WO2021057174A1 - 图像处理方法及装置、电子设备、存储介质和计算机程序 - Google Patents

图像处理方法及装置、电子设备、存储介质和计算机程序 Download PDF

Info

Publication number
WO2021057174A1
WO2021057174A1 PCT/CN2020/100692 CN2020100692W WO2021057174A1 WO 2021057174 A1 WO2021057174 A1 WO 2021057174A1 CN 2020100692 W CN2020100692 W CN 2020100692W WO 2021057174 A1 WO2021057174 A1 WO 2021057174A1
Authority
WO
WIPO (PCT)
Prior art keywords
image
processing
sub
result
classification network
Prior art date
Application number
PCT/CN2020/100692
Other languages
English (en)
French (fr)
Inventor
韩泓泽
黄宁
Original Assignee
北京市商汤科技开发有限公司
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 北京市商汤科技开发有限公司 filed Critical 北京市商汤科技开发有限公司
Priority to KR1020217037806A priority Critical patent/KR20210153700A/ko
Priority to JP2021563034A priority patent/JP2022530413A/ja
Publication of WO2021057174A1 publication Critical patent/WO2021057174A1/zh

Links

Images

Classifications

    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F18/00—Pattern recognition
    • G06F18/20—Analysing
    • G06F18/24—Classification techniques
    • G06F18/241—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches
    • 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
    • G06F18/213—Feature extraction, e.g. by transforming the feature space; Summarisation; Mappings, e.g. subspace methods
    • 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
    • G06N3/08—Learning methods
    • G06N3/084—Backpropagation, e.g. using gradient descent
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T7/00—Image analysis
    • G06T7/0002—Inspection of images, e.g. flaw detection
    • G06T7/0012—Biomedical image inspection
    • G—PHYSICS
    • G16—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR SPECIFIC APPLICATION FIELDS
    • G16H—HEALTHCARE INFORMATICS, i.e. INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR THE HANDLING OR PROCESSING OF MEDICAL OR HEALTHCARE DATA
    • G16H50/00—ICT specially adapted for medical diagnosis, medical simulation or medical data mining; ICT specially adapted for detecting, monitoring or modelling epidemics or pandemics
    • G16H50/20—ICT specially adapted for medical diagnosis, medical simulation or medical data mining; ICT specially adapted for detecting, monitoring or modelling epidemics or pandemics for computer-aided diagnosis, e.g. based on medical expert systems
    • G—PHYSICS
    • G16—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR SPECIFIC APPLICATION FIELDS
    • G16H—HEALTHCARE INFORMATICS, i.e. INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR THE HANDLING OR PROCESSING OF MEDICAL OR HEALTHCARE DATA
    • G16H50/00—ICT specially adapted for medical diagnosis, medical simulation or medical data mining; ICT specially adapted for detecting, monitoring or modelling epidemics or pandemics
    • G16H50/50—ICT specially adapted for medical diagnosis, medical simulation or medical data mining; ICT specially adapted for detecting, monitoring or modelling epidemics or pandemics for simulation or modelling of medical disorders
    • 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
    • G06T2207/10—Image acquisition modality
    • G06T2207/10072—Tomographic images
    • G06T2207/10081—Computed x-ray tomography [CT]
    • 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
    • G06T2207/20—Special algorithmic details
    • G06T2207/20081—Training; Learning
    • 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
    • G06T2207/20—Special algorithmic details
    • G06T2207/20084—Artificial neural networks [ANN]
    • 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
    • G06T2207/30061—Lung

Definitions

  • the embodiments of the present application relate to the field of computer vision technology, and relate to, but are not limited to, an image processing method and device, electronic equipment, computer storage media, and computer programs.
  • the rapid development of deep learning has made remarkable achievements in the field of image processing.
  • the current image processing technology based on deep learning requires multiple processing procedures to obtain a prediction of the extent of lesion development, which leads to cumbersome processing, and the accuracy of the resulting prediction results is not high.
  • the embodiment of the application proposes an image processing method and device, electronic equipment, computer storage medium, and computer program.
  • the embodiment of the application provides an image processing method, including:
  • the change data is predicted to obtain the prediction result; the change data characterizes: the change situation of the target object in the first image processed based on the area distribution in the second image.
  • the classification result can be obtained by using a target classification network, thereby simplifying the image processing process, and the classification result is obtained by inputting the first image and the second image into the target classification network jointly. , And predict the change of the target object in the first image based on the regional distribution in the second image according to the classification result, thereby improving the accuracy of the prediction result.
  • the method further includes:
  • the trained classification network is used as the target classification network.
  • the trained classification network is used as the target classification network, and the target classification network can be used to obtain the classification result.
  • the target object in the first image is based on the target object in the second image.
  • the regional distribution is processed to predict the changes obtained, thereby improving the accuracy of the prediction results.
  • the training the classification network to obtain the trained classification network includes:
  • an image to be processed is obtained; the first image and the second image are different types of image data;
  • the image to be processed is input into the classification network for training as a training sample, and the trained classification network is obtained.
  • the classification network is trained based on the image to be processed after the image superposition processing of the first image and the second image is performed. Since these two types of image data are comprehensively considered and training is performed in the classification network , Therefore, the accuracy of training can be obtained.
  • the obtaining the image to be processed includes:
  • the first sub-image data and the second sub-image data are used as the image to be processed.
  • the first image and the second image are respectively cut according to the contour of the target object, and the cut first sub-image data and second sub-image data are obtained and used for training of the classification network , Can improve training efficiency.
  • the first sub-image data and the second sub-image data are image data of the same size.
  • the image data of the same size is used, and the pixel position alignment may not be required in the image superposition process, and the first sub-image data and the second sub-image data can be directly used to realize the image superposition, thereby improving The processing efficiency of image overlay.
  • the inputting the to-be-processed image as a training sample into the classification network for training to obtain the trained classification network includes:
  • the image data is converted into the corresponding histogram, and the histogram equalization process is performed, and the histogram distribution of the image can be changed to an approximately uniform distribution, thereby enhancing the contrast of the image and making the image More clear.
  • the inputting the to-be-processed image as a training sample into the classification network for training to obtain the trained classification network includes:
  • the distribution of the pixels can be summarized, so that the pixels that need to be processed can be limited to a preset range after being normalized.
  • the normalization process is to make the subsequent series of processes more convenient and quicker, which is beneficial to accelerate the convergence speed of the classification network training.
  • the classification network includes at least one classification processing module
  • the inputting the to-be-processed image as a training sample into the classification network for training to obtain the trained classification network includes:
  • the image to be processed is subjected to feature extraction, dimensionality reduction processing, and global average pooling processing through at least one classification processing module to obtain a loss function, and train the classification according to the back propagation of the loss function Network, thus, the target classification network is trained.
  • each classification processing module includes at least a convolutional layer
  • the performing feature extraction, dimensionality reduction processing, and global average pooling processing on the image to be processed through the at least one classification processing module to obtain a loss function includes:
  • the loss function is obtained.
  • the second processing result can be obtained according to the first processing result obtained after the dimensionality reduction processing, and the loss function can be obtained according to the second processing result and the manual labeling result, so as The back propagation of the function trains the classification network, thereby training the target classification network.
  • each residual module when the classification processing module is a residual module, each residual module includes: a convolutional layer, a regularization layer, and an activation layer;
  • the method further includes:
  • the structure of the module includes: a convolutional layer, a regularization layer, and an activation layer, which are obtained after feature extraction corresponding to the convolutional layer by the residual module
  • the second extraction result is obtained.
  • the third extraction result can be obtained, so that according to the third extraction result
  • the first processing result for calculating the loss function is obtained, and after the loss function is obtained, the classification network can be trained according to the back propagation of the loss function, thereby training the target classification network.
  • the performing dimensionality reduction processing to obtain the first processing result includes:
  • the third extraction result can be subjected to dimensionality reduction processing to obtain the first processing result for calculating the loss function.
  • the training station can be trained according to the back propagation of the loss function.
  • the classification network is described, thus, the target classification network is obtained by training.
  • An embodiment of the present application also provides an image processing device, which includes:
  • the classification part is configured to input the first image and the second image into the target classification network to obtain a classification result
  • the prediction part is configured to predict the change data according to the classification result to obtain the prediction result; the change data characterization: the target object in the first image is processed based on the area distribution in the second image The change situation.
  • the device further includes a training part configured to:
  • the trained classification network is used as the target classification network.
  • the training part includes:
  • the superposition sub-part is configured to perform image superposition processing on the first image and the second image to obtain an image to be processed; the first image and the second image are different types of image data;
  • the training sub-part is configured to input the to-be-processed image as a training sample into the classification network for training to obtain the trained classification network.
  • the training part further includes:
  • the cutting sub-part is configured to perform image cutting on the first image and the second image respectively according to the contour of the target object to obtain the first sub-image data and the second sub-image data after being cut;
  • the first sub-image data and the second sub-image data are used as the image to be processed.
  • the first sub-image data and the second sub-image data are image data of the same size.
  • the training part further includes:
  • the equalization processing sub-part is configured to convert the first sub-image data and the second sub-image data into corresponding histograms respectively, and perform equalization processing of the histograms to obtain an equalization processing result; based on the The equalization processing result trains the classification network to obtain the trained classification network.
  • the training part further includes:
  • the normalization processing sub-part is configured to perform normalization processing on the corresponding pixels contained in the first sub-image data and the second sub-image data to obtain a normalization processing result; based on the normalization
  • the processing result trains the classification network to obtain the trained classification network.
  • the classification network includes at least one classification processing module
  • the training sub-part is configured as:
  • the image to be processed is subjected to feature extraction, dimensionality reduction processing, and global average pooling processing through the at least one classification processing module to obtain a loss function; the classification network is trained according to the back propagation of the loss function to obtain the Describe the classification network after training.
  • each classification processing module includes at least a convolutional layer
  • the training sub-part is configured as:
  • each residual module when the classification processing module is a residual module, each residual module includes: a convolution layer, a regularization layer, and an activation layer;
  • the training sub-part is configured as:
  • the first extraction result obtained by performing feature extraction on the image to be processed through the corresponding convolutional layer in at least one residual module is processed by the regularization layer and the activation layer to obtain the second extraction result; according to the second
  • the extraction result and the image to be processed are extracted to obtain a third extraction result.
  • the training sub-part is configured to perform dimensionality reduction processing according to the third extraction result to obtain the first processing result.
  • An embodiment of the application also provides an electronic device, including:
  • a memory configured to store executable instructions of the processor
  • the processor is configured to execute any one of the above-mentioned image processing methods.
  • An embodiment of the present application also provides a computer-readable storage medium on which computer program instructions are stored, and when the computer program instructions are executed by a processor, any one of the above-mentioned image processing methods is implemented.
  • the embodiment of the present application also provides a computer program, including computer-readable code, when the computer-readable code runs in an electronic device, the processor in the electronic device executes any one of the above-mentioned image processing method.
  • the first image and the second image are input into the target classification network to obtain the classification result; according to the classification result, the target object in the first image is based on the area distribution in the second image
  • the changes obtained by the processing are predicted, and the predicted results are obtained. Since the classification result can be obtained by using a target classification network, the technical solution of the embodiment of the present application simplifies the image processing process, because the classification result is obtained by jointly inputting the first image and the second image into the target classification network , And predict the change of the target object in the first image based on the regional distribution in the second image according to the classification result. Therefore, the technical solution of the embodiment of the application is adopted to improve the prediction The accuracy of the results.
  • FIG. 1 is a schematic flowchart of an image processing method provided by an embodiment of the application
  • Figure 2 is a schematic diagram of an application scenario of an embodiment of the application
  • FIG. 3 is a schematic diagram of a training process of a classification network provided by an embodiment of the application.
  • FIG. 4 is a schematic diagram of a classification network architecture for implementing an image processing method provided by an embodiment of the application
  • FIG. 5 is a schematic structural diagram of an image processing device provided by an embodiment of the application.
  • FIG. 6 is a schematic structural diagram of an electronic device according to an embodiment of the application.
  • FIG. 7 is a schematic structural diagram of another electronic device according to an embodiment of the application.
  • the prediction result may be a prediction result of the degree of influence on the designated area when the lesions in the designated area are processed.
  • the lesions can be in the abdomen, lungs, kidneys, brain, heart, etc. If the designated area is the lungs, the location where the lesions exist in the lungs needs to be surgically treated to affect the lungs (such as severity or No severity, etc.) to make predictions.
  • Radiation therapy to the lungs may cause radiation pneumonitis.
  • Radiation pneumonitis is caused by damage to normal lung tissue in the radiation field after radiotherapy (such as lung cancer, breast cancer, esophageal cancer, malignant lymphoma or other malignant tumors of the chest).
  • the inflammation caused by it Mild cases are asymptomatic, and inflammation can dissipate on its own; in severe cases, extensive fibrosis occurs in the lungs, leading to respiratory damage and even respiratory failure.
  • the degree of inflammatory response is closely related to the radiation dose and the state of the lesion before radiotherapy. It is necessary to predict the severity of radiation pneumonia after radiotherapy for lung cancer.
  • the extraction of image features can be achieved through radiomics.
  • the extraction of image features through radioomics is to extract image features through radiographic imaging methods, and then study the relationship between the image features and clinical symptoms (such as the prediction of the degree of influence of the designated area).
  • the support vector machine (SVM) and other classifiers can be used to predict the degree of impact (such as severity or non-severity, etc.) on the specified area. It can be seen that the entire image processing process includes multiple stages, which is not only cumbersome, but also the accuracy of the prediction results obtained therefrom is not high. The accuracy is not high because:
  • the prediction of radiation dose used in radiotherapy is carried out separately from the prediction of the image processing process.
  • the dose constant can be obtained through the method of averaging the radiation dose in the entire lung to achieve the above prediction process.
  • gray ( Gray, Gy) units are used to measure the absorbed dose of radiation.
  • Doctors can count the percentage of tissues whose absorbed dose in the lung exceeds a certain value to the entire lung as the dose constant.
  • V20 is the percentage of tissue volume whose absorbed dose in the lung exceeds 20Gy to the total lung volume.
  • the treatment method using this dose constant is too general, and does not consider the different dose sizes at different lesions.
  • the lesions are in different areas such as the abdomen, lungs, kidneys, brain, heart, etc., and the dose sizes used are different.
  • radiation The impact afterwards is also different.
  • the internal radiation dose of the entire lung is relatively small, and the statistical constant is also relatively small, when the radiation hits key organs, such as the main trachea, blood vessels, and heart, it will also cause serious consequences.
  • the constant used in the treatment method of the dose constant is only a statistic, and does not take into account the spatial distribution of the radiation in different regions. As a result, the prediction accuracy rate obtained by the treatment method of the dose constant is also inferior. not tall.
  • the task of predicting the severity of pneumonia after radiotherapy is mainly solved by radioomics, which has disadvantages such as low efficiency, low robustness, failure to consider radiation distribution, and low accuracy.
  • radioomics methods give ideal accuracy, the optimization of feature selection, dose constants, and hyperparameter selection of SVM in the process make the method less robust and difficult to apply to other data sets.
  • the radiation dose can be constant, that is, the radiation dose in the entire lung or cancer area is counted as a constant, but this calculation loses the radiation distribution characteristics.
  • the target classification network (such as a classification neural network, which may be three-dimensional) obtained after deep learning training can combine the lung image with
  • the ray distribution image (both images can be three-dimensional images) is input into the target classification network at the same time, so that the target classification network can be used to comprehensively obtain the image and the image of the designated area where the lesion is located or the associated area at each location.
  • Ray distribution to improve the prediction accuracy, and through the classification of the target classification network, the severity of pneumonia that will occur after radiotherapy can be directly output in one step.
  • the embodiment of this application not only considers the distribution of radiation dose in the image processing prediction process, but also can be widely used on similar task data sets.
  • the embodiment of this application can be directly applied Predict the severity of radiation pneumonia without changing any parameters and structures, and the application scenarios are not limited to the prediction of lesions in different areas or related areas such as the abdomen, lungs, kidneys, brain, heart, etc., and can be quickly obtained Accurate prediction results.
  • FIG. 1 is a schematic flow chart of an image processing method provided by an embodiment of the application.
  • the image processing method is applied to an image processing apparatus.
  • the image processing apparatus can be executed by a terminal device or a server or other processing device, where the terminal device can be a user Equipment (UE, User Equipment), mobile devices, cellular phones, cordless phones, personal digital processing (PDA, Personal Digital Assistant), handheld devices, computing devices, in-vehicle devices, wearable devices, etc.
  • the image processing method may be implemented by a processor calling computer-readable instructions stored in a memory. As shown in Figure 1, the process includes:
  • Step S101 Input the first image and the second image into a target classification network to obtain a classification result.
  • the first image may be an image with a lesion (it may be a CT image of the area where the lesion is located), such as the lesion in different areas such as the abdomen, lung, kidney, brain, heart, etc. or related
  • the image of the area may be a radiation dose distribution map used to perform radiotherapy on the area where the lesion is located or the associated area.
  • the two images, the image with the lesion and the radiation dose distribution map can be jointly input into the target classification network to obtain the classification result.
  • Step S102 Predict the change data according to the classification result to obtain the prediction result.
  • the change data represents: the change of the target object in the first image based on the area distribution in the second image.
  • the target object may be the organ where the lesion is located, such as the abdomen, lung, kidney, brain, heart, etc.
  • the regional distribution can be the distribution of radiation doses used for different lesions in different regions.
  • the change may be the severity (such as the probability of being severe or the probability of not being serious) that may cause inflammation to the organ (such as the lung) where the lesion is located once radiotherapy is performed on the lesion.
  • the severity of inflammation in the lungs in the image with the lesion based on the radiation distribution in the radiation dose distribution map can be predicted to obtain the prediction result.
  • prediction can be realized only by classification according to the target classification network, and the prediction result can be obtained end-to-end in one step, without the need for multi-stage cumbersome operations.
  • the different effects of different radiation doses at different locations improve the accuracy of prediction.
  • the target classification network does not require artificial hyperparameters to control, but can use the target classification network obtained after deep learning training, which realizes the adaptive control of prediction in the entire image processing process, which helps to improve the prediction accuracy.
  • FIG. 2 is a schematic diagram of an application scenario of an embodiment of the present application, as shown in FIG. 2, a CT image of a lung cancer patient 201 is the above first image, and the radiation dose distribution map 202 is the above second image.
  • the image processing method described in the foregoing embodiment can be used to predict the severity of radiation pneumonia after radiotherapy. The severity of radiation pneumonia will be predicted. Therefore, it can help doctors predict the postoperative risk in advance, take precautions in advance or revise the radiotherapy plan.
  • a classification network may be trained to obtain a trained classification network; the trained classification network is used as the target classification network.
  • FIG. 3 is a schematic diagram of the training process of the classification network provided by an embodiment of the application.
  • the training process of the classification network can be implemented based on an image processing device.
  • the image processing device can be executed by a terminal device or a server or other processing device. It can be a UE, a mobile device, a cellular phone, a cordless phone, a PDA, a handheld device, a computing device, a vehicle-mounted device, a wearable device, etc.
  • the image processing method may be implemented by a processor calling computer-readable instructions stored in a memory.
  • the training process of the classification network may include:
  • Step S301 After performing image superposition processing on the first image and the second image, an image to be processed is obtained.
  • the first image and the second image may be different types of image data
  • image cutting may be performed on the first image and the second image respectively according to the contour of the target object, Obtain the cut first sub-image data and second sub-image data.
  • the outline of the target object is not the outline of the lesion, but the outline of the area where the lesion is located or the associated area, such as the entire lung outline of the lung where the lesion is located, or the outline of the heart or kidney where the lesion is located, depending on the difference. Area, using different doses.
  • the first sub-image data and the second sub-image data may be image data of the same size. For example, according to the lung contour, the three-dimensional first image and the second image are cut, and the two are the same size. Then, the first sub-image data and the second sub-image data can be used as the image to be processed.
  • the first sub-image data and the second sub-image data may also be image data of different sizes.
  • the first sub-image data and the second sub-image data may be combined
  • the corresponding pixels contained in the second sub-image data are subjected to pixel position alignment processing to obtain the aligned first sub-image data and the aligned second sub-image data, and the aligned first sub-image data and the aligned second sub-image data are obtained.
  • the image to be processed is obtained.
  • Step S302 Input the to-be-processed image as a training sample into the classification network for training, to obtain the trained classification network.
  • the classification network is trained through step S301 to step S302, and a trained classification network can be obtained.
  • the trained classification network is used as the target classification network.
  • inputting the to-be-processed image as a training sample into the classification network for training, and obtaining the trained classification network may include: combining the first sub-image data and the second sub-image data Converting into corresponding histograms respectively, and performing equalization processing of the histograms to obtain equalization processing results; training the classification network based on the equalization processing results to obtain the trained classification network.
  • inputting the to-be-processed image as a training sample into the classification network for training, and obtaining the trained classification network may include: combining the first sub-image data and the second sub-image data Normalization processing is performed on the corresponding pixels contained in to obtain a normalization processing result; the classification network is trained based on the normalization processing result to obtain the trained classification network.
  • the first sub-image data and the second sub-image data may be respectively converted into corresponding histograms, and the histogram equalization processing is performed to obtain the equalization processing result. Then, the corresponding pixel points contained in the first sub-image data and the second sub-image data in the equalization processing result can be normalized. For example, after performing histogram equalization on the two sub-images, the two sub-images are normalized and then combined into image data represented by a two-channel four-dimensional matrix. The image data is input into the classification network, and the image data is extracted layer by layer through the convolutional layer of the classification network and the dimensionality is reduced. Finally, through the processing of the fully connected layer, the severity of radiation inflammation after radiotherapy is obtained. Probability.
  • the classification network may include at least one classification processing module
  • inputting the to-be-processed image as a training sample into the classification network for training to obtain the target classification network may include: extracting the features of the to-be-processed image through the at least one classification processing module (such as by The convolutional layer performs feature extraction), dimensionality reduction processing (such as pooling processing), and global average pooling processing to obtain the loss function; training the classification network according to the back propagation of the loss function (for example, according to the loss The error calculated by the function is back-propagated) to obtain the trained classification network.
  • the convolutional layer performs feature extraction
  • dimensionality reduction processing such as pooling processing
  • global average pooling processing global average pooling processing
  • each classification processing module includes at least a convolutional layer; the image to be processed is subjected to feature extraction, dimensionality reduction processing, and global average pooling processing through the at least one classification processing module to obtain a loss
  • the function may include: performing layer-by-layer feature extraction of the image to be processed through the corresponding convolutional layer in at least one classification processing module, and then performing layer-by-layer dimensionality reduction processing to obtain the first processing result; and performing the global processing result on the first processing result. After averaging pooling, it is input to the fully connected layer to obtain the second processing result.
  • the second processing result is the prediction result output by the classification network, which is used to characterize the predicted change of the extracted features; according to the second processing result and the manual annotation result (such as The actual changes marked by the doctor), the loss function is obtained.
  • the loss function can be obtained according to the prediction result output by the classification network and the actual changes marked by the doctor. If the error between the predicted change reflected by the loss function and the real situation is within the preset range (for example, the error is zero), it means that the difference between the generated predicted change and the real situation has reached the convergence condition, and the classification network After the training is over, the target classification network after training is obtained.
  • each residual module may include: a convolution layer, a regularization layer, and an activation layer.
  • the second extraction result is obtained after the regularization layer and the activation layer are processed; according to the second extraction result and the image to be processed, the third extraction result is obtained, and the third extraction result is used for dimensionality reduction deal with. That is, the input of the residual module is the "image to be processed", and the final extraction result obtained by adding the input of the residual module and the output of the last active layer in the residual module is the third extraction result.
  • layer-by-layer dimensionality reduction processing can also be performed to obtain the first processing result. For example, performing layer-by-layer dimensionality reduction processing according to the third extraction result to obtain the first processing result.
  • FIG. 4 is a schematic diagram of a classification network architecture for implementing an image processing method provided by an embodiment of the application.
  • the classification network (such as a classification neural network) may include at least one classification processing module 11.
  • the classification processing module 11 may be a residual module 12 and may also include a fully connected layer 13.
  • Each residual module 12 may include: at least one convolutional layer 121, at least one regularization layer 122, and at least one activation layer 123; the classification network automatically learns useful features in the extracted image, and uses these features to make predictions, Instead of extracting features and then selecting features, compared with related technologies, the prediction accuracy is improved.
  • the lung image and the radiation distribution image (the two images can be three-dimensional images) are cut into the same according to the lung contour. Two sub-images of the size. The two sub-images are histogram equalized, and the two sub-images are normalized and then combined into a two-channel four-dimensional matrix.
  • the four-dimensional matrix is input into the classification network, and the image is processed by the convolutional layer of the classification processing module 11 (specifically, the convolutional layer 121, the regularization layer 122, and the activation layer 123 in each residual module 12).
  • the layer performs feature extraction and dimensionality reduction processing.
  • the fully connected layer 13 obtains the probability of the severity of radiation pneumonia after radiotherapy.
  • the process of training a classification network may include:
  • the three-dimensional image of the lungs and the radiation dose distribution image are cropped into images of the same size (200x240x240) according to the lung contours, and then the image size (100x120x120) is obtained by downsampling to fit the video memory, and the lungs
  • the down-sampled image corresponding to the partial image and the down-sampled image corresponding to the ray distribution image are combined into a four-dimensional matrix (2x100x120x120).
  • Three-dimensional convolutional neural networks such as Res-Net (ResNeXt50 as shown in Figure 4), Dense-Net and other structures can be used to perform convolution, regularization and activation operations on the connected four-dimensional matrix, and change the feature channel from 2 One is increased to 2048, and then a one-dimensional vector is obtained by global average pooling of the features, and the one-dimensional vector is input into the fully connected layer to output two values (probability of serious or not serious), and finally through the softmax function Get the final prediction result (prediction probability).
  • Res-Net Res-Net
  • Dense-Net Dense-Net and other structures can be used to perform convolution, regularization and activation operations on the connected four-dimensional matrix, and change the feature channel from 2 One is increased to 2048, and then a one-dimensional vector is obtained by global average pooling of the features, and the one-dimensional vector is input into the fully connected layer to output two values (probability of serious or not serious), and finally through the softmax function Get the final prediction result (prediction
  • the classification network may adopt a serialized and modularized neural network.
  • Serialization refers to the sequential processing of data input to the neural network according to the serialized modules in the neural network (such as a four-dimensional matrix obtained by cutting two sub-images of the same size and connecting them), and modularity refers to the modules in the neural network can be free
  • modules in the neural network can be free
  • the image obtained by joining the two sub-images cut into the same size is equivalent to a four-dimensional matrix (it can be a four-dimensional matrix with two channels).
  • the feature extraction through the convolution layer can be to use at least one convolution kernel to convolve the input four-dimensional matrix, and the output channel is the four-dimensional matrix with the number of convolution kernels. As the number of convolution kernels increases, the matrix The number of channels has also increased until 2048. Regularization is performed through the regularization layer, and the four-dimensional matrix can be regularized using formula (1):
  • X is a four-dimensional matrix
  • u is the mean of the matrix
  • v is the variance of the matrix
  • the activation function can be used to add non-linear factors to improve the expressive ability of the neural network.
  • Global average pooling is to average the three-dimensional matrix of each channel to obtain a one-dimensional vector whose length is the number of channels. After the global average pooling of the features is performed to obtain a one-dimensional vector, the one-dimensional vector can be calculated through the neural network through the fully connected layer, and finally two values (probability of serious or not serious) are obtained, and finally the multi-classified by softmax The output value (probability of serious or not serious) is transformed into relative probability and used as the final prediction result.
  • the deep learning optimizer (such as the Adam optimizer) calculates the updated difference and adds it to the original parameters to achieve the update of the classification network parameters. This process is continuously iterated until the error is within the preset range (for example, the error is zero), and the classification network reaches Convergence. In this way, the trained target classification network can be obtained.
  • the writing order of the steps does not mean a strict execution order but constitutes any limitation on the implementation process.
  • the specific execution order of each step should be based on its function and possibility.
  • the inner logic is determined.
  • embodiments of the present application also provide image processing apparatuses, electronic equipment, computer-readable storage media, and programs, all of which can be used to implement any image processing method provided in the embodiments of the present application.
  • image processing apparatuses electronic equipment, computer-readable storage media, and programs, all of which can be used to implement any image processing method provided in the embodiments of the present application.
  • the corresponding technical solutions and descriptions and refer to the methods Part of the corresponding records will not be repeated here.
  • FIG. 5 is a schematic structural diagram of an image processing device provided by an embodiment of the application.
  • the image processing device of an embodiment of the application may include: a classification part 31 configured to input the first image and the second image into the target The classification network obtains the classification result; the prediction part 32 is configured to predict the change data according to the classification result to obtain the prediction result; the change data characterization: the target object in the first image is based on the second image The regional distribution in the processing of the changes obtained.
  • the device further includes a training part configured to:
  • the trained classification network is used as the target classification network.
  • the training part includes:
  • the superposition sub-part is configured to perform image superposition processing on the first image and the second image to obtain an image to be processed; the first image and the second image are different types of image data;
  • the training sub-part is configured to input the to-be-processed image as a training sample into the classification network for training to obtain the trained classification network.
  • the training part further includes:
  • the cutting sub-part is configured to perform image cutting on the first image and the second image respectively according to the contour of the target object to obtain the first sub-image data and the second sub-image data after being cut;
  • the first sub-image data and the second sub-image data are used as the image to be processed.
  • the first sub-image data and the second sub-image data are image data of the same size.
  • the training part further includes:
  • the equalization processing sub-part is configured to convert the first sub-image data and the second sub-image data into corresponding histograms respectively, and perform equalization processing of the histograms to obtain an equalization processing result; based on the The equalization processing result trains the classification network to obtain the trained classification network.
  • the training part further includes:
  • the normalization processing sub-part is configured to perform normalization processing on the corresponding pixels contained in the first sub-image data and the second sub-image data to obtain a normalization processing result; based on the normalization
  • the processing result trains the classification network to obtain the trained classification network.
  • the classification network includes at least one classification processing module
  • the training sub-part is configured to perform feature extraction, dimensionality reduction processing, and global average pooling processing on the image to be processed through the at least one classification processing module to obtain a loss function; back propagation training according to the loss function The classification network to obtain the trained classification network.
  • each of the classification processing modules includes at least a convolutional layer
  • the training sub-part is configured to perform feature extraction on the image to be processed through the corresponding convolutional layer in the at least one classification processing module, and then perform dimensionality reduction processing to obtain a first processing result;
  • the result is input to the fully connected layer after the global average pooling process, and the second processing result is obtained, and the second processing result is used to characterize the predicted change of the extracted features; according to the second processing result and the manual labeling result, the result is obtained.
  • the loss function is configured to perform feature extraction on the image to be processed through the corresponding convolutional layer in the at least one classification processing module, and then perform dimensionality reduction processing to obtain a first processing result;
  • the result is input to the fully connected layer after the global average pooling process, and the second processing result is obtained, and the second processing result is used to characterize the predicted change of the extracted features; according to the second processing result and the manual labeling result, the result is obtained.
  • the loss function is configured to perform feature extraction on the image to be processed through the corresponding convolutional layer in the at
  • each residual module when the classification processing module is a residual module, each residual module includes: a convolution layer, a regularization layer, and an activation layer;
  • the training sub-part is configured to perform feature extraction on the image to be processed through the corresponding convolutional layer in at least one residual module to obtain the first extraction result, after processing the regularization layer and the activation layer, to obtain the second extraction result.
  • Extraction result According to the second extraction result and the image to be processed, a third extraction result is obtained.
  • the training sub-part is configured to perform dimensionality reduction processing according to the third extraction result to obtain the first processing result.
  • the functions or modules contained in the apparatus provided in the embodiments of the present application can be used to execute the methods described in the above method embodiments.
  • the functions or modules contained in the apparatus provided in the embodiments of the present application can be used to execute the methods described in the above method embodiments.
  • the embodiment of the present application also proposes a computer-readable storage medium on which computer program instructions are stored, and when the computer program instructions are executed by a processor, any one of the above-mentioned image processing methods is implemented.
  • the computer-readable storage medium may be a non-volatile computer-readable storage medium.
  • An embodiment of the present application also proposes an electronic device, including: a processor; a memory configured to store executable instructions of the processor; wherein the processor is configured to be any of the foregoing image processing methods.
  • the electronic device can be a terminal, a server, or other types of devices.
  • An embodiment of the present application also proposes a computer program, including computer-readable code, and when the computer-readable code runs in an electronic device, a processor in the electronic device executes any one of the above-mentioned image processing methods.
  • FIG. 6 is a schematic structural diagram of an electronic device according to an embodiment of the application.
  • the electronic device 800 can be a mobile phone, a computer, a digital broadcasting terminal, a messaging device, a game console, a tablet device, a medical device, and a fitness device.
  • Terminals such as equipment, personal digital assistants, etc.
  • the electronic device 800 may include one or more of the following components: a first processing component 802, a first memory 804, a first power supply component 806, a multimedia component 808, an audio component 810, a first input/output (Input Output, I/O) interface 812, sensor component 814, and communication component 816.
  • a first processing component 802 a first memory 804, a first power supply component 806, a multimedia component 808, an audio component 810, a first input/output (Input Output, I/O) interface 812, sensor component 814, and communication component 816.
  • the first processing component 802 generally controls the overall operations of the electronic device 800, such as operations associated with display, telephone calls, data communication, camera operations, and recording operations.
  • the first processing component 802 may include one or more processors 820 to execute instructions to complete all or part of the steps of the foregoing method.
  • the first processing component 802 may include one or more modules to facilitate the interaction between the first processing component 802 and other components.
  • the first processing component 802 may include a multimedia module to facilitate the interaction between the multimedia component 808 and the first processing component 802.
  • the first memory 804 is configured to store various types of data to support operations in the electronic device 800. Examples of these data include instructions for any application or method operating on the electronic device 800, contact data, phone book data, messages, pictures, videos, etc.
  • the first memory 804 can be implemented by any type of volatile or non-volatile storage device or a combination thereof, such as static random access memory (Static Random-Access Memory, SRAM), electrically erasable programmable read-only memory (Electrically Erasable Programmable Read Only Memory, EEPROM), Erasable Programmable Read-Only Memory (Electrical Programmable Read Only Memory, EPROM), Programmable Read-Only Memory (Programmable Read-Only Memory, PROM), Read-Only Memory (Read-Only Memory) Only Memory, ROM), magnetic memory, flash memory, magnetic disk or optical disk.
  • SRAM static random access memory
  • SRAM static random access memory
  • EEPROM Electrically erasable programmable read-only memory
  • EEPROM Electrically Erasable Programmable
  • the first power supply component 806 provides power for various components of the electronic device 800.
  • the first power supply component 806 may include a power management system, one or more power supplies, and other components associated with generating, managing, and distributing power for the electronic device 800.
  • the multimedia component 808 includes a screen that provides an output interface between the electronic device 800 and the user.
  • the screen may include a liquid crystal display (LCD) and a touch panel (TP). If the screen includes a touch panel, the screen may be implemented as a touch screen to receive input signals from the user.
  • the touch panel includes one or more touch sensors to sense touch, sliding, and gestures on the touch panel. The touch sensor can not only sense the boundary of a touch or slide action, but also detect the duration and pressure related to the touch or slide operation.
  • the multimedia component 808 includes a front camera and/or a rear camera. When the electronic device 800 is in an operation mode, such as a shooting mode or a video mode, the front camera and/or the rear camera can receive external multimedia data. Each front camera and rear camera can be a fixed optical lens system or have focal length and optical zoom capabilities.
  • the audio component 810 is configured to output and/or input audio signals.
  • the audio component 810 includes a microphone (MIC), and when the electronic device 800 is in an operation mode, such as a call mode, a recording mode, and a voice recognition mode, the microphone is configured to receive an external audio signal.
  • the received audio signal may be further stored in the first memory 804 or transmitted via the communication component 816.
  • the audio component 810 further includes a speaker for outputting audio signals.
  • the first input/output interface 812 provides an interface between the first processing component 802 and a peripheral interface module.
  • the peripheral interface module may be a keyboard, a click wheel, a button, and the like. These buttons may include, but are not limited to: home button, volume button, start button, and lock button.
  • the sensor component 814 includes one or more sensors for providing the electronic device 800 with various aspects of state evaluation.
  • the sensor component 814 can detect the on/off status of the electronic device 800 and the relative positioning of the components.
  • the component is the display and the keypad of the electronic device 800.
  • the sensor component 814 can also detect the electronic device 800 or the electronic device 800.
  • the position of the component changes, the presence or absence of contact between the user and the electronic device 800, the orientation or acceleration/deceleration of the electronic device 800, and the temperature change of the electronic device 800.
  • the sensor component 814 may include a proximity sensor configured to detect the presence of nearby objects when there is no physical contact.
  • the sensor component 814 may also include a light sensor, such as a complementary metal oxide semiconductor (Complementary Metal Oxide Semiconductor, CMOS) or a charge coupled device (Charge Coupled Device, CCD) image sensor for use in imaging applications.
  • CMOS Complementary Metal Oxide Semiconductor
  • CCD Charge Coupled Device
  • the sensor component 814 may also include an acceleration sensor, a gyroscope sensor, a magnetic sensor, a pressure sensor, or a temperature sensor.
  • the communication component 816 is configured to facilitate wired or wireless communication between the electronic device 800 and other devices.
  • the electronic device 800 can access a wireless network based on a communication standard, such as WiFi, 2G, or 3G, or a combination thereof.
  • the communication component 816 receives a broadcast signal or broadcast related information from an external broadcast management system via a broadcast channel.
  • the communication component 816 further includes a Near Field Communication (NFC) module to facilitate short-range communication.
  • the NFC module can be based on Radio Frequency Identification (RFID) technology, Infrared Data Association (Infrared Data Association, IrDA) technology, Ultra Wide Band (UWB) technology, Bluetooth (Bluetooth, BT) technology and other technologies. Technology to achieve.
  • RFID Radio Frequency Identification
  • IrDA Infrared Data Association
  • UWB Ultra Wide Band
  • Bluetooth Bluetooth
  • the electronic device 800 may be used by one or more application specific integrated circuits (ASIC), digital signal processors (Digital Signal Processor, DSP), and digital signal processing equipment (Digital Signal Process, DSPD), programmable logic device (Programmable Logic Device, PLD), Field Programmable Gate Array (Field Programmable Gate Array, FPGA), controller, microcontroller, microprocessor or other electronic components to implement the above methods .
  • ASIC application specific integrated circuits
  • DSP Digital Signal Processor
  • DSPD digital signal processing equipment
  • PLD programmable logic device
  • FPGA Field Programmable Gate Array
  • controller microcontroller, microprocessor or other electronic components to implement the above methods .
  • a non-volatile computer-readable storage medium is also provided, such as the first memory 804 including computer program instructions, which can be executed by the processor 820 of the electronic device 800 to accomplish any of the foregoing.
  • An image processing method An image processing method.
  • FIG. 7 is a schematic structural diagram of another electronic device according to an embodiment of the application.
  • the electronic device 900 may be provided as a server.
  • the electronic device 900 includes a second processing component 922, which further includes one or more processors, and a memory resource represented by the second memory 932, for storing instructions that can be executed by the second processing component 922, For example, applications.
  • the application program stored in the second memory 932 may include one or more modules each corresponding to a set of instructions.
  • the second processing component 922 is configured to execute instructions to perform the above-mentioned method.
  • the electronic device 900 may also include a second power supply component 926 configured to perform power management of the electronic device 900, a wired or wireless network interface 950 configured to connect the electronic device 900 to the network, and a second input and output (I/O ) Interface 958.
  • the electronic device 900 may operate based on an operating system stored in the second storage 932, such as Windows ServerTM, Mac OS XTM, UnixTM, LinuxTM, FreeBSDTM or the like.
  • a non-volatile computer-readable storage medium is also provided, such as a second memory 932 including computer program instructions, which can be executed by the second processing component 922 of the electronic device 900 to complete The above method.
  • the embodiments of this application may be systems, methods and/or computer program products.
  • the computer program product may include a computer-readable storage medium loaded with computer-readable program instructions for enabling a processor to implement various aspects of the present application.
  • the computer-readable storage medium may be a tangible device that can hold and store instructions used by the instruction execution device.
  • the computer-readable storage medium may be, for example, but not limited to, an electrical storage device, a magnetic storage device, an optical storage device, an electromagnetic storage device, a semiconductor storage device, or any suitable combination of the foregoing.
  • Non-exhaustive list of computer-readable storage media include: portable computer disks, hard disks, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM) Or flash memory), static random access memory (SRAM), portable compact disk read-only memory (CD-ROM), digital versatile disk (Digital Video Disc, DVD), memory stick, floppy disk, mechanical encoding device, such as storage on it Commanded punch card or raised structure in the groove, and any suitable combination of the above.
  • RAM random access memory
  • ROM read-only memory
  • EPROM erasable programmable read-only memory
  • flash memory flash memory
  • SRAM static random access memory
  • CD-ROM compact disk read-only memory
  • DVD digital versatile disk
  • memory stick floppy disk
  • mechanical encoding device such as storage on it Commanded punch card or raised structure in the groove, and any suitable combination of the above.
  • the computer-readable storage medium used here is not interpreted as the instantaneous signal itself, such as radio waves or other freely propagating electromagnetic waves, electromagnetic waves propagating through waveguides or other transmission media (for example, light pulses through fiber optic cables), or through wires Transmission of electrical signals.
  • the computer-readable program instructions described herein can be downloaded from a computer-readable storage medium to various computing/processing devices, or downloaded to an external computer or external storage device via a network, such as the Internet, a local area network, a wide area network, and/or a wireless network.
  • the network may include copper transmission cables, optical fiber transmission, wireless transmission, routers, firewalls, switches, gateway computers, and/or edge servers.
  • the network adapter card or network interface in each computing/processing device receives computer-readable program instructions from the network, and forwards the computer-readable program instructions for storage in the computer-readable storage medium in each computing/processing device .
  • the computer program instructions used to perform the operations of the embodiments of the present application may be assembly instructions, instruction set architecture (ISA) instructions, machine instructions, machine-related instructions, microcode, firmware instructions, state setting data, or one or more programming Source code or object code written in any combination of languages, the programming language includes object-oriented programming languages such as Smalltalk, C++, etc., and conventional procedural programming languages such as "C" language or similar programming languages.
  • Computer-readable program instructions can be executed entirely on the user's computer, partly on the user's computer, executed as a stand-alone software package, partly on the user's computer and partly executed on a remote computer, or entirely on the remote computer or server carried out.
  • the remote computer can be connected to the user's computer through any kind of network-including Local Area Network (LAN) or Wide Area Network (WAN)-or it can be connected to an external computer (for example, Use an Internet service provider to connect via the Internet).
  • electronic circuits such as programmable logic circuits, FPGAs, or programmable logic arrays (Programmable Logic Array, PLA), can be customized by using the status information of computer-readable program instructions. Read the program instructions to realize all aspects of the embodiments of the present application.
  • These computer-readable program instructions can be provided to the processor of a general-purpose computer, a special-purpose computer, or other programmable data processing device, thereby producing a machine that makes these instructions when executed by the processor of the computer or other programmable data processing device , A device that implements the functions/actions specified in one or more blocks in the flowcharts and/or block diagrams is produced. It is also possible to store these computer-readable program instructions in a computer-readable storage medium. These instructions make computers, programmable data processing apparatuses, and/or other devices work in a specific manner. Thus, the computer-readable medium storing the instructions includes An article of manufacture, which includes instructions for implementing various aspects of the functions/actions specified in one or more blocks in the flowcharts and/or block diagrams.
  • each block in the flowchart or block diagram may represent a module, program segment, or part of an instruction, and the module, program segment, or part of an instruction contains one or more components for realizing the specified logical function.
  • Executable instructions may also occur in a different order than the order marked in the drawings. For example, two consecutive blocks can actually be executed substantially in parallel, or they can sometimes be executed in the reverse order, depending on the functions involved.
  • each block in the block diagram and/or flowchart, and the combination of the blocks in the block diagram and/or flowchart can be implemented by a dedicated hardware-based system that performs the specified functions or actions Or it can be realized by a combination of dedicated hardware and computer instructions.
  • the embodiment of the application provides an image processing method and device, electronic equipment, computer storage medium, and computer program.
  • the method includes: inputting a first image and a second image into a target classification network to obtain a classification result; As a result, the change data is predicted to obtain the prediction result; the change data characterizes: the change situation of the target object in the first image based on the area distribution in the second image.

Landscapes

  • Engineering & Computer Science (AREA)
  • Data Mining & Analysis (AREA)
  • Theoretical Computer Science (AREA)
  • Health & Medical Sciences (AREA)
  • Physics & Mathematics (AREA)
  • Medical Informatics (AREA)
  • General Physics & Mathematics (AREA)
  • Computer Vision & Pattern Recognition (AREA)
  • General Health & Medical Sciences (AREA)
  • Public Health (AREA)
  • Biomedical Technology (AREA)
  • Evolutionary Computation (AREA)
  • General Engineering & Computer Science (AREA)
  • Artificial Intelligence (AREA)
  • Life Sciences & Earth Sciences (AREA)
  • Evolutionary Biology (AREA)
  • Bioinformatics & Computational Biology (AREA)
  • Bioinformatics & Cheminformatics (AREA)
  • Databases & Information Systems (AREA)
  • Pathology (AREA)
  • Epidemiology (AREA)
  • Primary Health Care (AREA)
  • Biophysics (AREA)
  • Computational Linguistics (AREA)
  • Molecular Biology (AREA)
  • Computing Systems (AREA)
  • Mathematical Physics (AREA)
  • Software Systems (AREA)
  • Nuclear Medicine, Radiotherapy & Molecular Imaging (AREA)
  • Radiology & Medical Imaging (AREA)
  • Quality & Reliability (AREA)
  • Image Analysis (AREA)

Abstract

本申请实施例提供涉及一种图像处理方法及装置、电子设备、计算机存储介质和计算机程序，所述方法包括：将第一图像和第二图像输入目标分类网络，得到分类结果；根据所述分类结果，对变化数据进行预测，得到预测结果；所述变化数据表征：所述第一图像中的目标对象基于所述第二图像中的区域分布进行处理所得到的变化情况。

Description

图像处理方法及装置、电子设备、存储介质和计算机程序
相关申请的交叉引用
本申请基于申请号为201910918450.1、申请日为2019年09月26日的中国专利申请提出，并要求该中国专利申请的优先权，该中国专利申请的全部内容在此引入本申请作为参考。
技术领域
本申请实施例涉及计算机视觉技术领域，涉及但不限于一种图像处理方法及装置、电子设备、计算机存储介质和计算机程序。
背景技术
深度学习快速发展，在图像处理领域取得显著成就。目前基于深度学习的图像处理技术，要得到病灶发展程度的预测，需要经过多个处理过程，导致处理过程繁琐，而且由此得到的预测结果的准确性也不高。
发明内容
本申请实施例提出了一种图像处理方法及装置、电子设备、计算机存储介质和计算机程序。
本申请实施例提供了一种图像处理方法，包括：
将第一图像和第二图像输入目标分类网络，得到分类结果；
根据所述分类结果，对变化数据进行预测，得到预测结果；所述变化数据表征：所述第一图像中的目标对象基于所述第二图像中的区域分布进行处理所得到的变化情况。
本申请实施例的技术方案中，采用一个目标分类网络就可以得到分类结果，从而简化了图像处理过程，而且，该分类结果是通过第一图像和第二图像联合输入该目标分类网络所得到的，并根据分类结果对所述第一图像中的目标对象基于所述第二图像中的区域分布进行处理所得到的变化情况进行预测，从而，提高了预测结果的准确性。
在本申请的一些实施例中，所述方法还包括：
对分类网络进行训练，得到训练后的分类网络；
将所述训练后的分类网络作为所述目标分类网络。
采用本申请实施例的技术方案，将训练后的分类网络作为目标分类网络，则采用目标分类网络可以得到分类结果，根据分类结果对所述第一图像中的目标对象基于所述第二图像中的区域分布进行处理所得到的变化情况进行预测，从而，提高了预测结果的准确性。
在本申请的一些实施例中，所述对分类网络进行训练，得到训练后的分类网络，包括：
将第一图像和第二图像进行图像叠加处理后，得到待处理图像；所述第一图像和所述第二图像为不同种类的图像数据；
将所述待处理图像作为训练样本输入所述分类网络进行训练，得到所述训练后的分类网络。
采用本申请实施例的技术方案，根据第一图像和第二图像进行图像叠加处理后得到的待处理图像进行分类网络的训练，由于综合考虑了这两种图像数据并在该分类网络中进行训练，因此，可以得到训练的精确度。
在本申请的一些实施例中，所述得到待处理图像包括：
根据所述目标对象的轮廓，对所述第一图像及所述第二图像分别进行图像切割，得 到切割后的第一子图像数据和第二子图像数据；
将所述第一子图像数据和所述第二子图像数据作为所述待处理图像。
采用本申请实施例的技术方案，根据所述目标对象的轮廓分别切割第一图像及所述第二图像，得到切割后的第一子图像数据和第二子图像数据并用于该分类网络的训练，可以提高训练效率。
在本申请的一些实施例中，所述第一子图像数据和所述第二子图像数据，为相同尺寸的图像数据。
本申请实施例的技术方案中，采用相同尺寸的图像数据，在图像叠加处理过程中可以无需像素位置对齐，直接采用第一子图像数据和第二子图像数据实现图像叠加即可，从而提高了图像叠加的处理效率。
在本申请的一些实施例中，所述将所述待处理图像作为训练样本输入所述分类网络进行训练，得到所述训练后的分类网络，包括：
将所述第一子图像数据和所述第二子图像数据分别转换为对应的直方图，并进行直方图的均衡化处理，得到均衡化处理结果；
基于所述均衡化处理结果对所述分类网络进行训练，得到所述训练后的分类网络。
采用本申请实施例的技术方案，将图像数据转换为对应的直方图，并进行直方图的均衡化处理，可以将图像的直方图分布变成近似均匀分布，从而增强了图像的对比度，使图像更为清晰。
在本申请的一些实施例中，所述将所述待处理图像作为训练样本输入所述分类网络进行训练，得到所述训练后的分类网络，包括：
将所述第一子图像数据和所述第二子图像数据中包含的对应像素点进行归一化处理，得到归一化处理结果；
基于所述归一化处理结果对所述分类网络进行训练，得到所述训练后的分类网络。
采用本申请实施例的技术方案通过对像素点进行归一化处理，可以归纳出像素点的分布性，从而可以把需要处理的像素点经过归一化处理后限制在预设的定范围内，也就是说，归一化处理是为了后续的一系列处理更加方便快捷，有利于加速分类网络训练的收敛速度。
在本申请的一些实施例中，所述分类网络包括至少一个分类处理模块；
所述将所述待处理图像作为训练样本输入所述分类网络进行训练，得到所述训练后的分类网络，包括：
将所述待处理图像通过所述至少一个分类处理模块进行特征提取、降维处理及全局平均池化处理，得到损失函数；
根据所述损失函数的反向传播训练所述分类网络，以得到所述训练后的分类网络。
采用本申请实施例的技术方案，对待处理图像通过至少一个分类处理模块进行特征提取、降维处理及全局平均池化处理，可以得到损失函数，根据所述损失函数的反向传播训练所述分类网络，从而，训练得到该目标分类网络。
在本申请的一些实施例中，每个分类处理模块至少包括卷积层；
所述将所述待处理图像通过所述至少一个分类处理模块进行特征提取、降维处理及全局平均池化处理，得到损失函数，包括：
将所述待处理图像通过所述至少一个分类处理模块中的对应卷积层进行特征提取后，进行降维处理，得到第一处理结果；
将所述第一处理结果进行全局平均池化处理后输入全连接层，得到第二处理结果，所述第二处理结果用于表征所提取特征的预测变化情况；
根据所述第二处理结果和手工标注结果，得到所述损失函数。
采用本申请实施例的技术方案，可以根据降维处理后得到的第一处理结果得到第二 处理结果，根据所述第二处理结果和手工标注结果，得到所述损失函数，以便根据所述损失函数的反向传播训练所述分类网络，从而，训练得到该目标分类网络。
在本申请的一些实施例中，所述分类处理模块为残差模块的情况下，每个残差模块包括：卷积层、正则化层和激活层；
所述将所述待处理图像通过所述至少一个分类处理模块中的对应卷积层进行特征提取后，还包括：
将所述待处理图像通过至少一个残差模块中的对应卷积层进行特征提取后得到的第一提取结果，经过正则化层和激活层的处理后得到第二提取结果；
根据所述第二提取结果和所述待处理图像，得到第三提取结果。
采用本申请实施例的技术方案，分类处理模块为残差模块的情况下，该模块的结构包括：卷积层、正则化层和激活层，通过残差模块对应卷积层进行特征提取后得到的第一提取结果，经过正则化层和激活层的处理后得到第二提取结果，根据所述第二提取结果和所述待处理图像，可以得到第三提取结果，以便根据该第三提取结果得到用于计算损失函数的第一处理结果，得到损失函数后，可以根据所述损失函数的反向传播训练所述分类网络，从而，训练得到该目标分类网络。
在本申请的一些实施例中，所述进行降维处理，得到第一处理结果，包括：
根据所述第三提取结果，进行降维处理，得到所述第一处理结果。
采用本申请实施例的技术方案，可以对第三提取结果进行降维处理，以得到用于计算损失函数的第一处理结果，得到损失函数后，可以根据所述损失函数的反向传播训练所述分类网络，从而，训练得到该目标分类网络。
本申请实施例还提供了一种图像处理装置，所述装置包括：
分类部分，配置为将第一图像和第二图像输入目标分类网络，得到分类结果；
预测部分，配置为根据所述分类结果，对变化数据进行预测，得到预测结果；所述变化数据表征：所述第一图像中的目标对象基于所述第二图像中的区域分布进行处理所得到的变化情况。
在本申请的一些实施例中，所述装置还包括训练部分，配置为：
对分类网络进行训练，得到训练后的分类网络；
将所述训练后的分类网络作为所述目标分类网络。
在本申请的一些实施例中，所述训练部分，包括：
叠加子部分，配置为将第一图像和第二图像进行图像叠加处理后，得到待处理图像；所述第一图像和所述第二图像为不同种类的图像数据；
训练子部分，配置为将所述待处理图像作为训练样本输入所述分类网络进行训练，得到所述训练后的分类网络。
在本申请的一些实施例中，所述训练部分，还包括：
切割子部分，配置为根据所述目标对象的轮廓，对所述第一图像及所述第二图像分别进行图像切割，得到切割后的第一子图像数据和第二子图像数据；将所述第一子图像数据和所述第二子图像数据作为所述待处理图像。
在本申请的一些实施例中，所述第一子图像数据和所述第二子图像数据，为相同尺寸的图像数据。
在本申请的一些实施例中，所述训练部分，还包括：
均衡化处理子部分，配置为将所述第一子图像数据和所述第二子图像数据分别转换为对应的直方图，并进行直方图的均衡化处理，得到均衡化处理结果；基于所述均衡化处理结果对所述分类网络进行训练，得到所述训练后的分类网络。
在本申请的一些实施例中，所述训练部分，还包括：
归一化处理子部分，配置为将所述第一子图像数据和所述第二子图像数据中包含的 对应像素点进行归一化处理，得到归一化处理结果；基于所述归一化处理结果对所述分类网络进行训练，得到所述训练后的分类网络。
在本申请的一些实施例中，所述分类网络包括至少一个分类处理模块；
所述训练子部分，配置为：
将所述待处理图像通过所述至少一个分类处理模块进行特征提取、降维处理及全局平均池化处理，得到损失函数；根据所述损失函数的反向传播训练所述分类网络，以得到所述训练后的分类网络。
在本申请的一些实施例中，每个分类处理模块至少包括卷积层；
所述训练子部分，配置为：
将所述待处理图像通过所述至少一个分类处理模块中的对应卷积层进行特征提取后，进行降维处理，得到第一处理结果；将所述第一处理结果进行全局平均池化处理后输入全连接层，得到第二处理结果，所述第二处理结果用于表征所提取特征的预测变化情况；根据所述第二处理结果和手工标注结果，得到所述损失函数。
在本申请的一些实施例中，在所述分类处理模块为残差模块的情况下，每个残差模块包括：卷积层、正则化层和激活层；
所述训练子部分，配置为：
将所述待处理图像通过至少一个残差模块中的对应卷积层进行特征提取后得到的第一提取结果，经过正则化层和激活层的处理，得到第二提取结果；根据所述第二提取结果和所述待处理图像，得到第三提取结果。
在本申请的一些实施例中，所述训练子部分，配置为根据所述第三提取结果，进行降维处理，得到所述第一处理结果。
本申请实施例还提供了一种电子设备，包括：
处理器；
配置为存储处理器可执行指令的存储器；
其中，所述处理器被配置为：执行上述任意一种图像处理方法。
本申请实施例还提供了一种计算机可读存储介质，其上存储有计算机程序指令，所述计算机程序指令被处理器执行时实现上述任意一种图像处理方法。
本申请实施例还提供了一种计算机程序，包括计算机可读代码，当所述计算机可读代码在电子设备中运行时，所述电子设备中的处理器执行用于实现上述任意一种图像处理方法。
在本申请实施例中，将第一图像和第二图像输入目标分类网络，得到分类结果；根据所述分类结果，对所述第一图像中的目标对象基于所述第二图像中的区域分布进行处理所得到的变化情况进行预测，得到预测结果。由于采用一个目标分类网络就可以得到分类结果，因此，采用本申请实施例的技术方案，简化了图像处理过程，由于该分类结果是通过第一图像和第二图像联合输入该目标分类网络所得到的，并根据分类结果对所述第一图像中的目标对象基于所述第二图像中的区域分布进行处理所得到的变化情况进行预测，因此，采用本申请实施例的技术方案，提高了预测结果的准确性。
应当理解的是，以上的一般描述和后文的细节描述仅是示例性和解释性的，而非限制本申请。
根据下面参考附图对示例性实施例的详细说明，本申请的其它特征及方面将变得清楚。
附图说明
此处的附图被并入说明书中并构成本说明书的一部分，这些附图示出了符合本申请的实施例，并与说明书一起用于说明本申请实施例的技术方案。
图1为本申请实施例提供的图像处理方法的流程示意图；
图2为本申请实施例的一个应用场景的示意图；
图3为本申请实施例提供的分类网络的训练流程示意图；
图4为本申请实施例提供的实现图像处理方法的分类网络架构示意图；
图5为本申请实施例提供的图像处理装置的结构示意图；
图6为本申请实施例的一个电子设备的结构示意图；
图7为本申请实施例的另一个电子设备的结构示意图。
具体实施方式
以下将参考附图详细说明本申请的各种示例性实施例、特征和方面。附图中相同的附图标记表示功能相同或相似的元件。尽管在附图中示出了实施例的各种方面，但是除非特别指出，不必按比例绘制附图。
在这里专用的词“示例性”意为“用作例子、实施例或说明性”。这里作为“示例性”所说明的任何实施例不必解释为优于或好于其它实施例。
本文中术语“和/或”，仅仅是一种描述关联对象的关联关系，表示可以存在三种关系，例如，A和/或B，可以表示：单独存在A，同时存在A和B，单独存在B这三种情况。另外，本文中术语“至少一种”表示多种中的任意一种或多种中的至少两种的任意组合，例如，包括A、B、C中的至少一种，可以表示包括从A、B和C构成的集合中选择的任意一个或多个元素。
另外，为了更好的说明本申请实施例，在下文的具体实施方式中给出了众多的具体细节。本领域技术人员应当理解，没有某些具体细节，本申请实施例同样可以实施。在一些实例中，对于本领域技术人员熟知的方法、手段、元件和电路未作详细描述，以便于凸显本申请实施例的主旨。
图像处理的一个应用方向是：通过优化医学图像的处理，以更简化的处理过程，得到更准确的预测结果。该预测结果可以为针对指定区域病灶进行处理时会对该指定区域造成影响程度的预测结果。比如，病灶可以在腹部、肺部、肾部、脑部、心脏等，若指定区域是肺部，则需要对肺部存在病灶的位置经手术处置后对肺部造成影响程度(如严重程度或不严重程度等等)进行预测。
比如，对肺部病灶进行射线放疗之前，需要预测射线放疗对肺部造成影响程度。对肺部进行射线放疗可能会导致放射性肺炎，放射性肺炎是由于病灶(如肺癌、乳腺癌、食管癌、恶性淋巴瘤或胸部其他恶性肿瘤)经放射治疗后，在放射野内的正常肺组织受到损伤而引起的炎症反应。轻者无症状，炎症可自行消散；重者肺脏发生广泛纤维化，导致呼吸功能损害，甚致呼吸衰竭。炎症反应的程度与放射剂量以及放疗前病灶的状态密切相关，需要对诸如肺癌放疗后放射性肺炎严重程度进行预测，该过程比较繁琐，需要先从图像如包含肺癌病灶的电子计算机断层扫描(Computed Tomography，CT)图像中提取图像特征，然后从提取的图像特征中选择待处理的图像特征，之后在分类器中进行分类，以根据分类结果进行图像中对指定区域影响程度的预测。
在相关技术中，对于图像特征的提取，可以通过放射组学来实现。通过放射组学提取图像特征，是通过放射影像方法提取图像特征，再研究该图像特征与临床症状(如指定区域影响程度预测)等关系。提取图像特征后对特征进行选择，可以通过支持向量机(Support Vector Machines，SVM)等分类器预测对指定区域的影响程度(如严重程度或不严重程度等等)。由此可见，整个图像处理过程包括多个阶段，不仅繁琐，而且由此得到的预测结果的准确性也不高。准确性不高是因为：
1)在每个阶段都有许多人为设定的超参数，人为设定的超参数的选取是否准确，对最终预测结果有很大影响，也就是说，如果人为设定的超参数选取不准确，最终的预测 结果也不准确；
2)射线放疗采用的放射剂量的预测，与图像处理过程的预测分别进行，这里，可以通过整个肺内部放射剂量求均值等方法得到剂量常数，以实现上述预测过程，比如，一般用戈瑞(Gray，Gy)单位来衡量放射的吸收剂量，医生可以通过统计肺内吸收剂量超过一定值的组织占整个肺部的百分比作为该剂量常数。如V20就是肺内吸收剂量超过20Gy的组织体积占整个肺部体积的百分比。采用该剂量常数的处理方法过于笼统，没有考虑不同病灶处剂量大小也不同，显然病灶在腹部、肺部、肾部、脑部、心脏等不同区域，采用剂量大小是不同的，相应的，放射后造成的影响也不同。比如，虽然整个肺部的内部射线剂量比较少，通过统计的常数也比较小，但是，当射线照射到关键的器官，如主要的气管、血管、心脏等，也会导致严重后果。也就是说，该剂量常数的处理方法所采用的常数，只是一个统计量，没考虑到射线在不同区域空间上的分布，如此一来，采用该剂量常数的处理方法得到的预测准确率也并不高。
综上所述，相关技术中，放疗后肺炎严重程度预测任务主要通过放射组学的方式来解决，存在着效率不高、鲁棒性不强、未考虑射线分布和准确率低等缺陷。虽然有的放射组学方法给出了理想的准确率，但是其过程中的特征选择优化和剂量常数、SVM的超参数选择使其方法鲁棒性不强，难以在其他数据集上泛用。同时，可以对射线剂量进行了常数化处理，即将整个肺内的或癌症区域的射线剂量统计为一个常数，但是这样计算就失去了射线的分布特征。
采用本申请实施例，以通过射线放疗后对肺炎严重程度的预测为例，通过深度学习训练后得到的目标分类网络(如分类神经网络，该网络可以是三维的)，可以将肺部图像与射线分布图像(这两个图像都可以是三维的图像)同时输入到该目标分类网络中，从而，通过该目标分类网络可以综合得到病灶所在指定区域或相关联区域在每个位置上的图像及射线分布，以提高预测准确性，并通过该目标分类网络的分类，可以直接一步输出放疗后将发生肺炎的严重程度。本申请实施例不仅在图像处理的预测过程中考虑了射线剂量的分布，而且还可以将泛用在相似任务数据集上，如对于任一医院的放射性肺炎数据，都可以直接应用本申请实施例预测放射性肺炎严重的程度，而不需要更改任何参数和结构，且应用场景不限于病灶在腹部、肺部、肾部、脑部、心脏等不同区域或相关联区域的预测，并可以快速的得到准确的预测结果。
图1为本申请实施例提供的图像处理方法的流程示意图，该图像处理方法应用于图像处理装置，例如，图像处理装置可以由终端设备或服务器或其它处理设备执行，其中，终端设备可以为用户设备(UE，User Equipment)、移动设备、蜂窝电话、无绳电话、个人数字处理(PDA，Personal Digital Assistant)、手持设备、计算设备、车载设备、可穿戴设备等。在本申请的一些实施例中，该图像处理方法可以通过处理器调用存储器中存储的计算机可读指令的方式来实现。如图1所示，该流程包括：
步骤S101、将第一图像和第二图像输入目标分类网络，得到分类结果。
在本申请的一些实施例中，第一图像可以为存在病灶的图像(可以是病灶所在区域的CT图)，如病灶在腹部、肺部、肾部、脑部、心脏等不同区域或相关联区域的图像。第二图像可以为对病灶所在区域或相关联区域进行放疗所使用的射线剂量分布图。在实际实施时，可以将存在病灶的图像和射线剂量分布图这两个图像联合输入目标分类网络，以得到分类结果。
步骤S102、根据所述分类结果，对变化数据进行预测，得到预测结果。
这里，变化数据表征：所述第一图像中的目标对象基于所述第二图像中的区域分布进行处理所得到的变化情况。
在本申请的一些实施例中，该目标对象可以为病灶所在器官，如腹部、肺部、肾部、脑部、心脏等。该区域分布可以为在不同区域，针对不同病灶所采用的射线剂量的分布。 该变化情况可以为一旦对病灶进行射线放疗后可能对病灶所在器官(如肺部)产生炎症的严重程度(如严重的概率或者不严重的概率)。比如，可以根据分类结果，对存在病灶的图像中肺部基于射线剂量分布图中射线分布所得到的炎症严重程度进行预测，以得到预测结果。
采用本申请实施例的技术方案，只需要根据目标分类网络进行分类就可以实现预测，则采用一步即可端到端的得到预测结果，不需要多阶段的繁琐操作。综合考虑了存在病灶的图像和射线剂量分布图的彼此影响，并使用射线剂量分布图作为联合输入，而不是将二者分别割裂的处理，由于通过上述图像间的彼此影响，从而充分考虑了不同位置放射剂量不同带来的不同影响，提高了预测准确率。而且，该目标分类网络不需要人为的超参数来调控，而是可以采用深度学习训练后得到的目标分类网络，实现了整个图像处理过程中预测的自适应调控，有助于提高预测准确率。
本申请实施例的图像处理方法，能够应用于肺癌术前放疗规划、放射性肺炎预测等场景中；图2为本申请实施例的一个应用场景的示意图，如图2所示，肺癌病人的CT图像201为上述第一图像，放射剂量分布图202为上述第二图像，医师在得到肺癌病人的CT图像后，需要对放疗手术进行规划；此时，可以将放射剂量分布图和肺癌病人的CT图像输入至上述图像处理装置200中，在所述图像处理装置中，通过前述实施例记载的图像处理方法进行处理，可以得到放疗后会发生放射性肺炎的严重程度的预测结果，进而可以对放疗后将会发生放射性肺炎的严重程度进行预测，因此，可以帮助医生提前预知术后风险，进行提前防范或对放疗规划进行修正。
在本申请的一些实施例中，可以对分类网络进行训练，得到训练后的分类网络；将所述训练后的分类网络作为所述目标分类网络。
图3为本申请实施例提供的分类网络的训练流程示意图，该分类网络的训练流程可以基于图像处理装置实现，例如，图像处理装置可以由终端设备或服务器或其它处理设备执行，其中，终端设备可以为UE、移动设备、蜂窝电话、无绳电话、PDA、手持设备、计算设备、车载设备、可穿戴设备等。在本申请的一些实施例中，该图像处理方法可以通过处理器调用存储器中存储的计算机可读指令的方式来实现。如图3所示，该分类网络的训练流程可以包括：
步骤S301、将第一图像和第二图像进行图像叠加处理后，得到待处理图像。
本申请实施例中，第一图像和第二图像可以为不同种类的图像数据；
在本申请的一些实施例中，可以在将第一图像和第二图像进行图像叠加处理后，根据所述目标对象的轮廓，对所述第一图像及所述第二图像分别进行图像切割，得到切割后的第一子图像数据和第二子图像数据。
需要指出的是，目标对象的轮廓并不是病灶的轮廓，而是病灶所在区域或相关联区域的轮廓，如病灶所在肺部的整个肺部轮廓，或者，病灶所在心脏，肾脏等轮廓，根据不同区域，采用不同的剂量。
在本申请的一些实施例中，第一子图像数据和第二子图像数据，可以为相同尺寸的图像数据。比如，按照肺部轮廓，切割三维的第一图像及所述第二图像，且二者为相同大小。然后，可以将第一子图像数据和第二子图像数据作为所述待处理图像。
在本申请的一些实施例中，所述第一子图像数据和所述第二子图像数据，也可以为不相同尺寸的图像数据，这种情况下，可以将所述第一子图像数据和所述第二子图像数据中包含的对应像素点进行像素位置对齐处理，得到对齐后的第一子图像数据和对齐后的第二子图像数据，将对齐后的第一子图像数据和对齐后的第二子图像数据中包含的对应像素点进行叠加来实现图像叠加处理后，得到所述待处理图像。
步骤S302、将所述待处理图像作为训练样本输入所述分类网络进行训练，得到所述训练后的分类网络。
本申请实施例通过步骤S301-步骤S302，对分类网络进行训练，可以得到训练后的分类网络。将所述训练后的分类网络作为所述目标分类网络。
在本申请的一些实施例中，将所述待处理图像作为训练样本输入所述分类网络进行训练，得到所述训练后的分类网络，可以包括：将第一子图像数据和第二子图像数据分别转换为对应的直方图，并进行直方图的均衡化处理，得到均衡化处理结果；基于所述均衡化处理结果对所述分类网络进行训练，得到所述训练后的分类网络。
在本申请的一些实施例中，将所述待处理图像作为训练样本输入所述分类网络进行训练，得到所述训练后的分类网络，可以包括：将第一子图像数据和第二子图像数据中包含的对应像素点进行归一化处理，得到归一化处理结果；基于所述归一化处理结果对所述分类网络进行训练，得到所述训练后的分类网络。
在本申请的一些实施例中，首先可以将第一子图像数据和第二子图像数据分别转换为对应的直方图，并进行直方图的均衡化处理之后得到均衡化处理结果。然后，可以将该均衡化处理结果中对应第一子图像数据和第二子图像数据中包含的对应像素点进行归一化处理。比如，将这俩子图像进行直方图均衡化后，将二者进行归一化后联结为双通道四维矩阵表示的图像数据。将该图像数据输入到分类网络中，先通过该分类网络的卷积层对图像数据逐层进行特征提取并降维处理，最终，通过全连接层的处理得到射线放疗后发生放射性炎症严重程度的概率。
在本申请的一些实施例中，分类网络可以包括至少一个分类处理模块；
相应地，将所述待处理图像作为训练样本输入所述分类网络进行训练，得到所述目标分类网络，可以包括：将所述待处理图像通过所述至少一个分类处理模块进行特征提取(如通过卷积层进行特征提取)、降维处理(如池化处理)及全局平均池化处理，得到损失函数；根据所述损失函数的反向传播训练所述分类网络(例如可以是根据所述损失函数计算出的误差进行反向传播)，以得到所述训练后的分类网络。
在本申请的一些实施例中，每个分类处理模块至少包括卷积层；将所述待处理图像通过所述至少一个分类处理模块进行特征提取、降维处理及全局平均池化处理，得到损失函数，可以包括：将待处理图像通过至少一个分类处理模块中的对应卷积层进行逐层的特征提取后，进行逐层的降维处理，得到第一处理结果；将第一处理结果进行全局平均池化处理后输入全连接层，得到第二处理结果，第二处理结果为分类网络输出的预测结果，用于表征所提取特征的预测变化情况；根据第二处理结果和手工标注结果(如医生已标注的真实变化情况)，得到所述损失函数。
也就是说，可以根据分类网络输出的预测结果与医生已标注的真实变化情况得到该损失函数。如果损失函数所反映出的预测变化情况与真实情况之间的误差在预设范围内(例如误差为零)，说明生成的预测变化情况与真实情况之间的差异达到收敛条件，从而对分类网络的训练结束，得到训练后目标分类网络。
在本申请的一些实施例中，在所述分类处理模块为残差模块的情况下，每个残差模块可以包括：卷积层、正则化层和激活层。
相应地，将待处理图像通过至少一个分类处理模块中的对应卷积层进行逐层的特征提取后，还可以包括：将待处理图像通过至少一个残差模块中的对应卷积层进行特征提取后得到的第一提取结果，经过正则化层和激活层的处理后得到第二提取结果；根据所述第二提取结果和待处理图像，得到第三提取结果，第三提取结果用于降维处理。也就是说，残差模块的输入是“待处理图像”，将残差模块的输入和残差模块中最后一个激活层的输出相加所得到最终的提取结果，即为该第三提取结果。通过残差模块进行特征提取之后，还可以进行逐层的降维处理，以得到第一处理结果。比如，根据该第三提取结果进行逐层的降维处理，得到该第一处理结果。
图4为本申请实施例提供的实现图像处理方法的分类网络架构示意图，如图4所示， 该分类网络(如分类神经网络)可以包括至少一个分类处理模块11。分类处理模块11可以为残差模块12，还可以包括全连接层13。每个残差模块12可以包括：至少一个卷积层121、至少一个正则化层122和至少一个激活层123；通过该分类网络自动学习所提取图像中有用的特征，并应用这些特征进行预测，而不是提取特征后再选择特征，从而相比相关技术，提高了预测准确性。
以射线放疗后放射性肺炎严重程度预测为例进行阐述，训练分类网络的过程中，先将肺部图像与射线分布图像(这两个图像都可以是三维的图像)按照肺部轮廓分别切割为相同大小的两个子图像。将两个子图像进行直方图均衡化，并将二者进行归一化后联结为双通道的四维矩阵。将该四维矩阵输入到分类网络中，通过分类处理模块11的卷积层(具体可以为每个残差模块12中的卷积层121、正则化层122和激活层123的处理)对图像逐层进行特征提取，并进行降维处理，最终通过全连接层13得到射线放疗后发生放射性肺炎严重程度的概率。
在一个示例中，训练分类网络的流程可以包括：
一、参照图4，根据肺部轮廓将肺部三维图像与射线剂量分布图像裁剪为相同大小(200x240x240)的图像，然后，通过降采样得到大小为(100x120x120)的图像以适应显存，并将肺部图像对应的降采样后图像与射线分布图像对应的降采样后图像联结为四维矩阵(2x100x120x120)。
二、可以采用三维卷积神经网络如Res-Net(如图4所示的ResNeXt50)、Dense-Net等结构，对联结后的四维矩阵进行卷积、正则化及激活操作，将特征通道从2个提升到2048个，再通过对特征进行全局平均池化得到一维向量，将该一维向量输入到全连接层中，以输出两个值(严重或不严重的概率)，最后通过softmax函数得到最终的预测结果(预测概率)。
本申请实施例中，分类网络可以采用序列化、模块化的神经网络。
序列化指可以根据神经网络中序列化的模块来顺序处理输入神经网络中的数据(如切割为相同大小的两个子图像进行联结得到的四维矩阵)，而模块化指神经网络中的模块可以随意替换其他能实现本申请实施例的模块，模块替换后也能实现本申请都在本申请的保护范围之内。
需要指出的是，上述切割为相同大小的两个子图像进行联结后得到的图像，相当于一个四维矩阵(可以为2通道的四维矩阵)。通过卷积层进行特征提取，可以是采用至少一个卷积核对输入的四维矩阵进行卷积处理，输出通道数为卷积核个数的四维矩阵，随着卷积核个数的增加，矩阵的通道数也增加，直到2048个。通过正则化层进行正则化处理，可以将该四维矩阵采用公式(1)进行正则化处理：
X＝(X-u)/v     (1)
其中，X为四维矩阵，u为矩阵均值，v为矩阵方差。
对于通过激活层进行激活操作的实现方式，可以使用激活函数来加入非线性因素，以提高神经网络的表达能力。
全局平均池化，即为将每个通道的三维矩阵求均值，得到一个长为通道数的一维向量。在对特征进行全局平均池化得到一维向量后，可以通过全连接层将该一维向量通过神经网络计算，最终得到两个值(严重或不严重的概率)，最后通过softmax将多分类的输出数值(严重或不严重的概率)转化为相对概率并作为最终的预测结果。
三、将预测概率与真实概率通过带权重的交叉熵损失函数计算生成的预测变化情况与真实情况之间的误差，并反向求导后得到该分类网络中每个参数的梯度，并可以通过深度学习优化器(如Adam优化器)计算更新的差值加在原参数上，实现该分类网络参数的更新，不断迭代这个过程直至误差在预设范围内(例如误差为零)，该分类网络达到收敛，如此，可以得到训练后的目标分类网络。
本领域技术人员可以理解，在具体实施方式的上述方法中，各步骤的撰写顺序并不意味着严格的执行顺序而对实施过程构成任何限定，各步骤的具体执行顺序应当以其功能和可能的内在逻辑确定。
本申请提及的上述各个方法实施例，在不违背原理逻辑的情况下，均可以彼此相互结合形成结合后的实施例，限于篇幅，本申请实施例不再赘述。
此外，本申请实施例还提供了图像处理装置、电子设备、计算机可读存储介质、程序，上述均可用来实现本申请实施例提供的任一种图像处理方法，相应技术方案和描述和参见方法部分的相应记载，不再赘述。
图5为本申请实施例提供的图像处理装置的结构示意图，如图5所示，本申请实施例的图像处理装置，可以包括：分类部分31，配置为将第一图像和第二图像输入目标分类网络，得到分类结果；预测部分32，配置为根据所述分类结果，对变化数据进行预测，得到预测结果；所述变化数据表征：所述第一图像中的目标对象基于所述第二图像中的区域分布进行处理所得到的变化情况。
在本申请的一些实施例中，所述装置还包括训练部分，配置为：
对分类网络进行训练，得到训练后的分类网络；
将所述训练后的分类网络作为所述目标分类网络。
在本申请的一些实施例中，所述训练部分，包括：
叠加子部分，配置为将第一图像和第二图像进行图像叠加处理后，得到待处理图像；所述第一图像和所述第二图像为不同种类的图像数据；
训练子部分，配置为将所述待处理图像作为训练样本输入所述分类网络进行训练，得到所述训练后的分类网络。
在本申请的一些实施例中，所述训练部分，还包括：
切割子部分，配置为根据所述目标对象的轮廓，对所述第一图像及所述第二图像分别进行图像切割，得到切割后的第一子图像数据和第二子图像数据；将所述第一子图像数据和所述第二子图像数据作为所述待处理图像。
在本申请的一些实施例中，所述第一子图像数据和所述第二子图像数据，为相同尺寸的图像数据。
在本申请的一些实施例中，所述训练部分，还包括：
均衡化处理子部分，配置为将所述第一子图像数据和所述第二子图像数据分别转换为对应的直方图，并进行直方图的均衡化处理，得到均衡化处理结果；基于所述均衡化处理结果对所述分类网络进行训练，得到所述训练后的分类网络。
在本申请的一些实施例中，所述训练部分，还包括：
归一化处理子部分，配置为将所述第一子图像数据和所述第二子图像数据中包含的对应像素点进行归一化处理，得到归一化处理结果；基于所述归一化处理结果对所述分类网络进行训练，得到所述训练后的分类网络。
在本申请的一些实施例中，所述分类网络包括至少一个分类处理模块；
所述训练子部分，配置为将所述待处理图像通过所述至少一个分类处理模块进行特征提取、降维处理及全局平均池化处理，得到损失函数；根据所述损失函数的反向传播训练所述分类网络，以得到所述训练后的分类网络。
在本申请的一些实施例中，每个所述分类处理模块至少包括卷积层；
所述训练子部分，配置为将所述待处理图像通过所述至少一个分类处理模块中的对应卷积层进行特征提取后，进行降维处理，得到第一处理结果；将所述第一处理结果进行全局平均池化处理后输入全连接层，得到第二处理结果，所述第二处理结果用于表征所提取特征的预测变化情况；根据所述第二处理结果和手工标注结果，得到所述损失函数。
在本申请的一些实施例中，在所述分类处理模块为残差模块的情况下，每个残差模块包括：卷积层、正则化层和激活层；
所述训练子部分，配置为将所述待处理图像通过至少一个残差模块中的对应卷积层进行特征提取后得到的第一提取结果，经过正则化层和激活层的处理，得到第二提取结果；根据所述第二提取结果和所述待处理图像，得到第三提取结果。
在本申请的一些实施例中，所述训练子部分，配置为根据所述第三提取结果，进行降维处理，得到所述第一处理结果。
在一些实施例中，本申请实施例提供的装置具有的功能或包含的模块可以用于执行上文方法实施例描述的方法，其具体实现可以参照上文方法实施例的描述，为了简洁，这里不再赘述。
本申请实施例还提出一种计算机可读存储介质，其上存储有计算机程序指令，所述计算机程序指令被处理器执行时实现上述任意一种图像处理方法。计算机可读存储介质可以是非易失性计算机可读存储介质。
本申请实施例还提出一种电子设备，包括：处理器；配置为存储处理器可执行指令的存储器；其中，所述处理器被配置为上述任意一种图像处理方法。
电子设备可以为终端、服务器或其它形态的设备。
本申请实施例还提出一种计算机程序，包括计算机可读代码，当所述计算机可读代码在电子设备中运行时，所述电子设备中的处理器执行上述任意一种图像处理方法。
图6为本申请实施例的一个电子设备的结构示意图，如图6所示，电子设备800可以是移动电话、计算机、数字广播终端、消息收发设备、游戏控制台、平板设备、医疗设备、健身设备、个人数字助理等终端。
参照图6，电子设备800可以包括以下一个或多个组件：第一处理组件802，第一存储器804，第一电源组件806，多媒体组件808，音频组件810，第一输入/输出(Input Output，I/O)的接口812，传感器组件814，以及通信组件816。
第一处理组件802通常控制电子设备800的整体操作，诸如与显示，电话呼叫，数据通信，相机操作和记录操作相关联的操作。第一处理组件802可以包括一个或多个处理器820来执行指令，以完成上述的方法的全部或部分步骤。此外，第一处理组件802可以包括一个或多个模块，便于第一处理组件802和其他组件之间的交互。例如，第一处理组件802可以包括多媒体模块，以方便多媒体组件808和第一处理组件802之间的交互。
第一存储器804被配置为存储各种类型的数据以支持在电子设备800的操作。这些数据的示例包括用于在电子设备800上操作的任何应用程序或方法的指令，联系人数据，电话簿数据，消息，图片，视频等。第一存储器804可以由任何类型的易失性或非易失性存储设备或者它们的组合实现，如静态随机存取存储器(Static Random-Access Memory，SRAM)，电可擦除可编程只读存储器(Electrically Erasable Programmable Read Only Memory，EEPROM)，可擦除可编程只读存储器(Electrical Programmable Read Only Memory，EPROM)，可编程只读存储器(Programmable Read-Only Memory，PROM)，只读存储器(Read-Only Memory，ROM)，磁存储器，快闪存储器，磁盘或光盘。
第一电源组件806为电子设备800的各种组件提供电力。第一电源组件806可以包括电源管理系统，一个或多个电源，及其他与为电子设备800生成、管理和分配电力相关联的组件。
多媒体组件808包括在所述电子设备800和用户之间的提供一个输出接口的屏幕。在一些实施例中，屏幕可以包括液晶显示器(Liquid Crystal Display，LCD)和触摸面板(Touch Pad，TP)。如果屏幕包括触摸面板，屏幕可以被实现为触摸屏，以接收来自用户的输入信号。触摸面板包括一个或多个触摸传感器以感测触摸、滑动和触摸面板上的手势。所述触摸传感器可以不仅感测触摸或滑动动作的边界，而且还检测与所述触摸或滑动操作 相关的持续时间和压力。在一些实施例中，多媒体组件808包括一个前置摄像头和/或后置摄像头。当电子设备800处于操作模式，如拍摄模式或视频模式时，前置摄像头和/或后置摄像头可以接收外部的多媒体数据。每个前置摄像头和后置摄像头可以是一个固定的光学透镜系统或具有焦距和光学变焦能力。
音频组件810被配置为输出和/或输入音频信号。例如，音频组件810包括一个麦克风(MIC)，当电子设备800处于操作模式，如呼叫模式、记录模式和语音识别模式时，麦克风被配置为接收外部音频信号。所接收的音频信号可以被进一步存储在第一存储器804或经由通信组件816发送。在一些实施例中，音频组件810还包括一个扬声器，用于输出音频信号。
第一输入/输出接口812为第一处理组件802和外围接口模块之间提供接口，上述外围接口模块可以是键盘，点击轮，按钮等。这些按钮可包括但不限于：主页按钮、音量按钮、启动按钮和锁定按钮。
传感器组件814包括一个或多个传感器，用于为电子设备800提供各个方面的状态评估。例如，传感器组件814可以检测到电子设备800的打开/关闭状态，组件的相对定位，例如所述组件为电子设备800的显示器和小键盘，传感器组件814还可以检测电子设备800或电子设备800一个组件的位置改变，用户与电子设备800接触的存在或不存在，电子设备800方位或加速/减速和电子设备800的温度变化。传感器组件814可以包括接近传感器，被配置用来在没有任何的物理接触时检测附近物体的存在。传感器组件814还可以包括光传感器，如互补金属氧化物半导体(Complementary Metal Oxide Semiconductor，CMOS)或电荷耦合器件(Charge Coupled Device，CCD)图像传感器，用于在成像应用中使用。在一些实施例中，该传感器组件814还可以包括加速度传感器，陀螺仪传感器，磁传感器，压力传感器或温度传感器。
通信组件816被配置为便于电子设备800和其他设备之间有线或无线方式的通信。电子设备800可以接入基于通信标准的无线网络，如WiFi，2G或3G，或它们的组合。在一个示例性实施例中，通信组件816经由广播信道接收来自外部广播管理系统的广播信号或广播相关信息。在一个示例性实施例中，所述通信组件816还包括近场通信(Near Field Communication，NFC)模块，以促进短程通信。例如，在NFC模块可基于射频识别(Radio Frequency Identification，RFID)技术，红外数据协会(Infrared Data Association，IrDA)技术，超宽带(Ultra Wide Band，UWB)技术，蓝牙(Bluetooth，BT)技术和其他技术来实现。
在示例性实施例中，电子设备800可以被一个或多个应用专用集成电路(Application Specific Integrated Circuit，ASIC)、数字信号处理器(Digital Signal Processor，DSP)、数字信号处理设备(Digital Signal Process，DSPD)、可编程逻辑器件(Programmable Logic Device，PLD)、现场可编程门阵列(Field Programmable Gate Array，FPGA)、控制器、微控制器、微处理器或其他电子元件实现，用于执行上述方法。
在示例性实施例中，还提供了一种非易失性计算机可读存储介质，例如包括计算机程序指令的第一存储器804，上述计算机程序指令可由电子设备800的处理器820执行以完成上述任意一种图像处理方法。
图7为本申请实施例的另一个电子设备的结构示意图，如图7所示，电子设备900可以被提供为一服务器。参照图7，电子设备900包括第二处理组件922，其进一步包括一个或多个处理器，以及由第二存储器932所代表的存储器资源，用于存储可由第二处理组件922的执行的指令，例如应用程序。第二存储器932中存储的应用程序可以包括一个或一个以上的每一个对应于一组指令的模块。此外，第二处理组件922被配置为执行指令，以执行上述方法。
电子设备900还可以包括一个第二电源组件926被配置为执行电子设备900的电源管 理，一个有线或无线网络接口950被配置为将电子设备900连接到网络，和第二输入输出(I/O)接口958。电子设备900可以操作基于存储在第二存储器932的操作系统，例如Windows ServerTM，Mac OS XTM，UnixTM,LinuxTM，FreeBSDTM或类似。
在示例性实施例中，还提供了一种非易失性计算机可读存储介质，例如包括计算机程序指令的第二存储器932，上述计算机程序指令可由电子设备900的第二处理组件922执行以完成上述方法。
本申请实施例可以是系统、方法和/或计算机程序产品。计算机程序产品可以包括计算机可读存储介质，其上载有用于使处理器实现本申请的各个方面的计算机可读程序指令。
计算机可读存储介质可以是可以保持和存储由指令执行设备使用的指令的有形设备。计算机可读存储介质例如可以是但不限于电存储设备、磁存储设备、光存储设备、电磁存储设备、半导体存储设备或者上述的任意合适的组合。计算机可读存储介质的更具体的例子(非穷举的列表)包括：便携式计算机盘、硬盘、随机存取存储器(RAM)、只读存储器(ROM)、可擦式可编程只读存储器(EPROM或闪存)、静态随机存取存储器(SRAM)、便携式压缩盘只读存储器(CD-ROM)、数字多功能盘(Digital Video Disc，DVD)、记忆棒、软盘、机械编码设备、例如其上存储有指令的打孔卡或凹槽内凸起结构、以及上述的任意合适的组合。这里所使用的计算机可读存储介质不被解释为瞬时信号本身，诸如无线电波或者其他自由传播的电磁波、通过波导或其他传输媒介传播的电磁波(例如，通过光纤电缆的光脉冲)、或者通过电线传输的电信号。
这里所描述的计算机可读程序指令可以从计算机可读存储介质下载到各个计算/处理设备，或者通过网络、例如因特网、局域网、广域网和/或无线网下载到外部计算机或外部存储设备。网络可以包括铜传输电缆、光纤传输、无线传输、路由器、防火墙、交换机、网关计算机和/或边缘服务器。每个计算/处理设备中的网络适配卡或者网络接口从网络接收计算机可读程序指令，并转发该计算机可读程序指令，以供存储在各个计算/处理设备中的计算机可读存储介质中。
用于执行本申请实施例操作的计算机程序指令可以是汇编指令、指令集架构(ISA)指令、机器指令、机器相关指令、微代码、固件指令、状态设置数据、或者以一种或多种编程语言的任意组合编写的源代码或目标代码，所述编程语言包括面向对象的编程语言—诸如Smalltalk、C++等，以及常规的过程式编程语言—诸如“C”语言或类似的编程语言。计算机可读程序指令可以完全地在用户计算机上执行、部分地在用户计算机上执行、作为一个独立的软件包执行、部分在用户计算机上部分在远程计算机上执行、或者完全在远程计算机或服务器上执行。在涉及远程计算机的情形中，远程计算机可以通过任意种类的网络—包括局域网(Local Area Network，LAN)或广域网(Wide Area Network，WAN)—连接到用户计算机，或者，可以连接到外部计算机(例如利用因特网服务提供商来通过因特网连接)。在一些实施例中，通过利用计算机可读程序指令的状态信息来个性化定制电子电路，例如可编程逻辑电路、FPGA或可编程逻辑阵列(Programmable Logic Array，PLA)，该电子电路可以执行计算机可读程序指令，从而实现本申请实施例的各个方面。
这里参照根据本申请实施例的方法、装置(系统)和计算机程序产品的流程图和/或框图描述了本申请实施例的各个方面。应当理解，流程图和/或框图的每个方框以及流程图和/或框图中各方框的组合，都可以由计算机可读程序指令实现。
这些计算机可读程序指令可以提供给通用计算机、专用计算机或其它可编程数据处理装置的处理器，从而生产出一种机器，使得这些指令在通过计算机或其它可编程数据处理装置的处理器执行时，产生了实现流程图和/或框图中的一个或多个方框中规定的功能/动作的装置。也可以把这些计算机可读程序指令存储在计算机可读存储介质中，这些 指令使得计算机、可编程数据处理装置和/或其他设备以特定方式工作，从而，存储有指令的计算机可读介质则包括一个制造品，其包括实现流程图和/或框图中的一个或多个方框中规定的功能/动作的各个方面的指令。
也可以把计算机可读程序指令加载到计算机、其它可编程数据处理装置、或其它设备上，使得在计算机、其它可编程数据处理装置或其它设备上执行一系列操作步骤，以产生计算机实现的过程，从而使得在计算机、其它可编程数据处理装置、或其它设备上执行的指令实现流程图和/或框图中的一个或多个方框中规定的功能/动作。
附图中的流程图和框图显示了根据本申请的多个实施例的系统、方法和计算机程序产品的可能实现的体系架构、功能和操作。在这点上，流程图或框图中的每个方框可以代表一个模块、程序段或指令的一部分，所述模块、程序段或指令的一部分包含一个或多个用于实现规定的逻辑功能的可执行指令。在有些作为替换的实现中，方框中所标注的功能也可以以不同于附图中所标注的顺序发生。例如，两个连续的方框实际上可以基本并行地执行，它们有时也可以按相反的顺序执行，这依所涉及的功能而定。也要注意的是，框图和/或流程图中的每个方框、以及框图和/或流程图中的方框的组合，可以用执行规定的功能或动作的专用的基于硬件的系统来实现，或者可以用专用硬件与计算机指令的组合来实现。
以上已经描述了本申请的各实施例，上述说明是示例性的，并非穷尽性的，并且也不限于所披露的各实施例。在不偏离所说明的各实施例的范围和精神的情况下，对于本技术领域的普通技术人员来说许多修改和变更都是显而易见的。本文中所用术语的选择，旨在最好地解释各实施例的原理、实际应用或对市场中技术的技术改进，或者使本技术领域的其它普通技术人员能理解本文披露的各实施例。
工业实用性
本申请实施例提供涉及一种图像处理方法及装置、电子设备、计算机存储介质和计算机程序，所述方法包括：将第一图像和第二图像输入目标分类网络，得到分类结果；根据所述分类结果，对变化数据进行预测，得到预测结果；所述变化数据表征：所述第一图像中的目标对象基于所述第二图像中的区域分布进行处理所得到的变化情况。

Claims (25)

  1. 一种图像处理方法，所述方法包括：
    将第一图像和第二图像输入目标分类网络，得到分类结果；
    根据所述分类结果，对变化数据进行预测，得到预测结果；所述变化数据表征：所述第一图像中的目标对象基于所述第二图像中的区域分布进行处理所得到的变化情况。
  2. 根据权利要求1所述的方法，其中，所述方法还包括：
    对分类网络进行训练，得到训练后的分类网络；
    将所述训练后的分类网络作为所述目标分类网络。
  3. 根据权利要求2所述的方法，其中，所述对分类网络进行训练，得到训练后的分类网络，包括：
    将第一图像和第二图像进行图像叠加处理后，得到待处理图像；所述第一图像和所述第二图像为不同种类的图像数据；
    将所述待处理图像作为训练样本输入所述分类网络进行训练，得到所述训练后的分类网络。
  4. 根据权利要求3所述的方法，其中，所述得到待处理图像包括：
    根据所述目标对象的轮廓，对所述第一图像及所述第二图像分别进行图像切割，得到切割后的第一子图像数据和第二子图像数据；
    将所述第一子图像数据和所述第二子图像数据作为所述待处理图像。
  5. 根据权利要求4所述的方法，其中，所述第一子图像数据和所述第二子图像数据，为相同尺寸的图像数据。
  6. 根据权利要求4或5所述的方法，其中，所述将所述待处理图像作为训练样本输入所述分类网络进行训练，得到所述训练后的分类网络，包括：
    将所述第一子图像数据和所述第二子图像数据分别转换为对应的直方图，并进行直方图的均衡化处理，得到均衡化处理结果；
    基于所述均衡化处理结果对所述分类网络进行训练，得到所述训练后的分类网络。
  7. 根据权利要求4或5所述的方法，其中，所述将所述待处理图像作为训练样本输入所述分类网络进行训练，得到所述训练后的分类网络，包括：
    将所述第一子图像数据和所述第二子图像数据中包含的对应像素点进行归一化处理，得到归一化处理结果；
    基于所述归一化处理结果对所述分类网络进行训练，得到所述训练后的分类网络。
  8. 根据权利要求3-7任一项所述的方法，其中，所述分类网络包括至少一个分类处理模块；
    所述将所述待处理图像作为训练样本输入所述分类网络进行训练，得到所述训练后的分类网络，包括：
    将所述待处理图像通过所述至少一个分类处理模块进行特征提取、降维处理及全局平均池化处理，得到损失函数；
    根据所述损失函数的反向传播训练所述分类网络，以得到所述训练后的分类网络。
  9. 根据权利要求8所述的方法，其中，每个所述分类处理模块至少包括卷积层；
    所述将所述待处理图像通过所述至少一个分类处理模块进行特征提取、降维处理及全局平均池化处理，得到损失函数，包括：
    将所述待处理图像通过所述至少一个分类处理模块中的对应卷积层进行特征提取后，进行降维处理，得到第一处理结果；
    将所述第一处理结果进行全局平均池化处理后输入全连接层，得到第二处理结果，所述第二处理结果用于表征所提取特征的预测变化情况；
    根据所述第二处理结果和手工标注结果，得到所述损失函数。
  10. 根据权利要求9所述的方法，其中，在所述分类处理模块为残差模块的情况下，每个所述残差模块包括：卷积层、正则化层和激活层；
    所述将所述待处理图像通过所述至少一个分类处理模块中的对应卷积层进行特征提取后，还包括：
    将所述待处理图像通过至少一个残差模块中的对应卷积层进行特征提取后得到的第一提取结果，经过正则化层和激活层的处理后得到第二提取结果；
    根据所述第二提取结果和所述待处理图像，得到第三提取结果。
  11. 根据权利要求10所述的方法，其中，所述进行降维处理，得到第一处理结果，包括：
    根据所述第三提取结果，进行降维处理，得到所述第一处理结果。
  12. 一种图像处理装置，所述装置包括：
    分类部分，配置为将第一图像和第二图像输入目标分类网络，得到分类结果；
    预测部分，配置为根据所述分类结果，对变化数据进行预测，得到预测结果；所述变化数据表征：所述第一图像中的目标对象基于所述第二图像中的区域分布进行处理所得到的变化情况。
  13. 根据权利要求12所述的装置，其中，所述装置还包括训练部分，配置为：
    对分类网络进行训练，得到训练后的分类网络；
    将所述训练后的分类网络作为所述目标分类网络。
  14. 根据权利要求13所述的装置，其中，所述训练部分，包括：
    叠加子部分，配置为将第一图像和第二图像进行图像叠加处理后，得到待处理图像；所述第一图像和所述第二图像为不同种类的图像数据；
    训练子部分，配置为将所述待处理图像作为训练样本输入所述分类网络进行训练，得到所述训练后的分类网络。
  15. 根据权利要求14所述的装置，其中，所述训练部分，还包括：
    切割子部分，配置为根据所述目标对象的轮廓，对所述第一图像及所述第二图像分别进行图像切割，得到切割后的第一子图像数据和第二子图像数据；将所述第一子图像数据和所述第二子图像数据作为所述待处理图像。
  16. 根据权利要求15所述的装置，其中，所述第一子图像数据和所述第二子图像数据，为相同尺寸的图像数据。
  17. 根据权利要求15或16所述的装置，其中，所述训练部分，还包括：
    均衡化处理子部分，配置为将所述第一子图像数据和所述第二子图像数据分别转换为对应的直方图，并进行直方图的均衡化处理，得到均衡化处理结果；基于所述均衡化处理结果对所述分类网络进行训练，得到所述训练后的分类网络。
  18. 根据权利要求15或16所述的装置，其中，所述训练部分，还包括：
    归一化处理子部分，配置为将所述第一子图像数据和所述第二子图像数据中包含的对应像素点进行归一化处理，得到归一化处理结果；基于所述归一化处理结果对所述分类网络进行训练，得到所述训练后的分类网络。
  19. 根据权利要求14至18任一项所述的装置，其中，所述分类网络包括至少一个分类处理模块；
    所述训练子部分，配置为将所述待处理图像通过所述至少一个分类处理模块进行特征提取、降维处理及全局平均池化处理，得到损失函数；根据所述损失函数的反向传播训练所述分类网络，以得到所述训练后的分类网络。
  20. 根据权利要求19所述的装置，其中，每个所述分类处理模块至少包括卷积层；
    所述训练子部分，配置为将所述待处理图像通过所述至少一个分类处理模块中的对应卷积层进行特征提取后，进行降维处理，得到第一处理结果；将所述第一处理结果进行全局平均池化处理后输入全连接层，得到第二处理结果，所述第二处理结果用于表征所提取特征的预测变化情况；根据所述第二处理结果和手工标注结果，得到所述损失函数。
  21. 根据权利要求20所述的装置，其中，在所述分类处理模块为残差模块的情况下，每个残差模块包括：卷积层、正则化层和激活层；
    所述训练子部分，配置为将所述待处理图像通过至少一个残差模块中的对应卷积层进行特征提取后得到的第一提取结果，经过正则化层和激活层的处理，得到第二提取结果；根据所述第二提取结果和所述待处理图像，得到第三提取结果。
  22. 根据权利要求21所述的装置，其中，所述训练子部分，配置为根据所述第三提取结果，进行降维处理，得到所述第一处理结果。
  23. 一种电子设备，包括：
    处理器；
    配置为存储处理器可执行指令的存储器；
    其中，所述处理器被配置为：执行权利要求1至11中任意一项所述的方法。
  24. 一种计算机可读存储介质，其上存储有计算机程序指令，所述计算机程序指令被处理器执行时实现权利要求1至11中任意一项所述的方法。
  25. 一种计算机程序，包括计算机可读代码，当所述计算机可读代码在电子设备中运行时，所述电子设备中的处理器执行用于实现权利要求1至11中的任一权利要求所述的方法。
PCT/CN2020/100692 2019-09-26 2020-07-07 图像处理方法及装置、电子设备、存储介质和计算机程序 WO2021057174A1 (zh)

Priority Applications (2)

Application Number Priority Date Filing Date Title
KR1020217037806A KR20210153700A (ko) 2019-09-26 2020-07-07 이미지 처리 방법 및 장치, 전자 기기, 저장 매체 및 컴퓨터 프로그램
JP2021563034A JP2022530413A (ja) 2019-09-26 2020-07-07 画像処理方法及び装置、電子機器、記憶媒体並びにコンピュータプログラム

Applications Claiming Priority (2)

Application Number Priority Date Filing Date Title
CN201910918450.1 2019-09-26
CN201910918450.1A CN110705626A (zh) 2019-09-26 2019-09-26 一种图像处理方法及装置、电子设备和存储介质

Publications (1)

Publication Number Publication Date
WO2021057174A1 true WO2021057174A1 (zh) 2021-04-01

Family

ID=69197459

Family Applications (1)

Application Number Title Priority Date Filing Date
PCT/CN2020/100692 WO2021057174A1 (zh) 2019-09-26 2020-07-07 图像处理方法及装置、电子设备、存储介质和计算机程序

Country Status (5)

Country Link
JP (1) JP2022530413A (zh)
KR (1) KR20210153700A (zh)
CN (1) CN110705626A (zh)
TW (1) TW202113871A (zh)
WO (1) WO2021057174A1 (zh)

Cited By (6)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN113140012A (zh) * 2021-05-14 2021-07-20 北京字节跳动网络技术有限公司 图像处理方法、装置、介质及电子设备
CN113706583A (zh) * 2021-09-01 2021-11-26 上海联影医疗科技股份有限公司 图像处理方法、装置、计算机设备和存储介质
CN114120420A (zh) * 2021-12-01 2022-03-01 北京百度网讯科技有限公司 图像检测方法和装置
CN114298177A (zh) * 2021-12-16 2022-04-08 广州瑞多思医疗科技有限公司 一种适用于深度学习训练数据的扩充增强方法、系统及可读存储介质
CN117152128A (zh) * 2023-10-27 2023-12-01 首都医科大学附属北京天坛医院 神经影像的病灶识别方法、装置、电子设备和存储介质
CN113140012B (zh) * 2021-05-14 2024-05-31 北京字节跳动网络技术有限公司 图像处理方法、装置、介质及电子设备

Families Citing this family (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN110705626A (zh) * 2019-09-26 2020-01-17 北京市商汤科技开发有限公司 一种图像处理方法及装置、电子设备和存储介质
CN111368923B (zh) * 2020-03-05 2023-12-19 上海商汤智能科技有限公司 神经网络训练方法及装置、电子设备和存储介质
CN113689355B (zh) * 2021-09-10 2022-07-08 数坤(北京)网络科技股份有限公司 图像处理方法、装置、存储介质及计算机设备
KR20230125412A (ko) * 2022-02-21 2023-08-29 계명대학교 산학협력단 컴퓨터 단층 촬영 영상의 라디오믹스 파라미터를 이용한 요로결석의 요산석 분류 방법 및 분석장치

Citations (5)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US20180247410A1 (en) * 2017-02-27 2018-08-30 Case Western Reserve University Predicting immunotherapy response in non-small cell lung cancer with serial radiomics
CN108766563A (zh) * 2018-05-25 2018-11-06 戴建荣 基于剂量组学的放射治疗结果预测方法和系统
CN108830028A (zh) * 2017-05-04 2018-11-16 戴立言 一种设备及非诊断性确定对象功能等效均匀剂量的方法
CN109966662A (zh) * 2019-04-30 2019-07-05 四川省肿瘤医院 一种验证放射治疗剂量的方法及系统
CN110705626A (zh) * 2019-09-26 2020-01-17 北京市商汤科技开发有限公司 一种图像处理方法及装置、电子设备和存储介质

Family Cites Families (8)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
JPH02294786A (ja) * 1989-05-09 1990-12-05 Hitachi Ltd 画像輝度レベル値の自動正規化方法
KR102020383B1 (ko) * 2016-11-28 2019-09-10 고려대학교산학협력단 데이터마이닝을 이용한 방사선 치료 계획의 검증 방법
US11517768B2 (en) * 2017-07-25 2022-12-06 Elekta, Inc. Systems and methods for determining radiation therapy machine parameter settings
CN107403201A (zh) * 2017-08-11 2017-11-28 强深智能医疗科技（昆山）有限公司 肿瘤放射治疗靶区和危及器官智能化、自动化勾画方法
CN108717866B (zh) * 2018-04-03 2022-10-11 中国医学科学院肿瘤医院 一种预测放疗计划剂量分布的方法、装置、设备及存储介质
CN109166613A (zh) * 2018-08-20 2019-01-08 北京东方瑞云科技有限公司 基于机器学习的放射治疗计划评估系统及方法
CN109523532B (zh) * 2018-11-13 2022-05-03 腾讯医疗健康(深圳)有限公司 图像处理方法、装置、计算机可读介质及电子设备
CN110211664B (zh) * 2019-04-25 2022-11-04 安徽大学 一种基于机器学习自动设计放射治疗方案的系统

Patent Citations (5)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US20180247410A1 (en) * 2017-02-27 2018-08-30 Case Western Reserve University Predicting immunotherapy response in non-small cell lung cancer with serial radiomics
CN108830028A (zh) * 2017-05-04 2018-11-16 戴立言 一种设备及非诊断性确定对象功能等效均匀剂量的方法
CN108766563A (zh) * 2018-05-25 2018-11-06 戴建荣 基于剂量组学的放射治疗结果预测方法和系统
CN109966662A (zh) * 2019-04-30 2019-07-05 四川省肿瘤医院 一种验证放射治疗剂量的方法及系统
CN110705626A (zh) * 2019-09-26 2020-01-17 北京市商汤科技开发有限公司 一种图像处理方法及装置、电子设备和存储介质

Non-Patent Citations (1)

* Cited by examiner, † Cited by third party
Title
ZHANG JUNQIAN ,ZHANG YUAN ,YIN YONG , ZHU JIAN ,LI BAOSHENG: "A review of machine learning in tumor radiotherapy", JOURNAL OF BIOMEDICAL ENGINEERING, vol. 36, no. 5, 16 September 2019 (2019-09-16), pages 879 - 884, XP055794284, ISSN: 1001-5551, DOI: 10.7507/1001-5515.201810051 *

Cited By (9)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN113140012A (zh) * 2021-05-14 2021-07-20 北京字节跳动网络技术有限公司 图像处理方法、装置、介质及电子设备
CN113140012B (zh) * 2021-05-14 2024-05-31 北京字节跳动网络技术有限公司 图像处理方法、装置、介质及电子设备
CN113706583A (zh) * 2021-09-01 2021-11-26 上海联影医疗科技股份有限公司 图像处理方法、装置、计算机设备和存储介质
CN113706583B (zh) * 2021-09-01 2024-03-22 上海联影医疗科技股份有限公司 图像处理方法、装置、计算机设备和存储介质
CN114120420A (zh) * 2021-12-01 2022-03-01 北京百度网讯科技有限公司 图像检测方法和装置
CN114120420B (zh) * 2021-12-01 2024-02-13 北京百度网讯科技有限公司 图像检测方法和装置
CN114298177A (zh) * 2021-12-16 2022-04-08 广州瑞多思医疗科技有限公司 一种适用于深度学习训练数据的扩充增强方法、系统及可读存储介质
CN117152128A (zh) * 2023-10-27 2023-12-01 首都医科大学附属北京天坛医院 神经影像的病灶识别方法、装置、电子设备和存储介质
CN117152128B (zh) * 2023-10-27 2024-02-27 首都医科大学附属北京天坛医院 神经影像的病灶识别方法、装置、电子设备和存储介质

Also Published As

Publication number Publication date
JP2022530413A (ja) 2022-06-29
CN110705626A (zh) 2020-01-17
TW202113871A (zh) 2021-04-01
KR20210153700A (ko) 2021-12-17

Similar Documents

Publication Publication Date Title
WO2021057174A1 (zh) 图像处理方法及装置、电子设备、存储介质和计算机程序
WO2021147257A1 (zh) 网络训练、图像处理方法及装置、电子设备和存储介质
JP6134446B2 (ja) 画像分割方法、画像分割装置、画像分割デバイス、プログラム及び記録媒体
CN111310616B (zh) 图像处理方法及装置、电子设备和存储介质
CN109978886B (zh) 图像处理方法及装置、电子设备和存储介质
CN107798669B (zh) 图像去雾方法、装置及计算机可读存储介质
CN109871883B (zh) 神经网络训练方法及装置、电子设备和存储介质
EP3869456A1 (en) Image segmentation method and apparatus, computer device and storage medium
WO2020078268A1 (zh) 图像分割方法、装置、计算机设备及存储介质
CN112767329B (zh) 图像处理方法及装置、电子设备
WO2021135145A1 (zh) 图像分割方法及装置、电子设备和存储介质
JP7152598B2 (ja) 画像処理方法及び装置、電子機器並びに記憶媒体
WO2020134866A1 (zh) 关键点检测方法及装置、电子设备和存储介质
CN105554389B (zh) 拍摄方法及装置
CN110674719A (zh) 目标对象匹配方法及装置、电子设备和存储介质
CN112541928A (zh) 网络训练方法及装置、图像分割方法及装置和电子设备
JP2021503659A (ja) 生体検出方法、装置及びシステム、電子機器並びに記憶媒体
TW202203153A (zh) 圖像處理方法、電子設備和電腦可讀儲存介質
TW202209254A (zh) 圖像分割方法、電子設備和電腦可讀儲存介質
WO2022156235A1 (zh) 神经网络训练和图像处理方法及装置、电子设备及存储介质
US20200195905A1 (en) Method and apparatus for obtaining image, storage medium and electronic device
CN114494442A (zh) 图像处理方法、装置及设备
TWI765386B (zh) 神經網路訓練及圖像分割方法、電子設備和電腦儲存介質
CN115512116A (zh) 图像分割模型优化方法、装置、电子设备及可读存储介质
KR102374400B1 (ko) 이미지 처리 방법 및 장치, 전자 기기와 저장 매체

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: 20869215

Country of ref document: EP

Kind code of ref document: A1

ENP Entry into the national phase

Ref document number: 2021563034

Country of ref document: JP

Kind code of ref document: A

ENP Entry into the national phase

Ref document number: 20217037806

Country of ref document: KR

Kind code of ref document: A

NENP Non-entry into the national phase

Ref country code: DE

122 Ep: pct application non-entry in european phase

Ref document number: 20869215

Country of ref document: EP

Kind code of ref document: A1