图标记视为限制所涉及的权利要求。此外,显然“包括”一词不排除其他单元或,单数不排除复数。说明书中陈述的多个单元或装置也可以由一个单元或装置通过软件或者硬件来实现。第一,第二等词语用来表示名称,而并不表示任何特 定的顺序。
最后应说明的是,以上实施例仅用以说明本申请的技术方案而非限制,尽管参照较佳实施例对本申请进行了详细说明,本领域的普通技术人员应当理解,可以对本申请的技术方案进行修改或等同替换,而不脱离本申请技术方案的精神和范围。

Claims (20)

  1. 一种CT图像去噪处理方法,其中,所述CT图像去噪处理方法包括:
    加噪处理第一CT真实图像集,得到第一CT噪声图像集;
    根据预设部位分类所述第一CT噪声图像集,得到第一CT噪声图像子集,并根据所述第一CT噪声图像子集配置任务集;
    基于所述任务集进行元迁移学习,得到预设元模型的目标元模型参数;
    确定预设的对应目标部位的初始CT图像去噪模型,并根据所述目标元模型参数更新所述初始CT图像去噪模型,得到目标CT图像去噪模型;
    获取待去噪的第二CT噪声图像,并调用所述目标CT图像去噪模型处理所述第二CT噪声图像,得到第二CT去噪图像;
    计算对应所述目标部位的第二CT真实图像与所述第二CT去噪图像的损失值,并检测所述损失值是否小于预设损失阈值要求;
    当检测结果为所述损失值小于预设损失阈值要求时,调用所述目标CT图像去噪模型处理所述第二CT真实图像,得到目标CT图像。
  2. 根据权利要求1所述的CT图像去噪处理方法,其中,所述加噪处理第一CT真实图像集,得到第一CT噪声图像集包括:
    裁剪处理所述第一CT真实图像集,得到预设尺寸的第一CT真实图像集;
    获取预设强度的第一高斯噪声;
    添加所述第一高斯噪声至裁剪后的所述第一CT真实图像集中,得到第一CT噪声图像集。
  3. 根据权利要求1所述的CT图像去噪处理方法,其中,所述根据预设部位分类所述第一CT噪声图像集,得到第一CT噪声图像子集,并根据所述第一CT噪声图像子集配置任务集包括:
    获取所述第一CT噪声图像集中每一第一CT噪声图像的部位编码;
    选取所述部位编码一致的所述第一CT噪声图像为第一CT噪声图像子集;
    确定每一所述第一CT噪声图像子集为一个任务;
    组合所述任务,得到任务集。
  4. 根据权利要求1所述的CT图像去噪处理方法,其中,所述基于所述任务集进行元迁移学习,得到预设元模型的目标元模型参数包括:
    拆分所述任务集,得到训练任务集与验证任务集;
    计算所述训练任务集中每一训练任务对应所述预设元模型的初始元模型参数,得到初始元模型参数集;
    基于所述初始元模型参数集调用所述验证任务集计算所述预设元模型的总损失函数;
    采用随机梯度下降算法优化所述总损失函数,并检测优化后的所述总损失函数是否收敛;
    当检测结果为优化后的所述总损失函数收敛时,遍历所述预设元模型的网络层,得到各个网络层的预设参数作为目标元模型参数。
  5. 根据权利要求1所述的CT图像去噪处理方法,其中,所述根据所述目标元模型参数更新所述初始CT图像去噪模型,得到目标CT图像去噪模型包括:
    获取所述初始CT图像去噪模型的初始模型参数;
    对比所述初始模型参数与所述目标元模型参数,得到差异模型参数;
    根据所述差异模型参数调整所述初始模型参数以更新所述初始CT图像去噪模型,得到目标CT图像去噪模型。
  6. 根据权利要求1所述的CT图像去噪处理方法,其中,所述获取待去噪的第二CT噪声图像包括:
    获取对应所述目标部位的第二CT真实图像;
    确定预定强度的第二高斯噪声;
    添加所述第二高斯噪声至所述第二CT真实图像中,得到待去噪的第二CT噪声图像。
  7. 根据权利要求1所述的CT图像去噪处理方法,其中,所述方法还包括:
    当检测结果为所述损失值未满足预设损失阈值要求时,根据所述第二CT真实图像与所述第二CT去噪图像获取所述目标CT图像去噪模型的模型损失函数;
    采用梯度下降算法优化所述模型损失函数,直至所述模型损失函数收敛;
    获取所述模型损失函数收敛时的模型参数,并根据所述模型参数建立目标CT图像去噪模型。
  8. 一种CT图像去噪处理装置,其中,所述CT图像去噪处理装置包括:
    图像加噪模块,用于加噪处理第一CT真实图像集,得到第一CT噪声图像集;
    图片分类模块,用于根据预设部位分类所述第一CT噪声图像集,得到第一CT噪声图像子集,并根据所述第一CT噪声图像子集配置任务集;
    参数获取模块,用于基于所述任务集进行元迁移学习,得到预设元模型的目标元模型参数;
    模型确定模块,用于确定预设的对应目标部位的初始CT图像去噪模型,并根据所述目标元模型参数更新所述初始CT图像去噪模型,得到目标CT图像去噪模型;
    去噪处理模块,用于获取待去噪的第二CT噪声图像,并调用所述目标CT图像去噪模型处理所述第二CT噪声图像,得到第二CT去噪图像;
    损失检测模块,用于计算对应所述目标部位的第二CT真实图像与所述第二CT去噪图像的损失值,并检测所述损失值是否小于预设损失阈值要求;
    图像确定模块,用于当检测结果为所述损失值小于预设损失阈值要求时,调用所述目标CT图像去噪模型处理所述第二CT真实图像,得到目标CT图像。
  9. 一种计算机设备,其中,所述计算机设备包括处理器,所述处理器用于执行存储器中 存储的计算机可读指令以实现以下步骤:
    加噪处理第一CT真实图像集,得到第一CT噪声图像集;
    根据预设部位分类所述第一CT噪声图像集,得到第一CT噪声图像子集,并根据所述第一CT噪声图像子集配置任务集;
    基于所述任务集进行元迁移学习,得到预设元模型的目标元模型参数;
    确定预设的对应目标部位的初始CT图像去噪模型,并根据所述目标元模型参数更新所述初始CT图像去噪模型,得到目标CT图像去噪模型;
    获取待去噪的第二CT噪声图像,并调用所述目标CT图像去噪模型处理所述第二CT噪声图像,得到第二CT去噪图像;
    计算对应所述目标部位的第二CT真实图像与所述第二CT去噪图像的损失值,并检测所述损失值是否小于预设损失阈值要求;
    当检测结果为所述损失值小于预设损失阈值要求时,调用所述目标CT图像去噪模型处理所述第二CT真实图像,得到目标CT图像。
  10. 根据权利要求9所述的计算机设备,其中,所述处理器执行所述计算机可读指令以实现所述加噪处理第一CT真实图像集,得到第一CT噪声图像集时,包括:
    裁剪处理所述第一CT真实图像集,得到预设尺寸的第一CT真实图像集;
    获取预设强度的第一高斯噪声;
    添加所述第一高斯噪声至裁剪后的所述第一CT真实图像集中,得到第一CT噪声图像集。
  11. 根据权利要求9所述的计算机设备,其中,所述处理器执行所述计算机可读指令以实现所述根据预设部位分类所述第一CT噪声图像集,得到第一CT噪声图像子集,并根据所述第一CT噪声图像子集配置任务集时,包括:
    获取所述第一CT噪声图像集中每一第一CT噪声图像的部位编码;
    选取所述部位编码一致的所述第一CT噪声图像为第一CT噪声图像子集;
    确定每一所述第一CT噪声图像子集为一个任务;
    组合所述任务,得到任务集。
  12. 根据权利要求9所述的计算机设备,其中,所述处理器执行所述计算机可读指令以实现所述基于所述任务集进行元迁移学习,得到预设元模型的目标元模型参数时,包括:
    拆分所述任务集,得到训练任务集与验证任务集;
    计算所述训练任务集中每一训练任务对应所述预设元模型的初始元模型参数,得到初始元模型参数集;
    基于所述初始元模型参数集调用所述验证任务集计算所述预设元模型的总损失函数;
    采用随机梯度下降算法优化所述总损失函数,并检测优化后的所述总损失函数是否收敛;
    当检测结果为优化后的所述总损失函数收敛时,遍历所述预设元模型的网络层,得 到各个网络层的预设参数作为目标元模型参数。
  13. 根据权利要求9所述的计算机设备,其中,所述处理器执行所述计算机可读指令以实现所述根据所述目标元模型参数更新所述初始CT图像去噪模型,得到目标CT图像去噪模型时,包括:
    获取所述初始CT图像去噪模型的初始模型参数;
    对比所述初始模型参数与所述目标元模型参数,得到差异模型参数;
    根据所述差异模型参数调整所述初始模型参数以更新所述初始CT图像去噪模型,得到目标CT图像去噪模型。
  14. 根据权利要求9所述的计算机设备,其中,所述处理器执行所述计算机可读指令以实现所述获取待去噪的第二CT噪声图像时,包括:
    获取对应所述目标部位的第二CT真实图像;
    确定预定强度的第二高斯噪声;
    添加所述第二高斯噪声至所述第二CT真实图像中,得到待去噪的第二CT噪声图像。
  15. 根据权利要求9所述的计算机设备,其中,所述处理器执行所述计算机可读指令还用以实现以下步骤:
    当检测结果为所述损失值未满足预设损失阈值要求时,根据所述第二CT真实图像与所述第二CT去噪图像获取所述目标CT图像去噪模型的模型损失函数;
    采用梯度下降算法优化所述模型损失函数,直至所述模型损失函数收敛;
    获取所述模型损失函数收敛时的模型参数,并根据所述模型参数建立目标CT图像去噪模型。
  16. 一种计算机可读存储介质,所述计算机可读存储介质上存储有计算机可读指令,其中,所述计算机可读指令被处理器执行时实现以下步骤:
    加噪处理第一CT真实图像集,得到第一CT噪声图像集;
    根据预设部位分类所述第一CT噪声图像集,得到第一CT噪声图像子集,并根据所述第一CT噪声图像子集配置任务集;
    基于所述任务集进行元迁移学习,得到预设元模型的目标元模型参数;
    确定预设的对应目标部位的初始CT图像去噪模型,并根据所述目标元模型参数更新所述初始CT图像去噪模型,得到目标CT图像去噪模型;
    获取待去噪的第二CT噪声图像,并调用所述目标CT图像去噪模型处理所述第二CT噪声图像,得到第二CT去噪图像;
    计算对应所述目标部位的第二CT真实图像与所述第二CT去噪图像的损失值,并检测所述损失值是否小于预设损失阈值要求;
    当检测结果为所述损失值小于预设损失阈值要求时,调用所述目标CT图像去噪模型处理所述第二CT真实图像,得到目标CT图像。
  17. 根据权利要求16所述的计算机可读存储介质,其中,所述计算机可读指令被处理器执行以实现所述加噪处理第一CT真实图像集,得到第一CT噪声图像集包括:
    裁剪处理所述第一CT真实图像集,得到预设尺寸的第一CT真实图像集;
    获取预设强度的第一高斯噪声;
    添加所述第一高斯噪声至裁剪后的所述第一CT真实图像集中,得到第一CT噪声图像集。
  18. 根据权利要求16所述的计算机可读存储介质,其中,所述计算机可读指令被处理器执行以实现所述根据预设部位分类所述第一CT噪声图像集,得到第一CT噪声图像子集,并根据所述第一CT噪声图像子集配置任务集包括:
    获取所述第一CT噪声图像集中每一第一CT噪声图像的部位编码;
    选取所述部位编码一致的所述第一CT噪声图像为第一CT噪声图像子集;
    确定每一所述第一CT噪声图像子集为一个任务;
    组合所述任务,得到任务集。
  19. 根据权利要求16所述的计算机可读存储介质,其中,所述计算机可读指令被处理器执行以实现所述基于所述任务集进行元迁移学习,得到预设元模型的目标元模型参数包括:
    拆分所述任务集,得到训练任务集与验证任务集;
    计算所述训练任务集中每一训练任务对应所述预设元模型的初始元模型参数,得到初始元模型参数集;
    基于所述初始元模型参数集调用所述验证任务集计算所述预设元模型的总损失函数;
    采用随机梯度下降算法优化所述总损失函数,并检测优化后的所述总损失函数是否收敛;
    当检测结果为优化后的所述总损失函数收敛时,遍历所述预设元模型的网络层,得到各个网络层的预设参数作为目标元模型参数。
  20. 根据权利要求16所述的计算机可读存储介质,其中,所述计算机可读指令被处理器执行以实现所述根据所述目标元模型参数更新所述初始CT图像去噪模型,得到目标CT图像去噪模型包括:
    获取所述初始CT图像去噪模型的初始模型参数;
    对比所述初始模型参数与所述目标元模型参数,得到差异模型参数;
    根据所述差异模型参数调整所述初始模型参数以更新所述初始CT图像去噪模型,得到目标CT图像去噪模型。
PCT/CN2021/096638 2021-02-26 2021-05-28 Ct图像去噪处理方法、装置、计算机设备及介质 WO2022178995A1 (zh)

Applications Claiming Priority (2)

Application Number Priority Date Filing Date Title
CN202110220871.4A CN112927152B (zh) 2021-02-26 2021-02-26 Ct图像去噪处理方法、装置、计算机设备及介质
CN202110220871.4 2021-02-26

Publications (1)

Publication Number Publication Date
WO2022178995A1 true WO2022178995A1 (zh) 2022-09-01

Family

ID=76172451

Family Applications (1)

Application Number Title Priority Date Filing Date
PCT/CN2021/096638 WO2022178995A1 (zh) 2021-02-26 2021-05-28 Ct图像去噪处理方法、装置、计算机设备及介质

Country Status (2)

Country Link
CN (1) CN112927152B (zh)
WO (1) WO2022178995A1 (zh)

Families Citing this family (1)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN113362259B (zh) * 2021-07-13 2024-01-09 商汤集团有限公司 图像降噪处理方法、装置、电子设备及存储介质

Citations (3)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US20190156524A1 (en) * 2017-11-20 2019-05-23 ClariPI Inc. Apparatus and method for ct image denoising based on deep learning
CN110163827A (zh) * 2019-05-28 2019-08-23 腾讯科技(深圳)有限公司 图像去噪模型的训练方法、图像去噪方法、装置及介质
CN111311506A (zh) * 2020-01-21 2020-06-19 辽宁师范大学 基于双残差网络的低剂量ct牙齿图像去噪方法

Family Cites Families (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
KR102184278B1 (ko) * 2018-11-21 2020-11-30 한국과학기술원 메타학습에 기반한 임의의 목표 데이터셋과 모델 구조로의 전이학습 방법 및 시스템
CN111476292B (zh) * 2020-04-03 2021-02-19 北京全景德康医学影像诊断中心有限公司 医学图像分类处理人工智能的小样本元学习训练方法

Patent Citations (3)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US20190156524A1 (en) * 2017-11-20 2019-05-23 ClariPI Inc. Apparatus and method for ct image denoising based on deep learning
CN110163827A (zh) * 2019-05-28 2019-08-23 腾讯科技(深圳)有限公司 图像去噪模型的训练方法、图像去噪方法、装置及介质
CN111311506A (zh) * 2020-01-21 2020-06-19 辽宁师范大学 基于双残差网络的低剂量ct牙齿图像去噪方法

Also Published As

Publication number Publication date
CN112927152A (zh) 2021-06-08
CN112927152B (zh) 2024-02-02

Similar Documents

Publication Publication Date Title
Vásquez-Morales et al. Explainable prediction of chronic renal disease in the colombian population using neural networks and case-based reasoning
CN112037912B (zh) 基于医疗知识图谱的分诊模型训练方法、装置及设备
Liz et al. Ensembles of convolutional neural network models for pediatric pneumonia diagnosis
US20210225463A1 (en) System and Method with Federated Learning Model for Medical Research Applications
Yue et al. Comparison and validation of deep learning models for the diagnosis of pneumonia
WO2019157214A2 (en) Deep learning-based diagnosis and referral of diseases and disorders
Ghayvat et al. AI-enabled radiologist in the loop: novel AI-based framework to augment radiologist performance for COVID-19 chest CT medical image annotation and classification from pneumonia
WO2016073776A1 (en) System for management of health resources
US20230395196A1 (en) Method and system for quantifying cellular activity from high throughput sequencing data
US20210225513A1 (en) Method to Create Digital Twins and use the Same for Causal Associations
Patel et al. Artificial intelligence (AI) in Monkeypox infection prevention
Bushra et al. Automated detection of COVID-19 from X-ray images using CNN and Android mobile
Mahajan et al. COVID‐19 detection using hybrid deep learning model in chest x‐rays images
Robinson et al. Deep learning models for COVID-19 chest x-ray classification: Preventing shortcut learning using feature disentanglement
WO2022178995A1 (zh) Ct图像去噪处理方法、装置、计算机设备及介质
Wong et al. Applying machine learning in distributed data networks for pharmacoepidemiologic and pharmacovigilance studies: opportunities, challenges, and considerations
CN112435745B (zh) 就诊策略推荐方法、装置、电子设备及存储介质
Syed et al. Artificial intelligence methods in computer-aided diagnostic tools and decision support analytics for clinical informatics
Rawat et al. Case based Reasoning Technique in Digital Diagnostic System for Lung Cancer Detection
Khaniki et al. Enhancing Pneumonia Detection using Vision Transformer with Dynamic Mapping Re-Attention Mechanism
Yang et al. Automated facial recognition for Noonan syndrome using novel deep convolutional neural network with additive angular margin loss
Wahid et al. Pneumonia Detection in Chest X‐Ray Images Using Enhanced Restricted Boltzmann Machine
CN112802573B (zh) 药品包推荐方法、装置、计算机系统及可读存储介质
Joshi et al. Analysis of Smart Lung Tumour Detector and Stage Classifier Using Deep Learning Techniques with Internet of Things
US20210209489A1 (en) Processing a classifier

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: 21927428

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: 21927428

Country of ref document: EP

Kind code of ref document: A1