WO2022247004A1 - Procédé et appareil d'apprentissage de modèle de prédiction d'image, procédé et appareil d'application de modèle de prédiction d'image, dispositif et support de stockage - Google Patents

Procédé et appareil d'apprentissage de modèle de prédiction d'image, procédé et appareil d'application de modèle de prédiction d'image, dispositif et support de stockage Download PDF

Info

Publication number
WO2022247004A1
WO2022247004A1 PCT/CN2021/109475 CN2021109475W WO2022247004A1 WO 2022247004 A1 WO2022247004 A1 WO 2022247004A1 CN 2021109475 W CN2021109475 W CN 2021109475W WO 2022247004 A1 WO2022247004 A1 WO 2022247004A1
Authority
WO
WIPO (PCT)
Prior art keywords
treatment
level
image
predicted
oct image
Prior art date
Application number
PCT/CN2021/109475
Other languages
English (en)
Chinese (zh)
Inventor
张潇月
张成奋
吕彬
吕传峰
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 WO2022247004A1 publication Critical patent/WO2022247004A1/fr

Links

Images

Classifications

    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T7/00—Image analysis
    • G06T7/30—Determination of transform parameters for the alignment of images, i.e. image registration
    • 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
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/04—Architecture, e.g. interconnection topology
    • G06N3/045—Combinations of networks
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T3/00—Geometric image transformations in the plane of the image
    • G06T3/40—Scaling of whole images or parts thereof, e.g. expanding or contracting
    • G06T3/4023—Scaling of whole images or parts thereof, e.g. expanding or contracting based on decimating pixels or lines of pixels; based on inserting pixels or lines of pixels
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T2207/00—Indexing scheme for image analysis or image enhancement
    • G06T2207/30—Subject of image; Context of image processing
    • G06T2207/30004—Biomedical image processing
    • G06T2207/30041—Eye; Retina; Ophthalmic

Definitions

  • the present application relates to the technical field of image processing, and in particular to a training and application method, device, equipment, and storage medium of an image prediction model.
  • AMD age-related macular degeneration
  • nAMD wet AMD
  • VEGF anti-vascular endothelial growth factor
  • Optical coherence tomography is a high-resolution non-invasive imaging technology, which uses the principle of low-coherence interference to obtain tomographic capabilities in the depth direction, and can provide the most abundant fundus structure information. Applied fundus and retina disease examination means. Based on the prediction of OCT images in the industry, the pre-treatment OCT images are used as input, and the predicted post-treatment OCT images are used as output to provide intuitive assistance for doctors' decision-making. The inventors found that the existing technology often uses direct, pixel-level generation , the image prediction effect is not real and accurate enough.
  • the present application provides a training and application method, device, device, and storage medium of an image prediction model.
  • the trained first-level GAN model and the second-level GAN model are cascaded to obtain an image prediction model, and the therapeutic
  • the previous OCT image is input to the image prediction model, and the predicted OCT image after treatment is output, which realizes more accurate prediction at the image level and provides an intuitive reference for doctors to make decisions.
  • the present application provides a training method for an image prediction model, the training method comprising:
  • the first training set is input to the first-level generative confrontation network, and the first-level generative confrontation network is subjected to iterative optimization training to obtain the first-level generative confrontation network model;
  • the OCT images before and after the treatment are used as the second training set;
  • the second training set is input to the second-level generative confrontation network, and the second-level generative confrontation network is subjected to iterative optimization training to obtain the second-level generative confrontation network model;
  • the first-level generative confrontation network model and the second-level generative confrontation network model are cascaded to obtain an image prediction model.
  • the present application also provides an application method of an image prediction model, where the image prediction model is trained according to the above-mentioned training method, and the application method includes:
  • the deformed OCT image is input into the second-level generation confrontation network model of the image prediction model, and the final predicted OCT image after treatment is output.
  • the present application also provides a training device for an image prediction model, the training device comprising:
  • the first training set acquisition module is used to acquire the OCT images before and after treatment, perform image registration on the OCT images before and after the treatment, and obtain the first deformation field as the first training set;
  • the first training module is used to input the first training set to the first-level generative confrontation network, and perform iterative optimization training on the first-level generative confrontation network to obtain the first-level generative confrontation network model;
  • the second training set acquisition module is used to use the OCT images before and after the treatment as the second training set;
  • the second training module is used to input the second training set to the second-level generation confrontation network, and perform iterative optimization training on the second-level generation confrontation network to obtain the second-level generation confrontation network model;
  • a model cascading module configured to cascade the first-level generative adversarial network model and the second-level generative adversarial network model to obtain an image prediction model.
  • the present application also provides an application device for an image prediction model, the image prediction model is trained according to the above-mentioned training method, and the application device includes:
  • An image acquisition module configured to acquire an OCT image before treatment to be predicted
  • the deformation field output module is used to input the OCT image before the treatment to be predicted into the first stage of the image prediction model to generate a confrontation network model, and output the predicted deformation field;
  • An interpolation module configured to interpolate the pre-treatment OCT image to be predicted based on the predicted deformation field, and output a deformed OCT image
  • the image output module is used to input the deformed OCT image into the second-level generation confrontation network model of the image prediction model, and output the final predicted OCT image after treatment.
  • the present application also provides a computer device, the computer device includes a memory and a processor; the memory is used to store a computer program; the processor is used to execute the computer program and execute the The computer program implements the following steps:
  • the first training set is input to the first-level generative confrontation network, and the first-level generative confrontation network is subjected to iterative optimization training to obtain the first-level generative confrontation network model;
  • the OCT images before and after the treatment are used as the second training set;
  • the second training set is input to the second-level generative confrontation network, and the second-level generative confrontation network is subjected to iterative optimization training to obtain the second-level generative confrontation network model;
  • the deformed OCT image is input into the second-level generation confrontation network model of the image prediction model, and the final predicted OCT image after treatment is output.
  • the present application also provides a computer-readable storage medium, the computer-readable storage medium stores a computer program, and when the computer program is executed by a processor, the processor implements the following steps:
  • the first training set is input to the first-level generative confrontation network, and the first-level generative confrontation network is subjected to iterative optimization training to obtain the first-level generative confrontation network model;
  • the OCT images before and after the treatment are used as the second training set;
  • the second training set is input to the second-level generative confrontation network, and the second-level generative confrontation network is subjected to iterative optimization training to obtain the second-level generative confrontation network model;
  • the deformed OCT image is input into the second-level generation confrontation network model of the image prediction model, and the final predicted OCT image after treatment is output.
  • This application discloses a training and application method, device, equipment, and storage medium for an image prediction model.
  • an image prediction model Through the cascading of the two-stage generative adversarial network model for the prediction of the image deformation field and the prediction of the pixel, more accurate image prediction is realized.
  • the image prediction model trained by this method has the advantage of easier convergence, which improves the training efficiency of the model.
  • Fig. 1 is a schematic flowchart of a training method for an image prediction model provided by an embodiment of the present application
  • Fig. 2 is a schematic diagram of obtaining a deformation field by image registration provided by an embodiment of the present application
  • Fig. 3 is a schematic diagram of the network structure of the first-level generative confrontation network provided by the embodiment of the present application.
  • FIG. 4 is a schematic diagram of the network structure of the second-level generative confrontation network provided by the embodiment of the present application.
  • Fig. 5 is a schematic flowchart of an application method of an image prediction model provided by an embodiment of the present application
  • FIG. 6 is a schematic diagram of image prediction using an image prediction model provided by an embodiment of the present application.
  • FIG. 7 is a schematic block diagram of a training device for an image prediction model provided by an embodiment of the present application.
  • FIG. 8 is a schematic block diagram of an application device for an image prediction model provided by an embodiment of the present application.
  • FIG. 9 is a schematic structural block diagram of a computer device involved in an embodiment of the present application.
  • Embodiments of the present application provide a method, device, device, and storage medium for training and applying an image prediction model.
  • the training and application method of the image prediction model can be applied to the server, and the image prediction can be realized through the training and application of the image prediction model.
  • the server may be an independent server or a server cluster.
  • FIG. 1 is a schematic flowchart of a method for training an image prediction model provided by an embodiment of the present application.
  • the training method of the image prediction model can be applied to a server to obtain a trained image prediction model.
  • the training method of the image prediction model specifically includes steps S101 to S105.
  • the deformation field of the first deformation field is a field as large as the OCT images before and after treatment and is used to represent the coordinate mapping relationship of the OCT images before and after treatment.
  • the deformation field includes a plurality of two-dimensional vectors, and the two-dimensional vectors are in The pixel position saves the horizontal and vertical coordinates of the pixel to be sampled.
  • a plurality of optical coherence tomography (OCT) images of the macula are acquired before and after a diagnosis of wet age-related macular degeneration (nAMD) and anti-vascular endothelial growth factor (VEGF) treatment. Then image registration is performed on the multiple OCT images before and after treatment, and the first deformation field is obtained as the first training set.
  • OCT optical coherence tomography
  • image correction is performed on the OCT images before and after treatment, and the image correction includes image brightness correction and/or tilt correction.
  • the algorithm used for image registration may adopt the "VoxelMorph" algorithm.
  • the OCT image before treatment is used as the image to be registered (M)
  • the OCT image after treatment is used as the target image (F) which is input into the image registration network of unsupervised learning.
  • the image registration network defines a Parameterize the ⁇ function and optimize the parameter ⁇ of this function in the given dataset.
  • the registration domain can be obtained by directly calculating the ⁇ function through the parameter ⁇ obtained after optimization.
  • the registration domain is a first deformation field, and the first deformation field constitutes the first training set.
  • the network structure of the first-level generative adversarial network includes a first-level generator and a first-level discriminator.
  • the first-stage generator is a down-sampling and up-sampling cascaded structure.
  • the first-level generator adopts a U-Net structure, and the U-Net network structure includes a convolutional layer, a maximum pooling layer (down-sampling), a deconvolution layer (up-sampling) and a nonlinear activation function (ReLU).
  • the convolutional layer in the first-level generator is used to extract the multi-scale features in the pre-treatment OCT image, wherein the features of different scales reflect different image features, and the features of the shallower scale reflect the images of the shallower layers Features such as edges, etc., and deeper-scale features reflect deeper image features such as object outlines.
  • the first-level generator network structure of this application skips connections between the same scales, which can alleviate the gradient disappearance problem caused by increasing the depth in the deep neural network.
  • the first-level discriminator is a downsampling structure.
  • the first-level discriminator may adopt a VGGNet network structure.
  • the VGGNet network structure replaces the large-size filter with a small-size filter, which reduces the parameters of deep training, increases the number of nonlinear transformations, and improves the feature learning ability of the convolutional neural network.
  • the first-level discriminator may adopt a residual network structure (ResNet).
  • ResNet residual network structure
  • the characteristic of the residual network structure is that it is easy to optimize, and the accuracy can be improved by increasing the depth.
  • the residual block inside the residual network uses skip connections, which can alleviate the gradient disappearance problem caused by increasing the depth in the deep neural network.
  • Iterative optimization training is carried out on the first-level generative confrontation network to obtain the first-level generative confrontation network model, which specifically includes: inputting the OCT image before the treatment to the first-level generator, and using the first deformation field as The training target is to predict the deformation of the image by generating the second deformation field; the loss function calculation is performed on the first deformation field and the second deformation field by the first-level discriminator to obtain the first loss function value; according to The first loss function value optimizes the parameters of the first-level generative confrontation network, and through iterative optimization training, when the first loss function value is lower than the first preset threshold, the first-level generative confrontation network model is obtained .
  • Iterative optimization training is the process of generating an adversarial network to obtain optimal parameters through learning.
  • the specific iterative optimization rules are: the generator generates a predicted value under the initial parameters, and the discriminator uses the loss function to calculate the similarity between the predicted value and the real value to evaluate the parameters. , to optimize the parameters according to the evaluation results.
  • the generator regenerates the predicted value under the optimized parameters, and the discriminator uses the loss function calculation to evaluate the similarity between the predicted value and the real value again to evaluate and optimize the parameters. Through such an iterative calculation process, the optimal parameters are obtained.
  • the gradient descent method may be used for the iterative optimization training, which is not specifically limited in this application.
  • the OCT images before and after treatment acquired in step S101 are used as the second training set.
  • the OCT images before and after treatment that have undergone the above image correction processing may also be used as the second training set, and the image correction processing includes image correction including image brightness correction and/or tilt correction.
  • the network structure of the second-level generative adversarial network includes a second-level generator and a second-level discriminator.
  • the second-level generator adopts a U-Net network structure. In another embodiment, the second-level generator adopts a residual network structure (ResNet), and of course other network structures can also be used.
  • ResNet residual network structure
  • the second-level discriminator adopts a VGGNet network structure. In another embodiment, the second-level discriminator adopts a residual network structure (ResNet), and of course other network structures can also be used.
  • ResNet residual network structure
  • Carrying out iterative optimization training on the second-level generative confrontation network to obtain the second-level generative confrontation network model which specifically includes: inputting the OCT image before the treatment to the second-level generator to obtain the predicted OCT image after the treatment , taking the OCT image after treatment as the training target, predicting the pixel change of the image through the predicted OCT image after treatment; Perform loss function calculation on the final OCT image to obtain a second loss function value; optimize the parameters of the second-level generating confrontation network according to the second loss function value, when the second loss function value is lower than the second preset When the threshold value is , the second-level generative confrontation network model is obtained.
  • first preset threshold and second preset threshold may be set based on actual conditions, which is not specifically limited in the present application.
  • the loss function calculation uses the mean absolute error function, and the calculation method of the mean absolute error function is:
  • MAE is the loss function value
  • f(xi ) is the predicted value
  • y i is the real value
  • i corresponds to the i-th group of data in the training set
  • n is the total number of training sets
  • y i is used as a set of real values to represent the first deformation field corresponding to the i-th OCT image before treatment
  • f(xi ) is For the predicted second deformation field generated from the i-th OCT image before treatment
  • MAE is the average value of the sum of errors between the predicted second deformation field and the first deformation field.
  • y i is represented as a set of real values as the post-treatment OCT image corresponding to the i-th OCT image before treatment
  • f( xi ) is the predicted post-treatment OCT image generated for the i-th pre-treatment OCT image
  • MAE is the average value of the sum of errors between the predicted post-treatment OCT image and the real post-treatment OCT image.
  • S105 Concatenate the first-level GAN model and the second-level GAN model to obtain an image prediction model.
  • the cascading of the first-level GAN model and the second-level GAN model specifically refers to the result of interpolating the input of the first-level GAN model based on the output of the first-level GAN model, as the result of the first-level GAN model
  • the input of the second-level generative confrontation network model, and the output of the second-level generative confrontation network model are used as the output of the entire image prediction model.
  • Image interpolation is the use of known gray values of adjacent pixels to generate gray values of unknown pixels, so as to reproduce an image with higher resolution from the original image.
  • the application of image interpolation in this application is to interpolate the pixels of the pre-treatment OCT image based on the deformation field output by the first-level GAN model to obtain the pixel value of the deformed OCT image, and to obtain the pixel value of the deformed OCT image.
  • the image serves as the input to the second-level Generative Adversarial Network model.
  • the image prediction model of the present application is obtained through the cascade of the two-stage generative adversarial network model of the prediction of the image deformation field and the prediction of the pixel, more accurate image prediction is achieved, and the image prediction model trained by this training method It has the advantage of easier convergence and improves the training efficiency of the model.
  • FIG. 5 is a schematic flowchart of an application method of an image prediction model provided by an embodiment of the present application.
  • the image prediction model is trained according to the above method.
  • the application method of the image prediction model can be applied in a server, so as to realize OCT image prediction after treatment on the OCT image before treatment.
  • the application method of the image prediction model specifically includes steps S201 to S204.
  • FIG. 6 is a schematic diagram of an application method of an image prediction model provided by an embodiment of the present application.
  • OCT optical coherence tomography
  • nAMD wet age-related macular degeneration
  • the OCT image to be predicted before treatment is input to the first-level generative confrontation network model of the image prediction model, and the predicted deformation field is output.
  • an interpolation operation is performed to output a deformed OCT image.
  • the deformed OCT image is input to the second-level generative confrontation network model of the image prediction model, and the final predicted OCT image after treatment is output.
  • the interpolation operation is specifically: according to the abscissa and ordinate of the pixel to be sampled stored at the pixel position in the predicted deformation field, double-line the OCT image before treatment in the abscissa and ordinate directions.
  • Linear interpolation the pixel value obtained by bilinear interpolation is the pixel value of the deformed OCT image, and each pixel of the deformed OCT image is traversed to obtain the pixel value according to the above bilinear interpolation method, and the deformed OCT image can be obtained OCT images.
  • This application uses the first-level generative confrontation network model to predict the deformation after treatment, and the second-level generative confrontation network model to predict the pixel-level changes of partial effusion and high anti-points from nothing to existence or from existence to non-existence after treatment. More accurate prediction at the image level; at the same time, it is more in line with the characteristics of biological drug responses through the method of generation from deformation to pixel level.
  • the image prediction results output by applying the image prediction model of the present application provide accurate and comprehensive references for doctors.
  • an embodiment of the present application provides a schematic block diagram of an apparatus for training an image prediction model, and the apparatus for training an image prediction model is used to implement the aforementioned method for training an image prediction model.
  • the training device of the image prediction model can be configured in a server.
  • the training device 400 of the image prediction model specifically includes:
  • the first training set acquisition module 401 is configured to acquire OCT images before and after treatment, perform image registration on the OCT images before and after treatment, and obtain a first deformation field as a first training set;
  • the first training module 402 is used to input the first training set into the first-level generative confrontation network, and perform iterative optimization training on the first-level generative confrontation network to obtain the first-level generative confrontation network model;
  • the second training set acquisition module 403 is used to use the OCT images before and after the treatment as the second training set;
  • the second training module 404 is used to input the second training set into the second-level generative confrontation network, and perform iterative optimization training on the second-level generative confrontation network to obtain the second-level generative confrontation network model;
  • the model cascading module 405 is used for cascading the first-level generation confrontation network model and the second-level generation confrontation network model to obtain an image prediction model.
  • the above-mentioned apparatus can be implemented in the form of a computer program, and the computer program can be run on the computer device as shown in FIG. 9 .
  • an embodiment of the present application provides a schematic block diagram of an image prediction model application device, and the image prediction model application device is used to execute the aforementioned image prediction model application method.
  • the application device of the image prediction model can be configured in a server.
  • the training device 500 of the image prediction model includes:
  • An image acquisition module 501 configured to acquire an OCT image before treatment to be predicted
  • the deformation field output module 502 is used to input the OCT image before the treatment to be predicted into the first level of the image prediction model to generate the confrontation network model, and output the predicted deformation field;
  • An interpolation module 503, configured to interpolate the OCT image to be predicted before treatment based on the predicted deformation field, and output a deformed OCT image;
  • the image output module 504 is configured to input the deformed OCT image into the second-level generative confrontation network model of the image prediction model, and output the final predicted OCT image after treatment.
  • the above-mentioned apparatus can be implemented in the form of a computer program, and the computer program can be run on the computer device as shown in FIG. 9 .
  • FIG. 9 is a schematic structural block diagram of a computer device provided by an embodiment of the present application.
  • the computer device may be a server.
  • the computer device includes a processor, a memory, and a network interface connected through a system bus, where the memory may include a non-volatile storage medium and an internal memory.
  • Non-volatile storage media can store operating systems and computer programs.
  • the computer program includes program instructions.
  • the processor can execute any one of the image prediction model training method or the image prediction model application method.
  • the processor is used to provide computing and control capabilities and support the operation of the entire computer equipment.
  • the internal memory provides an environment for running the computer program in the non-volatile storage medium.
  • the processor can execute any one of the image prediction model training method or the image prediction model application method.
  • This network interface is used for network communication, such as sending assigned tasks, etc.
  • FIG. 9 is only a block diagram of a part of the structure related to the solution of this application, and does not constitute a limitation on the computer equipment on which the solution of this application is applied.
  • the specific computer equipment can be More or fewer components than shown in the figures may be included, or some components may be combined, or have a different arrangement of components.
  • the processor may be a central processing unit (Central Processing Unit, CPU), and the processor may also be other general processors, digital signal processors (Digital Signal Processor, DSP), application specific integrated circuits (Application Specific Integrated Circuit, ASIC), Field-Programmable Gate Array (Field-Programmable Gate Array, FPGA) or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc.
  • the general-purpose processor may be a microprocessor or the processor may be any conventional processor or the like.
  • the processor is used to run a computer program stored in the memory to implement the following steps:
  • the first training set is input to the first-level generative confrontation network, and the first-level generative confrontation network is subjected to iterative optimization training to obtain the first-level generative confrontation network model;
  • the OCT images before and after the treatment are used as the second training set;
  • the second training set is input to the second-level generative confrontation network, and the second-level generative confrontation network is subjected to iterative optimization training to obtain the second-level generative confrontation network model;
  • the first-level generative confrontation network model and the second-level generative confrontation network model are cascaded to obtain an image prediction model.
  • the processor when the processor performs image registration on the OCT images before and after treatment, and obtains the first deformation field as the first training set, it is used to realize:
  • the registration domain is a first deformation field
  • the first deformation field constitutes the first training set.
  • the processor performs iterative optimization training on the first-level generative confrontation network for realizing:
  • the network structure of the first-level generated confrontation network includes: a first-level generator and a first-level discriminator;
  • the OCT image before the treatment is input to the first-level generator, and the first deformation field is used as a training target to predict the deformation of the image by generating a second deformation field;
  • the first loss function value optimizes the parameters of the first-level generative confrontation network, and through iterative optimization training, when the first loss function value is lower than the first preset threshold, the first-level generative confrontation network is obtained.
  • the processor when the processor implements iterative optimization training on the second-level generative confrontation network, it is used to implement:
  • the iterative optimization training is carried out on the second-level generating confrontation network to obtain the second-level generating confrontation network model, including:
  • the parameters of the second-level generative adversarial network are optimized according to the second loss function value, and when the second loss function value is lower than a second preset threshold, a second-level generative adversarial network model is obtained.
  • the processor when the processor calculates the loss function, it is used to realize:
  • the calculation of the loss function uses the mean absolute error function, and the mean absolute error function is:
  • MAE is the loss function value
  • f( xi ) is the predicted value
  • y i is the real value
  • i corresponds to the i-th group of data in the training set
  • f( xi ) is the predicted second deformation field generated for the i-th OCT image before treatment
  • y i is the i-th OCT image
  • MAE is the average value of the error sum between the predicted second deformation field and the first deformation field
  • f( xi ) is the predicted OCT image after treatment generated for the i-th OCT image before treatment
  • y i is the OCT image after the treatment.
  • MAE is the average value of the sum of errors between the predicted post-treatment OCT image and the real post-treatment OCT image.
  • the processor when the processor realizes applying the image prediction model for image prediction, it is used to realize:
  • the deformed OCT image is input into the second-level generation confrontation network model of the image prediction model, and the final predicted OCT image after treatment is output.
  • Embodiments of the present application also provide a computer-readable storage medium, the computer-readable storage medium stores a computer program, the computer program includes program instructions, and the processor executes the program instructions to implement the present application. Any one of the methods for training an image prediction model or the method for applying an image prediction model provided in the embodiments.
  • the computer-readable storage medium may be an internal storage unit of the computer device described in the foregoing embodiments, such as a hard disk or a memory of the computer device.
  • the computer-readable storage medium can also be an external storage device of the computer device, such as a plug-in hard disk equipped on the computer device, a smart memory card (Smart Media Card, SMC), a secure digital (Secure Digital, SD ) card, flash memory card (Flash Card), etc.
  • the computer-readable storage medium may be non-volatile or volatile.

Landscapes

  • Engineering & Computer Science (AREA)
  • Theoretical Computer Science (AREA)
  • Physics & Mathematics (AREA)
  • General Physics & Mathematics (AREA)
  • Data Mining & Analysis (AREA)
  • Evolutionary Computation (AREA)
  • Life Sciences & Earth Sciences (AREA)
  • Artificial Intelligence (AREA)
  • General Engineering & Computer Science (AREA)
  • Computer Vision & Pattern Recognition (AREA)
  • Health & Medical Sciences (AREA)
  • Mathematical Physics (AREA)
  • Biophysics (AREA)
  • General Health & Medical Sciences (AREA)
  • Molecular Biology (AREA)
  • Computing Systems (AREA)
  • Biomedical Technology (AREA)
  • Computational Linguistics (AREA)
  • Software Systems (AREA)
  • Bioinformatics & Cheminformatics (AREA)
  • Bioinformatics & Computational Biology (AREA)
  • Evolutionary Biology (AREA)
  • Image Processing (AREA)
  • Eye Examination Apparatus (AREA)
  • Image Analysis (AREA)
  • Management, Administration, Business Operations System, And Electronic Commerce (AREA)

Abstract

La présente demande se rapporte aux domaines techniques de l'intelligence artificielle et du traitement d'image et concerne en particulier un procédé et un appareil d'apprentissage de modèle de prédiction d'image, un procédé et un appareil d'application de modèle de prédiction d'image, ainsi qu'un dispositif et un support de stockage, permettant d'effectuer une prédiction d'image OCT à l'aide de réseaux antagonistes génératifs en cascade. Le procédé d'apprentissage de modèle de prédiction d'image consiste à : obtenir des images OCT avant et après le traitement, puis effectuer un enregistrement d'image sur les images OCT avant et après le traitement afin d'obtenir des premiers champs de déformation en tant que premier ensemble d'apprentissage ; entrer le premier ensemble d'apprentissage dans un réseau antagoniste génératif de premier niveau, puis effectuer un apprentissage d'optimisation itératif sur le réseau antagoniste génératif de premier niveau afin d'obtenir un modèle de réseau antagoniste de premier niveau ; utiliser les images OCT avant et après le traitement comme second ensemble d'apprentissage ; entrer le second ensemble d'apprentissage dans un réseau antagoniste génératif de second niveau, puis effectuer un apprentissage d'optimisation itératif sur le réseau antagoniste génératif de second niveau afin d'obtenir un modèle de réseau antagoniste génératif de second niveau ; et mettre en cascade le modèle de réseau antagoniste génératif de premier niveau et le modèle de réseau antagoniste génératif de second niveau afin d'obtenir un modèle de prédiction d'image.
PCT/CN2021/109475 2021-05-25 2021-07-30 Procédé et appareil d'apprentissage de modèle de prédiction d'image, procédé et appareil d'application de modèle de prédiction d'image, dispositif et support de stockage WO2022247004A1 (fr)

Applications Claiming Priority (2)

Application Number Priority Date Filing Date Title
CN202110572884.8 2021-05-25
CN202110572884.8A CN113269812B (zh) 2021-05-25 2021-05-25 图像预测模型的训练及应用方法、装置、设备、存储介质

Publications (1)

Publication Number Publication Date
WO2022247004A1 true WO2022247004A1 (fr) 2022-12-01

Family

ID=77232797

Family Applications (1)

Application Number Title Priority Date Filing Date
PCT/CN2021/109475 WO2022247004A1 (fr) 2021-05-25 2021-07-30 Procédé et appareil d'apprentissage de modèle de prédiction d'image, procédé et appareil d'application de modèle de prédiction d'image, dispositif et support de stockage

Country Status (2)

Country Link
CN (1) CN113269812B (fr)
WO (1) WO2022247004A1 (fr)

Families Citing this family (1)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN117952961B (zh) * 2024-03-25 2024-06-07 深圳大学 图像预测模型的训练及应用方法、设备及可读存储介质

Citations (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN111047629A (zh) * 2019-11-04 2020-04-21 中国科学院深圳先进技术研究院 多模态图像配准的方法、装置、电子设备及存储介质
CN112512632A (zh) * 2019-07-22 2021-03-16 北京市肿瘤防治研究所 一种放射治疗出射束监测方法和系统

Family Cites Families (1)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US10636141B2 (en) * 2017-02-09 2020-04-28 Siemens Healthcare Gmbh Adversarial and dual inverse deep learning networks for medical image analysis

Patent Citations (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN112512632A (zh) * 2019-07-22 2021-03-16 北京市肿瘤防治研究所 一种放射治疗出射束监测方法和系统
CN111047629A (zh) * 2019-11-04 2020-04-21 中国科学院深圳先进技术研究院 多模态图像配准的方法、装置、电子设备及存储介质

Non-Patent Citations (2)

* Cited by examiner, † Cited by third party
Title
LIU YUTONG, YANG JINGYUAN, ZHOU YANG, WANG WEISEN, ZHAO JIANCHUN, YU WEIHONG, ZHANG DINGDING, DING DAYONG, LI XIRONG, CHEN YOUXIN: "Conclusion", BRITISH JOURNAL OF OPHTHALMOLOGY, BMJ PUBLISHING GROUP, GB, vol. 104, no. 12, 23 December 2020 (2020-12-23), GB , pages 1735 - 1740, XP055798665, ISSN: 0007-1161, DOI: 10.1136/bjophthalmol-2019-315338 *
LIU YUTONG: "Prediction of OCT Images of Short-Term Response to Anti-VEGF Treatment for Neovascular Age-related Macular Degeneration (AMD) Using Generative Adversarial Network", CHINESE DOCTORAL DISSERTATIONS FULL-TEXT DATABASE, UNIVERSITY OF CHINESE ACADEMY OF SCIENCES, CN, no. 5, 15 May 2021 (2021-05-15), CN , XP093008562, ISSN: 1674-022X *

Also Published As

Publication number Publication date
CN113269812B (zh) 2024-05-07
CN113269812A (zh) 2021-08-17

Similar Documents

Publication Publication Date Title
CN109829894B (zh) 分割模型训练方法、oct图像分割方法、装置、设备及介质
JP7058373B2 (ja) 医療画像に対する病変の検出及び位置決め方法、装置、デバイス、及び記憶媒体
CN110782421B (zh) 图像处理方法、装置、计算机设备及存储介质
JP7155271B2 (ja) 画像処理システム及び画像処理方法
CN109389552B (zh) 一种基于上下文相关多任务深度学习的图像超分辨算法
JP2024056701A (ja) 深層学習法を使用する３ｄ歯顎顔面構造の分類および３ｄモデリング
CN108198184B (zh) 造影图像中血管分割的方法和系统
CA3078095A1 (fr) Classification et taxonomie automatisees de donnees de dents 3d a l'aide de procedes d'apprentissage profond
US11669729B2 (en) Model training method and apparatus
CN112215755B (zh) 一种基于反投影注意力网络的图像超分辨率重建方法
WO2022089079A1 (fr) Procédé, appareil et système de traitement d'image, et dispositif et support de stockage lisible par ordinateur
CN111161269B (zh) 图像分割方法、计算机设备和可读存储介质
WO2021136368A1 (fr) Procédé et appareil pour détecter automatiquement une zone principale pectorale dans une image cible de molybdène
WO2021017006A1 (fr) Procédé et appareil de traitement d'image, réseau neuronal et procédé d'apprentissage, et support d'enregistrement
WO2021082819A1 (fr) Procédé et appareil de génération d'image, et dispositif électronique
US20210358129A1 (en) Machine learning method, machine learning device, and machine learning program
US20230377314A1 (en) Out-of-distribution detection of input instances to a model
JP7083015B2 (ja) 超解像処理装置、方法及びプログラム
WO2024045442A1 (fr) Procédé d'entraînement de modèle de correction d'image, procédé de correction d'image, dispositif et support de stockage
Wannenwetsch et al. Probabilistic pixel-adaptive refinement networks
WO2022247004A1 (fr) Procédé et appareil d'apprentissage de modèle de prédiction d'image, procédé et appareil d'application de modèle de prédiction d'image, dispositif et support de stockage
CN109961435B (zh) 脑图像获取方法、装置、设备及存储介质
CN111339993A (zh) 一种x射线图像金属检测方法和系统
KR102476888B1 (ko) 디지털 병리이미지의 인공지능 진단 데이터 처리 장치 및 그 방법
WO2020241337A1 (fr) Dispositif de traitement d'image

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: 21942573

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: 21942573

Country of ref document: EP

Kind code of ref document: A1