WO2020258507A1 - X光片的分类方法、装置、终端及存储介质 - Google Patents

X光片的分类方法、装置、终端及存储介质 Download PDF

Info

Publication number
WO2020258507A1
WO2020258507A1 PCT/CN2019/103655 CN2019103655W WO2020258507A1 WO 2020258507 A1 WO2020258507 A1 WO 2020258507A1 CN 2019103655 W CN2019103655 W CN 2019103655W WO 2020258507 A1 WO2020258507 A1 WO 2020258507A1
Authority
WO
WIPO (PCT)
Prior art keywords
ray film
classification
label
sample set
result
Prior art date
Application number
PCT/CN2019/103655
Other languages
English (en)
French (fr)
Inventor
刘新卉
王健宗
Original Assignee
平安科技（深圳）有限公司
Priority date (The priority date is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the date listed.)
Filing date
Publication date
Application filed by 平安科技（深圳）有限公司 filed Critical 平安科技（深圳）有限公司
Publication of WO2020258507A1 publication Critical patent/WO2020258507A1/zh

Links

Images

Classifications

    • 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
    • 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
    • G06F18/2413—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches based on distances to training or reference patterns
    • G06F18/24133—Distances to prototypes
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/04—Architecture, e.g. interconnection topology
    • G06N3/045—Combinations of networks
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/08—Learning methods
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
    • G06V10/00—Arrangements for image or video recognition or understanding
    • G06V10/20—Image preprocessing
    • G06V10/25—Determination of region of interest [ROI] or a volume of interest [VOI]
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
    • G06V10/00—Arrangements for image or video recognition or understanding
    • G06V10/20—Image preprocessing
    • G06V10/30—Noise filtering
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
    • G06V2201/00—Indexing scheme relating to image or video recognition or understanding
    • G06V2201/03—Recognition of patterns in medical or anatomical images
    • Y—GENERAL TAGGING OF NEW TECHNOLOGICAL DEVELOPMENTS; GENERAL TAGGING OF CROSS-SECTIONAL TECHNOLOGIES SPANNING OVER SEVERAL SECTIONS OF THE IPC; TECHNICAL SUBJECTS COVERED BY FORMER USPC CROSS-REFERENCE ART COLLECTIONS [XRACs] AND DIGESTS
    • Y02—TECHNOLOGIES OR APPLICATIONS FOR MITIGATION OR ADAPTATION AGAINST CLIMATE CHANGE
    • Y02A—TECHNOLOGIES FOR ADAPTATION TO CLIMATE CHANGE
    • Y02A90/00—Technologies having an indirect contribution to adaptation to climate change
    • Y02A90/10—Information and communication technologies [ICT] supporting adaptation to climate change, e.g. for weather forecasting or climate simulation

Definitions

  • This application relates to the field of neural network technology, and in particular to a classification method, device, terminal and storage medium of X-ray films.
  • X-ray film is the most commonly used medical imaging examination tool.
  • One X-ray film can detect a variety of diseases, and it occupies a vital position in the process of screening, diagnosis and management of diseases including pneumonia.
  • the existing X-ray film classification method usually adopts a method of obtaining a target result from one label information in the X-ray film.
  • Each label information is independent of each other, the classification accuracy is low, and the classification effect is poor. .
  • This application provides a classification method, device, terminal and storage medium for X-ray films to solve the current X-ray film classification methods, which cannot make full use of the relationship between the label information of X-ray films, and the classification accuracy is low. The problem of poor results.
  • This application provides a method for classifying X-ray films, including the following steps:
  • Obtain a target X-ray film to be classified extract all label information in the target X-ray film, and input the label information into the X-ray film classification model for classification and recognition, and obtain a classification result.
  • An X-ray film classification device provided in this application includes:
  • the acquisition module is used to acquire the X-ray film sample set, perform noise reduction processing on the X-ray film in the X-ray film sample set, and extract the key information area of the X-ray film after the noise reduction processing by a positioning method Multiple label information;
  • the calculation module is used to calculate the degree of association between the two pieces of label information in the X-ray film, and the degree of association greater than a set threshold and the two label information corresponding to the degree of association form a label sample, and multiple label samples form Label sample set;
  • the construction module is used to construct a neural network model of the X-ray film, and train the neural network model according to the label sample set to obtain the X-ray film classification model;
  • the classification and recognition module is used to obtain a target X-ray film to be classified, extract all label information in the target X-ray film, and input the label information into the X-ray film classification model for classification and recognition, and obtain a classification result.
  • the present application provides a terminal, including a memory and a processor.
  • the memory stores computer-readable instructions.
  • the processor executes an X-ray film The steps of the classification method;
  • the X-ray film classification method includes the following steps:
  • Obtain a target X-ray film to be classified extract all label information in the target X-ray film, and input the label information into the X-ray film classification model for classification and recognition, and obtain a classification result.
  • This application provides a storage medium on which a computer program is stored, and when the computer program is executed by a processor, a method for classifying X-ray films is realized;
  • the X-ray film classification method includes the following steps:
  • Obtain a target X-ray film to be classified extract all label information in the target X-ray film, and input the label information into the X-ray film classification model for classification and recognition, and obtain a classification result.
  • the X-ray film classification method provided in this application firstly obtains an X-ray film sample set, and performs noise reduction processing on the X-ray film in the X-ray film sample set to minimize the interference information in the obtained X-ray film. Remove; and extract multiple label information from the key information area of the X-ray film after the noise reduction process by positioning; calculate the correlation between the two label information in the X-ray film, and will be greater than the set.
  • the correlation degree with a predetermined threshold and the two label information corresponding to the correlation degree form a label sample, and multiple label samples form a label sample set; then a neural network model of the X-ray film is constructed, and the neural network is compared according to the label sample set.
  • the model is trained to obtain the X-ray film classification model; finally the target X-ray film to be classified is obtained, all the label information in the target X-ray film is extracted, and the label information is input into the X-ray film classification model for classification and recognition , Get the classification result.
  • the application classifies the X-ray film based on the degree of association between each label information, it makes full use of the associated information between each label information to improve the accuracy of classification.
  • FIG. 1 is an implementation environment diagram of the X-ray film classification method provided in an embodiment of the application
  • FIG. 2 is a flowchart of an embodiment of a method for classifying X-ray films of this application
  • FIG. 3 is a block diagram of an embodiment of an apparatus for classifying X-ray films of this application
  • Fig. 4 is a block diagram of the internal structure of a terminal in an embodiment of the application.
  • FIG. 1 is an implementation environment diagram of the X-ray film classification method provided in an embodiment.
  • the implementation environment includes a server 110 and a terminal 120.
  • the terminal 120 is connected to a server through a network.
  • a client or a browser is installed on the terminal 120.
  • the user can upload the target X-ray film to the server 110 through the client or the browser.
  • the server 110 is classified and processed, the classification result will be obtained Return to the terminal 120.
  • the aforementioned network may include the Internet, 2G/3G/4G, wifi, and so on.
  • the server 110 may be an independent physical server or terminal, or a server cluster composed of multiple physical servers, and may be a cloud server that provides basic cloud computing services such as cloud servers, cloud databases, cloud storage, and CDN.
  • the terminal 120 may be a smart phone, a tablet computer, a notebook computer, a desktop computer, a smart watch, etc., but is not limited to this.
  • the classification method of X-ray film provided by this application is to solve the current classification method of X-ray film, which cannot make full use of the relationship between the label information of X-ray film, and the classification accuracy is low.
  • One of the embodiments includes the following steps:
  • a collection of pictures can be downloaded through the network, a number of X-ray films can be screened out from the collection as X-ray film sample sets, and then the X-ray films in the X-ray film sample sets are subjected to noise reduction processing to The interference information in the obtained X-ray film is removed as much as possible to improve the accuracy of subsequent identification of the label information in the X-ray film. And by positioning, extract multiple label information from the key information area of each X-ray film after noise reduction processing.
  • the label information includes human organ information and health physiological information or pathological information reflected by the human organ information.
  • the X-ray sample set can also be obtained in other ways. Taking chest X-rays as an example, the chest X-ray data set ChestX-ray14 provided by NIH (National Institutes of Health) can be used as the chest X-ray film sample set.
  • the nearest neighbor sampling interpolation algorithm may be used to automatically scale the X-ray film; and an image decontamination algorithm may be used to remove stains on the X-ray film after the automatic scaling process. And scratch processing; then use angle detection algorithm and image fast rotation algorithm for angle processing and reflection processing.
  • the angle detection algorithm is an algorithm for finding the angle feature on the image through mathematical calculation, and it has the characteristic of rotation invariance.
  • the interference information in the obtained X-ray film is removed as much as possible, and the accuracy of subsequent identification of the label information in the X-ray film is improved.
  • the contour area reflecting the human tissue in the X-ray film can be located, and A number of position coordinates are set on the contour area, and then the label information within a predetermined range is extracted with the position coordinates as the center, so as to obtain the label information of the key information area in the X-ray film.
  • the label information may include text information.
  • the label information includes: atrial septal defect, ventricular septal defect, patent ductus arteriosus, etc.; label information corresponding to heart valve disease includes: mitral valve stenosis and insufficiency, aortic valve stenosis and insufficiency, etc., calculate the label
  • the correlation between information the correlation between atrial septal defect and ventricular septal defect is higher, while the correlation between atrial septal defect and mitral valve stenosis is relatively low, and so on, we get the difference between every two label information
  • the degree of relevance and then filter out the degree of relevance with a value greater than the set threshold and the two label information corresponding to the degree of relevance to form a label sample, and multiple label samples into a label sample set, that is, the label sample set includes multiple Label
  • a convolutional neural network model can be constructed, and a label sample set is input into the convolutional neural network model for training, until the convolutional neural network model converges, a trained X-ray film classification model is obtained.
  • the classification effect of the X-ray film classification model obtained by training is better. Therefore, before training the convolutional neural network model, obtain as much as possible Label information.
  • all label information is extracted from the target X-ray film to be classified, and the label information is input into the X-ray film classification model obtained by training for classification and recognition.
  • the X-ray film classification model calculates the target X-ray film The degree of association between every two pieces of label information, and the target X-ray film is classified according to the degree of association and label information to obtain the classification result.
  • the label information When the label information is throat inflammation, the corresponding classification result may be hot or cold. In the classification, other label information needs to be combined to form multiple label information with a relatively high degree of correlation with throat inflammation.
  • Sample set input the label set into the classification model to classify, and get the classification result. For example, when the label information of throat inflammation is accompanied by runny nose and other label information, the final classification result can be obtained as a cold instead of getting angry. Improve classification accuracy.
  • the X-ray film classification method provided in this application firstly obtains an X-ray film sample set, and performs noise reduction processing on the X-ray film in the X-ray film sample set to minimize the interference information in the obtained X-ray film. Remove; and extract multiple label information from the key information area of the X-ray film after the noise reduction process by positioning; calculate the correlation between the two label information in the X-ray film, and will be greater than the set.
  • the correlation degree with a predetermined threshold and the two label information corresponding to the correlation degree form a label sample, and multiple label samples form a label sample set; then a neural network model of the X-ray film is constructed, and the neural network is compared according to the label sample set.
  • the model is trained to obtain the X-ray film classification model; finally the target X-ray film to be classified is obtained, all the label information in the target X-ray film is extracted, and the label information is input into the X-ray film classification model for classification and recognition , Get the classification result.
  • the application classifies the X-ray film based on the degree of association between each label information, it makes full use of the associated information between each label information to improve the accuracy of classification.
  • the step of constructing the neural network model of the X-ray film, and training the neural network model according to the label sample set to obtain the X-ray film classification model may specifically include:
  • the X-ray film classification model is obtained.
  • the classifier is a general term for methods for classifying samples in data mining, and includes algorithms such as decision trees, logistic regression, naive Bayes, and neural networks.
  • the loss function is used to estimate the degree of inconsistency between the predicted value f(x) of your model and the true value Y. It is a non-negative real-valued function, usually represented by L(Y, f(x)), The smaller the result of the loss function, the better the robustness of the neural network model. Therefore, the classification effect of the X-ray film classification model can be judged by the loss function.
  • the MSML loss function is a boundary sample mining loss function, which can push the boundary between positive and negative samples. It also takes into account the relative distance and the absolute distance and introduces the metric learning method of difficult sample sampling.
  • the neural network model of this embodiment constructs two classifiers: a first classifier and a second classifier.
  • first input the label information of the label sample set into the first classifier to obtain the first classification result
  • second classifier uses the MSML loss function, which can be used to learn the relationship between different labels to improve the accuracy of classification.
  • the first classifier uses a first classification function
  • the first classification loss function is:
  • the MSML loss function is:
  • l and k represent label samples
  • Y i represents a positive sample in the label sample set
  • the step of inputting the correlation between the first classification result and the label sample set into the second classifier to obtain the target training result may specifically include:
  • the bilinear pooling processing result is subjected to convolution processing to obtain the target training result.
  • the first classifier After the first classifier classifies, it simply classifies the label information, and each label information is independent of each other, and does not consider the associated information between the label information. Therefore, in this embodiment, the second classifier is combined with the correlation degree of the label sample set to adjust the first classification result to obtain the second classification result, so as to consider the correlation information between the independent label information to classify the first classifier. After the first classification result is calibrated, the accuracy of classification is improved.
  • the bilinear pooling process can merge the features obtained by the two classifiers.
  • the process of merging can obtain the paired correlation relationship of the feature maps of the two classifiers, so that the features of the two classifiers can be used in combination. Then fully consider the relationship between the label information, so that using richer features can get a better classification effect.
  • the step of obtaining an X-ray film classification model when the target training result meets a preset requirement may specifically include:
  • the classification results of the two classifiers can be subjected to bilinear pooling operation and convolution processing, and then use the fine-grained cross-entropy loss function to calculate the total loss of the neural network model of the X-ray film, and determine the total loss Whether the loss is lower than the preset value; if so, the X-ray film classification model is obtained; otherwise, the relevant parameters of the two classifiers are adjusted, and the training of the two classifiers continues until the neural network model of the X-ray film is qualified.
  • the formula for the total loss is as follows:
  • E total ⁇ (E MSML +E CE )+ ⁇ E FCE ;
  • the E total represents the total loss
  • the E CE represents the first loss function
  • the E MSML represents the MSML loss function
  • the E FCE represents the fine-grained cross-entropy loss function.
  • the loss calculation results of the first classifier and the second classifier are combined to obtain an X-ray film classification model with better training effect.
  • the step of the bilinear pooling process may include the following formula:
  • the output of one pooling layer combines the results of two classifiers, thereby improving the classification accuracy of the X-ray film classification model.
  • the method may further include:
  • the X-ray film classification model with successful verification is generated.
  • ChestX-ray14 a chest X-ray film data set provided by NIH. It contains a total of 112,120 frontal chest X-rays and a total of 14 diseases. The size of each image is 1024 ⁇ 1024.
  • the first part contains 70% of the X-rays in the data set as a training set; 10% of the X-rays are used for verification; and the last 20% of the X-rays are used for testing.
  • the purpose of dividing the original data into three sets is to select the X-ray film classification model with the best effect and the best generalization ability.
  • the function of the training set is to fit the X-ray film classification model, and train the X-ray film classification model by setting the parameters of the classifier in the X-ray film classification.
  • different values of the same parameter will be selected to fit the classifier.
  • the verification set is used when multiple classification models are trained through the training set, in order to find the best X-ray film classification model, each model is used to predict the verification set data, and the accuracy of the model is recorded.
  • the parameters corresponding to the best model are selected to adjust the model parameters.
  • the ratio between the test set, training set and validation set can also be set according to the actual classification effect.
  • the method further includes:
  • the X-ray film test set is used to test the successfully verified X-ray film classification model, and when the test result is qualified, the tested X-ray film classification model is obtained.
  • the test set is used to finally evaluate the performance and classification ability of the pattern recognition system. That is, the test set can be regarded as a data set that never exists. After the model parameters have been determined, the test set can be used to make model predictions and evaluate the performance of the model.
  • an embodiment of the present application also provides a classification device for X-ray films.
  • it includes an acquisition module 31, a calculation module 32, a construction module 33, and a classification recognition module 34. among them,
  • the acquiring module 31 is configured to acquire an X-ray film sample set, perform noise reduction processing on the X-ray film in the X-ray film sample set, and extract the key information area of the X-ray film after the noise reduction process by a positioning method Multiple label information;
  • a collection of pictures can be downloaded through the network, a number of X-ray films can be screened out from the collection as X-ray film sample sets, and then the X-ray films in the X-ray film sample sets are subjected to noise reduction processing to The interference information in the obtained X-ray film is removed as much as possible to improve the accuracy of subsequent identification of the label information in the X-ray film. And by positioning, extract multiple label information from the key information area of each X-ray film after noise reduction processing.
  • the label information includes human organ information and health physiological information or pathological information reflected by the human organ information.
  • the X-ray sample set can also be obtained in other ways. Taking chest X-rays as an example, the chest X-ray data set ChestX-ray14 provided by NIH (National Institutes of Health) can be used as the chest X-ray film sample set.
  • the nearest neighbor sampling interpolation algorithm may be used to automatically scale the X-ray film; and an image decontamination algorithm may be used to remove stains on the X-ray film after the automatic scaling process. And scratch processing; then use angle detection algorithm and image fast rotation algorithm for angle processing and reflection processing.
  • the angle detection algorithm is an algorithm for finding the angle feature on the image through mathematical calculation, and it has the characteristic of rotation invariance.
  • the interference information in the obtained X-ray film is removed as much as possible, and the accuracy of subsequent identification of the label information in the X-ray film is improved.
  • the contour area reflecting the human tissue in the X-ray film can be located, and A number of position coordinates are set on the contour area, and then the label information within a predetermined range is extracted with the position coordinates as the center, so as to obtain the label information of the key information area in the X-ray film.
  • the label information to be extracted can also be classified in advance.
  • the type of the label information is determined first, and then the position coordinates corresponding to the label information type are obtained.
  • the label information in a predetermined range centered on the position coordinates can be used to obtain the required type of label information, thereby reducing the extraction of irrelevant information and improving the efficiency and accuracy of subsequent neural network model training.
  • the calculation module 32 is configured to calculate the degree of association between the pair of label information in the X-ray film, and form a label sample with the degree of association greater than a set threshold and the two label information corresponding to the degree of association. Form a label sample set;
  • the label information may include text information.
  • the label information includes: atrial septal defect, ventricular septal defect, patent ductus arteriosus, etc.; label information corresponding to heart valve disease includes: mitral valve stenosis and insufficiency, aortic valve stenosis and insufficiency, etc., calculate the label
  • the correlation between information the correlation between atrial septal defect and ventricular septal defect is higher, while the correlation between atrial septal defect and mitral valve stenosis is relatively low, and so on, we get the difference between every two label information
  • the degree of relevance and then filter out the degree of relevance with a value greater than the set threshold and the two label information corresponding to the degree of relevance to form a label sample, and multiple label samples into a label sample set, that is, the label sample set includes multiple Label
  • the construction module 33 is configured to construct a neural network model of the X-ray film, and train the neural network model according to the label sample set to obtain the X-ray film classification model;
  • a convolutional neural network model can be constructed, and a label sample set is input into the convolutional neural network model for training, until the convolutional neural network model converges, a trained X-ray film classification model is obtained.
  • the classification effect of the X-ray film classification model obtained by training is better. Therefore, before training the convolutional neural network model, obtain as much as possible Label information.
  • the classification and recognition module 34 is configured to obtain a target X-ray film to be classified, extract all label information in the target X-ray film, and input the label information into the X-ray film classification model for classification and recognition, and obtain a classification result.
  • all label information is extracted from the target X-ray film to be classified, and the label information is input into the X-ray film classification model obtained by training for classification and recognition.
  • the X-ray film classification model calculates the target X-ray film The degree of association between every two pieces of label information, and the target X-ray film is classified according to the degree of association and label information to obtain the classification result.
  • the label information When the label information is inflammation of the throat, the corresponding classification result may be an angry or a cold.
  • other label information needs to be combined to collect information labels with a relatively high correlation with throat inflammation to form a label set ,
  • the label set is input into the classification model for classification, and the classification result is obtained.
  • label information of throat inflammation there is also label information such as runny nose, and the final classification result can be obtained as a cold instead of getting angry, thereby improving the classification Accuracy.
  • the X-ray film classification device first obtains an X-ray film sample set, and performs noise reduction processing on the X-ray film in the X-ray film sample set to minimize the interference information in the obtained X-ray film. Remove; and extract multiple label information from the key information area of the X-ray film after the noise reduction process by positioning; calculate the correlation between the two label information in the X-ray film, and will be greater than the set.
  • the correlation degree with a predetermined threshold and the two label information corresponding to the correlation degree form a label sample, and multiple label samples form a label sample set; then a neural network model of the X-ray film is constructed, and the neural network is compared according to the label sample set.
  • the model is trained to obtain the X-ray film classification model; finally the target X-ray film to be classified is obtained, all the label information in the target X-ray film is extracted, and the label information is input into the X-ray film classification model for classification and recognition , Get the classification result.
  • the application classifies the X-ray film based on the degree of association between each label information, it makes full use of the associated information between each label information to improve the accuracy of classification.
  • construction module 33 is specifically configured as:
  • the X-ray film classification model is obtained.
  • construction module 33 is further configured to:
  • the bilinear pooling processing result is subjected to convolution processing to obtain the target training result.
  • the bilinear pooling process includes the following formula:
  • the MSML loss function is:
  • l and k represent label samples
  • Y i represents a positive sample in the label sample set
  • the classification recognition module 34 is further configured to:
  • the X-ray film classification model with successful verification is generated.
  • the classification recognition module 34 is further configured to:
  • the X-ray film test set is used to test the successfully verified X-ray film classification model, and when the test result is qualified, the tested X-ray film classification model is obtained.
  • a terminal provided by the present application includes a memory and a processor, and computer-readable instructions are stored in the memory.
  • the processor executes any of the above The steps of the X-ray film classification method described.
  • the terminal is a computer device, as shown in FIG. 4.
  • the computer equipment described in this embodiment may be equipment such as servers, personal computers, and network equipment.
  • the computer equipment includes a processor 402, a memory 403, an input unit 404, a display unit 405 and other devices.
  • the memory 403 may be used to store a computer program 401 and various functional modules, and the processor 402 runs the computer program 401 stored in the memory 403 to execute various functional applications and data processing of the device.
  • the memory may be internal memory or external memory, or include both internal memory and external memory.
  • the internal memory may include read only memory (ROM), programmable ROM (PROM), electrically programmable ROM (EPROM), electrically erasable programmable ROM (EEPROM), flash memory, or random access memory.
  • ROM read only memory
  • PROM programmable ROM
  • EPROM electrically programmable ROM
  • EEPROM electrically erasable programmable ROM
  • flash memory or random access memory.
  • External storage can include hard disks, floppy disks, ZIP disks, U disks, tapes, etc.
  • the memory disclosed in this application includes but is not limited to these types of memory.
  • the memory disclosed in this application is only an example and not a limitation.
  • the input unit 404 is used for receiving input of signals and receiving keywords input by the user.
  • the input unit 404 may include a touch panel and other input devices.
  • the touch panel can collect the user's touch operations on or near it (for example, the user uses any suitable objects or accessories such as fingers, stylus, etc., to operate on the touch panel or near the touch panel), and according to preset
  • the program drives the corresponding connection device; other input devices may include, but are not limited to, one or more of a physical keyboard, function keys (such as playback control buttons, switch buttons, etc.), trackball, mouse, and joystick.
  • the display unit 405 can be used to display information input by the user or information provided to the user and various menus of the computer device.
  • the display unit 405 can take the form of a liquid crystal display, an organic light emitting diode, or the like.
  • the processor 402 is the control center of the computer equipment. It uses various interfaces and lines to connect the various parts of the entire computer. It executes by running or executing the software programs and/or modules stored in the memory 402 and calling the data stored in the memory. Various functions and processing data.
  • the computer device includes: one or more processors 402, a memory 403, and one or more computer programs 401, wherein the one or more computer programs 401 are stored in the memory 403 and configured to Executed by the one or more processors 402, the one or more computer programs 401 are configured to execute an X-ray film classification method, wherein the X-ray film classification method includes the following steps: The X-ray film sample set is subjected to noise reduction processing on the X-ray film samples in the X-ray film sample set, and multiple label information is extracted from the key information area of the X-ray film after the noise reduction processing by a positioning method; The correlation degree between the two pieces of label information in the X-ray film, the correlation degree greater than the set threshold and the two label information corresponding to the correlation degree form a label sample, and multiple label samples form a label sample set; construct the X-ray film The neural network model is trained according to the label sample set to obtain the X-ray film classification model; the target X-ray film to be classified
  • this application also proposes a storage medium storing computer-readable instructions.
  • the one or more processors execute an X-ray X-ray film classification method, wherein the X-ray film classification method includes the following steps: obtaining an X-ray film sample set, performing noise reduction processing on X-ray films in the X-ray film sample set, and performing noise reduction processing on After extracting multiple label information from the key information area of the X-ray film; calculate the correlation degree between the two label information in the X-ray film, and calculate the correlation degree greater than the set threshold and the correlation degree corresponding The two pieces of label information form a label sample, and multiple label samples form a label sample set; construct a neural network model of the X-ray film, and train the neural network model according to the label sample set to obtain an X-ray film classification model; obtain For the target X-ray film to be classified, extract all label information in the target X-ray film, and input the label information into the X-
  • the aforementioned storage medium may be a non-volatile storage medium such as a magnetic disk, an optical disc, a read-only memory (Read-Only Memory, ROM), or a random access memory (Random Access Memory, RAM), etc.
  • the X-ray film sample set is first obtained, and the X-ray film in the X-ray film sample set is subjected to noise reduction processing, so as to obtain the X-ray film sample set.
  • the interference information in the optical film is removed as much as possible; and multiple label information is extracted from the key information area of the X-ray film after the noise reduction process by positioning; the difference between the two labels in the X-ray film is calculated
  • the degree of association is greater than the set threshold and the two label information corresponding to the degree of association form a label sample, and a plurality of label samples form a label sample set; then a neural network model of the X-ray film is constructed, according to the label
  • the sample set trains the neural network model to obtain the X-ray film classification model; finally obtains the target X-ray film to be classified, extracts all the label information in the target X-ray film, and inputs the label information into the X-ray film
  • the light sheet classification model performs classification recognition and obtains classification results.
  • the application classifies the X-ray film based on the degree of association between each label information, it makes full use of the associated information between each label information to improve the accuracy of classification.

Landscapes

  • Engineering & Computer Science (AREA)
  • Theoretical Computer Science (AREA)
  • Physics & Mathematics (AREA)
  • General Physics & Mathematics (AREA)
  • Data Mining & Analysis (AREA)
  • Evolutionary Computation (AREA)
  • Life Sciences & Earth Sciences (AREA)
  • Artificial Intelligence (AREA)
  • General Engineering & Computer Science (AREA)
  • Computing Systems (AREA)
  • Bioinformatics & Cheminformatics (AREA)
  • Molecular Biology (AREA)
  • Computational Linguistics (AREA)
  • Biophysics (AREA)
  • Biomedical Technology (AREA)
  • Mathematical Physics (AREA)
  • Software Systems (AREA)
  • Health & Medical Sciences (AREA)
  • General Health & Medical Sciences (AREA)
  • Bioinformatics & Computational Biology (AREA)
  • Computer Vision & Pattern Recognition (AREA)
  • Evolutionary Biology (AREA)
  • Multimedia (AREA)
  • Information Retrieval, Db Structures And Fs Structures Therefor (AREA)
  • Image Analysis (AREA)
  • Analysing Materials By The Use Of Radiation (AREA)

Abstract

一种X光片的分类方法、装置、终端及存储介质。X光片的分类方法包括：获取X光片样本集，从降噪处理后的X光片中提取出多个标签信息（S21）；计算X光片中两两标签信息之间的关联度，并将大于设定阈值的关联度及与关联度对应的两个标签信息形成标签样本，多个标签样本形成标签样本集（S22）；构建X光片的神经网络模型，根据标签样本集对神经网络模型进行训练，获得X光片分类模型（S23）；获取待分类的目标X光片，提取目标X光片中的所有标签信息，将标签信息输入X光片分类模型进行分类识别，获得分类结果（S24）。充分利用了各标签信息之间的关联信息，提高分类的准确性。

Description

X光片的分类方法、装置、终端及存储介质
本申请要求于2019年6月26日提交中国专利局、申请号为201910563462.7，发明名称为“X光片的分类方法、装置、终端及存储介质”的中国专利申请的优先权，其全部内容通过引用结合在本申请中。
技术领域
本申请涉及神经网络技术领域，尤其涉及一种X光片的分类方法、装置、终端及存储介质。
背景技术
X光片是最常用的医疗影像检查工具，一张X光片可以检测出多种疾病，在对涵盖肺炎在内的疾病进行筛查诊断和管理的过程中占据着至关重要的地位。
随着人工智能和机器学习的发展，深度学习技术可以有效解决对X光片进行分类的问题。发明人意识到，现有的X光片的分类方法，通常采用X光片中一个标签信息得到一个目标结果的方式，各标签信息之间相互独立，分类的准确性较低，分类效果较差。
发明内容
本申请提供一种X光片的分类方法、装置、终端及存储介质，以解决当前X光片的分类方法，无法充分利用X光片的各标签信息的关系，分类的准确性较低，分类效果较差的问题。
为解决上述问题，本申请采用如下技术方案：
本申请提供一种X光片的分类方法，包括如下步骤：
获取X光片样本集，对所述X光片样本集中的X光片进行降噪处理，通过定位方式从降噪处理后的所述X光片的关键信息区域中提取出多个标签信息；
计算所述X光片中两两标签信息之间的关联度，将大于设定阈值的关联度及所述关联度对应的两个标签信息形成标签样本，多个标签样本形成标签样本集；
构建X光片的神经网络模型，根据所述标签样本集对所述神经网络模型进行训练，获得X光片分类模型；
获取待分类的目标X光片，提取所述目标X光片中的所有标签信息，将所述标签信息输入所述X光片分类模型进行分类识别，获得分类结果。
本申请提供的一种X光片的分类装置，包括：
获取模块，用于获取X光片样本集，对所述X光片样本集中的X光片进 行降噪处理，通过定位方式从降噪处理后的所述X光片的关键信息区域中提取出多个标签信息；
计算模块，用于计算所述X光片中两两标签信息之间的关联度，将大于设定阈值的关联度及所述关联度对应的两个标签信息形成标签样本，多个标签样本形成标签样本集；
构建模块，用于构建X光片的神经网络模型，根据所述标签样本集对所述神经网络模型进行训练，获得X光片分类模型；
分类识别模块，用于获取待分类的目标X光片，提取所述目标X光片中的所有标签信息，将所述标签信息输入所述X光片分类模型进行分类识别，获得分类结果。
本申请提供一种终端，包括存储器和处理器，所述存储器中存储有计算机可读指令，所述计算机可读指令被所述处理器执行时，使得所述处理器执行一种X光片的分类方法的步骤；
其中，所述X光片的分类方法包括以下步骤：
获取X光片样本集，对所述X光片样本集中的X光片进行降噪处理，通过定位方式从降噪处理后的所述X光片的关键信息区域中提取出多个标签信息；
计算所述X光片中两两标签信息之间的关联度，将大于设定阈值的关联度及所述关联度对应的两个标签信息形成标签样本，多个标签样本形成标签样本集；
构建X光片的神经网络模型，根据所述标签样本集对所述神经网络模型进行训练，获得X光片分类模型；
获取待分类的目标X光片，提取所述目标X光片中的所有标签信息，将所述标签信息输入所述X光片分类模型进行分类识别，获得分类结果。
本申请提供一种存储介质，其上存储有计算机程序，所述计算机程序被处理器执行时，实现一种X光片的分类方法；
其中，所述X光片的分类方法包括以下步骤：
获取X光片样本集，对所述X光片样本集中的X光片进行降噪处理，通过定位方式从降噪处理后的所述X光片的关键信息区域中提取出多个标签信息；
计算所述X光片中两两标签信息之间的关联度，将大于设定阈值的关联度及所述关联度对应的两个标签信息形成标签样本，多个标签样本形成标签样本集；
构建X光片的神经网络模型，根据所述标签样本集对所述神经网络模型进行训练，获得X光片分类模型；
获取待分类的目标X光片，提取所述目标X光片中的所有标签信息，将所述标签信息输入所述X光片分类模型进行分类识别，获得分类结果。
本申请提供的X光片的分类方法，首先通过获取X光片样本集，对所述X光片样本集中的X光片进行降噪处理，以将获取到的X光片中的干扰信息尽量 去除；并通过定位方式从降噪处理后的所述X光片的关键信息区域中提取出多个标签信息；计算所述X光片中两两标签信息之间的关联度，并将大于设定阈值的关联度及所述关联度对应的两个标签信息形成标签样本，多个标签样本形成标签样本集；然后构建X光片的神经网络模型，根据所述标签样本集对所述神经网络模型进行训练，获得X光片分类模型；最后获取待分类的目标X光片，提取所述目标X光片中的所有标签信息，将所述标签信息输入所述X光片分类模型进行分类识别，获得分类结果。本申请根据X光片各标签信息之间的关联度进行分类时，以充分利用各标签信息之间的关联信息，提高分类的准确性。
附图说明
图1为本申请一个实施例中提供的X光片的分类方法的实施环境图；
图2为本申请X光片的分类方法一种实施例流程框图；
图3为本申请X光片的分类装置一种实施例模块框图；
图4为本申请一个实施例中终端的内部结构框图。
具体实施方式
为了使本技术领域的人员更好地理解本申请方案，下面将结合本申请实施例中的附图，对本申请实施例中的技术方案进行清楚、完整地描述。
在本申请的说明书和权利要求书及上述附图中的描述的一些流程中，包含了按照特定顺序出现的多个操作，但是应该清楚了解，这些操作可以不按照其在本文中出现的顺序来执行或并行执行，操作的序号如S11、S12等，仅仅是用于区分开各个不同的操作，序号本身不代表任何的执行顺序。另外，这些流程可以包括更多或更少的操作，并且这些操作可以按顺序执行或并行执行。需要说明的是，本文中的“第一”、“第二”等描述，是用于区分不同的消息、设备、模块等，不代表先后顺序，也不限定“第一”和“第二”是不同的类型。
本领域普通技术人员可以理解，除非特意声明，这里使用的单数形式“一”、“一个”、“所述”和“该”也可包括复数形式。应该进一步理解的是，本申请的说明书中使用的措辞“包括”是指存在所述特征、整数、步骤、操作、元件和/或组件，但是并不排除存在或添加一个或多个其他特征、整数、步骤、操作、元件、组件和/或它们的组。应该理解，当我们称元件被“连接”或“耦接”到另一元件时，它可以直接连接或耦接到其他元件，或者也可以存在中间元件。此外，这里使用的“连接”或“耦接”可以包括无线连接或无线耦接。这里使用的措辞“和/或”包括一个或更多个相关联的列出项的全部或任一单元和全部组合。
本领域普通技术人员可以理解，除非另外定义，这里使用的所有术语(包括技术术语和科学术语)，具有与本申请所属领域中的普通技术人员的一般理解相同的意义。还应该理解的是，诸如通用字典中定义的那些术语，应该被 理解为具有与现有技术的上下文中的意义一致的意义，并且除非像这里一样被特定定义，否则不会用理想化或过于正式的含义来解释。
下面将结合本申请实施例中的附图，对本申请实施例中的技术方案进行清楚、完整地描述，其中自始至终相同或类似的标号表示相同或类似的元件或具有相同或类似功能的元件。显然，所描述的实施例仅仅是本申请一部分实施例，而不是全部的实施例。基于本申请中的实施例，本领域技术人员在没有作出创造性劳动前提下所获得的所有其他实施例，都属于本申请保护的范围。
图1为一个实施例中提供的X光片的分类方法的实施环境图，如图1所示，在该实施环境中，包括服务器110、终端120。终端120通过网络与服务器连接，所述终端120上安装有客户端或浏览器，用户可通过客户端或浏览器上传目标X光片至服务器110，经服务器110分类处理后，将得到的分类结果返回给终端120。其中，上述网络可以包括因特网、2G/3G/4G、wifi等。
需要说明的是，服务器110可以是独立的物理服务器或终端，也可以是多个物理服务器构成的服务器集群，可以是提供云服务器、云数据库、云存储和CDN等基础云计算服务的云服务器。
终端120可以是智能手机、平板电脑、笔记本电脑、台式计算机、智能手表等，但并不局限于此。
请参阅图2，本申请所提供的一种X光片的分类方法，以解决当前X光片的分类方法，无法充分利用X光片的各标签信息的关系，分类的准确性较低，分类效果较差的问题。其中一种实施方式中，包括如下步骤：
S21、获取X光片样本集，对所述X光片样本集中的X光片进行降噪处理，通过定位方式从降噪处理后的所述X光片的关键信息区域中提取出多个标签信息；
在本实施例中，可通过网络下载图片集，从图片集中筛选出若干张X光片作为X光片样本集，然后对所述X光片样本集中的X光片进行降噪处理，以将获取到的X光片中的干扰信息尽量去除，提高后续对X光片中的标签信息识别的准确率。并通过定位方式分别从降噪处理后的每张X光片的关键信息区域中提取出多个标签信息。其中，所述标签信息包括人体器官信息及人体器官信息反映的健康生理信息或病理信息。此外，X光片样本集还可通过其他方式获取，以胸部X光片为例，可将NIH(National Institutes of Health，美国国立卫生研究院)提供的胸部X光片数据集ChestX-ray14作为胸部X光片样本集。
在一实施例中，对X光片进行降噪处理时，可利用近邻取样插值算法对所述X光片进行自动缩放处理；利用图像去污算法对自动缩放处理后的X光片进行去除污点和划痕处理；然后利用角度检测算法和图像快速旋转算法进行角度处理和反光处理。其中，所述角度检测算法是通过数学计算在图像上发现角度特征的一种算法，而且其具有旋转不变性的特质。
通过上述降噪处理，以将获取到的X光片中的干扰信息尽量去除，提高 后续对X光片中的标签信息识别的准确率。
在一实施例中，通过定位方式从降噪处理后的所述X光片的关键信息区域中提取出多个标签信息时，可通过对X光片中反映人体组织的轮廓区域进行定位，并在轮廓区域上设置若干位置坐标，然后以该位置坐标为中心提取出预定范围内的标签信息，从而得到X光片中关键信息区域的标签信息。
提取标签信息时，还可预先对待提取的标签信息进行分类，当需要提取某一类型的标签信息时，先确定该标签信息的类型，然后获取该标签信息类型对应的位置坐标，提取出以所述位置坐标为中心的预定范围内的标签信息，以得到所需类型的标签信息，从而减少了无关信息的提取，以提高后续对神经网络模型训练的效率及准确性。
S22、计算所述X光片中两两标签信息之间的关联度，将大于设定阈值的关联度及所述关联度对应的两个标签信息形成标签样本，多个标签样本形成标签样本集；
在本实施例中，所述标签信息可包括文字信息，计算X光片中两两标签信息之间的关联度时，可根据文字信息中关键词的历史搭配频率计算，如先天性心脏病对应的标签信息包括：房间隔缺损、室间隔缺损、动脉导管未闭等等；心脏瓣膜病对应的标签信息则包括：二尖瓣狭窄和关闭不全、主动脉瓣狭窄和关闭不全等等，计算标签信息之间的关联度时，则房间隔缺损与室间隔缺损的关联度较高，而房间隔缺损与二尖瓣狭窄的关联度比较低，依此类推，得到每两个标签信息之间的关联度，然后筛选出数值大于设定阈值的关联度及所述关联度对应的两个标签信息，以形成标签样本，并将多个标签样本形成标签样本集，即该标签样本集中包括多个标签信息及每两个标签信息之间的关联度。
S23、构建X光片的神经网络模型，根据所述标签样本集对所述神经网络模型进行训练，获得X光片分类模型；
本实施例可构建卷积神经网络模型，将标签样本集输入所述卷积神经网络模型中进行训练，直至卷积神经网络模型收敛时，得到训练合格的X光片分类模型。在本实施例中，当标签样本集中标签信息的数量越多，则训练得到的X光片分类模型的分类效果越好，因此，在对卷积神经网络模型进行训练前，尽可能获取更多的标签信息。
S24、获取待分类的目标X光片，提取所述目标X光片中的所有标签信息，将所述标签信息输入所述X光片分类模型进行分类识别，获得分类结果。
本实施例从待分类的目标X光片中提取出所有标签信息，并将所述标签信息输入训练得到的X光片分类模型中进行分类识别，X光片分类模型通过计算目标X光片中每两个标签信息之间的关联度，并根据关联度及标签信息对目标X光片进行分类，得到分类结果。
为了更好地理解本技术方案，下面以感冒为例进行说明：
当标签信息为喉咙发炎时，则对应的分类结果可能是上火也可能是感冒，在进行分类时，还需结合其他标签信息，将与喉咙发炎关联度比较高的多个 标签信息，形成标签样本集，将标签集合输入分类模型进行分类，得到分类结果，如除了喉咙发炎这一标签信息时，还伴有流鼻涕等标签信息，则可以得到最终分类结果为感冒，而非上火，从而提高分类准确率。
本申请提供的X光片的分类方法，首先通过获取X光片样本集，对所述X光片样本集中的X光片进行降噪处理，以将获取到的X光片中的干扰信息尽量去除；并通过定位方式从降噪处理后的所述X光片的关键信息区域中提取出多个标签信息；计算所述X光片中两两标签信息之间的关联度，并将大于设定阈值的关联度及所述关联度对应的两个标签信息形成标签样本，多个标签样本形成标签样本集；然后构建X光片的神经网络模型，根据所述标签样本集对所述神经网络模型进行训练，获得X光片分类模型；最后获取待分类的目标X光片，提取所述目标X光片中的所有标签信息，将所述标签信息输入所述X光片分类模型进行分类识别，获得分类结果。本申请根据X光片各标签信息之间的关联度进行分类时，以充分利用各标签信息之间的关联信息，提高分类的准确性。
在一实施例中，所述构建X光片的神经网络模型，根据所述标签样本集对所述神经网络模型进行训练，获得X光片分类模型的步骤，可具体包括：
构建第一分类器和第二分类器；其中，所述第二分类器采用MSML损失函数；
将所述标签样本集的标签信息输入第一分类器中，得到第一分类结果；
将第一分类结果及标签样本集的关联度输入第二分类器中，得到目标训练结果；
当目标训练结果满足预设要求时，得到X光片分类模型。
在本实施例中，分类器是数据挖掘中对样本进行分类的方法的统称，包含决策树、逻辑回归、朴素贝叶斯、神经网络等算法。损失函数(loss function)是用来估量你模型的预测值f(x)与真实值Y的不一致程度，它是一个非负实值函数,通常使用L(Y,f(x))来表示，损失函数的结果越小，神经网络模型的鲁棒性就越好。因此，可通过损失函数判断X光片分类模型的分类效果。MSML损失函数为边界样本挖掘损失函数，可把正负样本的边界推开，其同时兼顾相对距离和绝对距离并引入了难样本采样思想的度量学习方法。
本实施例的神经网络模型构建了两个分类器：第一分类器和第二分类器。对神经网络模型进行训练时，先将标签样本集的标签信息输入第一分类器中，得到第一分类结果，再对第一分类结果进一步分类，结合每两个标签信息之间的关联度，将第一分类结果及标签样本集的关联度输入第二分类器中，得到满足预设要求的目标分类结果，进而得到X光片分类模型。其中，第二分类器使用了MSML损失函数，可用来学习不同标签之间的关系，以提高分类的准确性。
在一实施例中，所述第一分类器采用第一分类函数，所述第一分类损失函数为：
Figure PCTCN2019103655-appb-000001
其中，
Figure PCTCN2019103655-appb-000002
表示第i个输入X光片的特征属于c类的预测概率；
Figure PCTCN2019103655-appb-000003
表示第i个输入X光片的特征属于c类的标签信息。
在一实施例中，所述MSML损失函数为：
Figure PCTCN2019103655-appb-000004
其中，l和k表示标签样本，Y i表示标签样本集中的正样本，
Figure PCTCN2019103655-appb-000005
表示标签样本集的负样本，|Y i|为归一化的基数,
Figure PCTCN2019103655-appb-000006
为标签样本集的特征。
在一实施例中，所述将第一分类结果及标签样本集的关联度输入第二分类器中，得到目标训练结果的步骤，可具体包括：
利用标签样本集的关联度对所述第一分类结果进行调整，得到第二分类结果；
将第一分类结果和第二分类器分类得到的第二分类结果进行双线性池化处理，得到双线性池化处理结果；
将双线性池化处理结果进行卷积处理，得到目标训练结果。
第一分类器分类后，只是对标签信息进行简单的分类，各标签信息相互独立，并未考虑标签信息之间的关联信息。因此，本实施例利用第二分类器结合标签样本集的关联度对第一分类结果进行调整，得到第二分类结果，以考虑各独立的标签信息之间的关联信息，对第一分类器分类后的第一分类结果进行校准，提高分类的准确性。
其中，双线性池化处理可以把两个分类器分类得到的特征汇合起来，汇合的过程可以得到两个分类器的特征图成对的相关关系，使两个分类器的特征结合起来使用，进而充分考虑标签信息之间的关系，这样使用更丰富的特征可以得到更好的分类效果。
在一实施例中，所述当目标训练结果满足预设要求时，得到X光片分类模型的步骤，可具体包括：
利用细粒度交叉熵损失函数计算经过训练后的X光片的神经网络模型的总损失；
判断所述总损失是否低于预设值；
当所述总损失低于预设值时，得到X光片分类模型。
本实施例可将两个分类器的分类结果经过双线性池化操作和卷积处理后，然后使用细粒度交叉熵损失函数计算X光片的神经网络模型的总损失，并判断所述总损失是否低于预设值；若是，则得到X光片分类模型；否则调整两个分类器的相关参数，继续对两个分类器进行训练，直至X光片的神经网络模型训练合格。
在一实施例中，所述总损失的公式如下：
E total＝α(E MSML+E CE)+βE FCE；
其中，所述E total表示总损失；所述E CE表示第一损失函数；所述E MSML表示MSML损失函数；所述E FCE表示细粒度交叉熵损失函数。
本实施例计算X光片分类模型的总损失时，结合了第一分类器及第二分类器的损失计算结果，以得到训练效果更佳的X光片分类模型。
在一实施例中，所述双线性池化处理的步骤中，可包括如下公式：
Figure PCTCN2019103655-appb-000007
其中，
Figure PCTCN2019103655-appb-000008
是第一分类器中池化层的局部特征，
Figure PCTCN2019103655-appb-000009
是第二分类器中池化层的局部特征，P i,j是对局部特征进行向量化的结果，Vec是向量化操作处理。
在本实施例中，为了产生二阶局部特征的统计差异，一个池化层的输出结合了两个分类器的结果，从而提高X光片分类模型的分类准确性。
在一实施例中，所述获得X光片分类模型之后，还可包括：
获取X光片验证集；
将所述X光片验证集输入所述X光片分类模型中，得到验证结果；
当验证结果满足预设条件时，生成验证成功的X光片分类模型。
在本实施例中，我们可使用由NIH提供的胸部X光片数据集ChestX-ray14。它总共包含112120个正面胸部X光片，共包含14种病。每个图像大小为1024×1024。我们可将整个数据集分成3个部分。第一部分含有数据集中70％的X光片，作为训练集；10％的X光片用于验证；最后20％的X光片用于测试。对原始数据进行三个集合的划分，是为了能够选出效果最好的、泛化能力最佳的X光片分类模型。
其中，训练集的作用是用来拟合X光片分类模型，通过设置X光片分类中分类器的参数，训练X光片分类模型。后续结合验证集作用时，会选出同一参数的不同取值，拟合出分类器。
所述验证集是用来当通过训练集训练出多个分类模型后，为了能找出效果最佳的X光片分类模型，使用各个模型对验证集数据进行预测，并记录模型准确率。选出效果最佳的模型所对应的参数，即用来调整模型参数。当然，测试集、训练集和验证集三者之间的比例，还可根据实际分类效果，进行设定。
在一实施例中，所述生成验证成功的X光片分类模型之后，还包括：
获取X光片测试集；
利用所述X光片测试集对所述验证成功的X光片分类模型进行测试，当测试结果合格时，得到测试合格的X光片分类模型。
在本实施例中，所述测试集是用来最终评估模式识别系统的性能和分类能力。即可以把测试集当做从来不存在的数据集，当已经确定模型参数后，可以使用测试集进行模型预测并评估模型的性能。
请参考图3，本申请的实施例还提供一种X光片的分类装置，一种本实施例中，包括获取模块31、计算模块32、构建模块33及分类识别模块34。其中，
获取模块31，用于获取X光片样本集，对所述X光片样本集中的X光片进行降噪处理，通过定位方式从降噪处理后的所述X光片的关键信息区域中提取出多个标签信息；
在本实施例中，可通过网络下载图片集，从图片集中筛选出若干张X光片作为X光片样本集，然后对所述X光片样本集中的X光片进行降噪处理，以将获取到的X光片中的干扰信息尽量去除，提高后续对X光片中的标签信息识别的准确率。并通过定位方式分别从降噪处理后的每张X光片的关键信息区域中提取出多个标签信息。其中，所述标签信息包括人体器官信息及人体器官信息反映的健康生理信息或病理信息。此外，X光片样本集还可通过其他方式获取，以胸部X光片为例，可将NIH(National Institutes of Health，美国国立卫生研究院)提供的胸部X光片数据集ChestX-ray14作为胸部X光片样本集。
在一实施例中，对X光片进行降噪处理时，可利用近邻取样插值算法对所述X光片进行自动缩放处理；利用图像去污算法对自动缩放处理后的X光片进行去除污点和划痕处理；然后利用角度检测算法和图像快速旋转算法进行角度处理和反光处理。其中，所述角度检测算法是通过数学计算在图像上发现角度特征的一种算法，而且其具有旋转不变性的特质。
通过上述降噪处理，以将获取到的X光片中的干扰信息尽量去除，提高后续对X光片中的标签信息识别的准确率。
在一实施例中，通过定位方式从降噪处理后的所述X光片的关键信息区域中提取出多个标签信息时，可通过对X光片中反映人体组织的轮廓区域进行定位，并在轮廓区域上设置若干位置坐标，然后以该位置坐标为中心提取出预定范围内的标签信息，从而得到X光片中关键信息区域的标签信息。
提取标签信息时，还可预先对待提取的标签信息进行分类，当需要提取某一类型的标签信息时，先确定该标签信息的类型，然后获取该标签信息类型对应的位置坐标提取出以所述位置坐标为中心的预定范围内的标签信息，以得到所需类型的标签信息，从而减少了无关信息的提取，以提高后续对神经网络模型训练的效率及准确性。
计算模块32，用于计算所述X光片中两两标签信息之间的关联度，将大于设定阈值的关联度及所述关联度对应的两个标签信息形成标签样本，多个标签样本形成标签样本集；
在本实施例中，所述标签信息可包括文字信息，计算X光片中两两标签信息之间的关联度时，可根据文字信息中关键词的历史搭配频率计算，如先天性心脏病对应的标签信息包括：房间隔缺损、室间隔缺损、动脉导管未闭等等；心脏瓣膜病对应的标签信息则包括：二尖瓣狭窄和关闭不全、主动脉瓣狭窄和关闭不全等等，计算标签信息之间的关联度时，则房间隔缺损与室间隔缺损的关联度较高，而房间隔缺损与二尖瓣狭窄的关联度比较低，依此类推，得到每两个标签信息之间的关联度，然后筛选出数值大于设定阈值的关联度及所述关联度对应的两个标签信息，以形成标签样本，并将多个标签样本形成标签样本集，即该标签样本集中包括多个标签信息及每两个标签信息之间的关联度。
构建模块33，用于构建X光片的神经网络模型，根据所述标签样本集对 所述神经网络模型进行训练，获得X光片分类模型；
本实施例可构建卷积神经网络模型，将标签样本集输入所述卷积神经网络模型中进行训练，直至卷积神经网络模型收敛时，得到训练合格的X光片分类模型。在本实施例中，当标签样本集中标签信息的数量越多，则训练得到的X光片分类模型的分类效果越好，因此，在对卷积神经网络模型进行训练前，尽可能获取更多的标签信息。
分类识别模块34，用于获取待分类的目标X光片，提取所述目标X光片中的所有标签信息，将所述标签信息输入所述X光片分类模型进行分类识别，获得分类结果。
本实施例从待分类的目标X光片中提取出所有标签信息，并将所述标签信息输入训练得到的X光片分类模型中进行分类识别，X光片分类模型通过计算目标X光片中每两个标签信息之间的关联度，并根据关联度及标签信息对目标X光片进行分类，得到分类结果。
为了更好地理解本技术方案，下面以感冒为例进行说明：
当标签信息为喉咙发炎时，则对应的分类结果可能是上火也可能是感冒，在进行分类时，还需结合其他标签信息，将与喉咙发炎关联度比较高的信息标签集合，形成标签集合，将标签集合输入分类模型进行分类，得到分类结果，如除了喉咙发炎这一标签信息时，还伴有流鼻涕等标签信息，则可以得到最终分类结果为感冒，而非上火，从而提高分类准确率。
本申请提供的X光片的分类装置，首先通过获取X光片样本集，对所述X光片样本集中的X光片进行降噪处理，以将获取到的X光片中的干扰信息尽量去除；并通过定位方式从降噪处理后的所述X光片的关键信息区域中提取出多个标签信息；计算所述X光片中两两标签信息之间的关联度，并将大于设定阈值的关联度及所述关联度对应的两个标签信息形成标签样本，多个标签样本形成标签样本集；然后构建X光片的神经网络模型，根据所述标签样本集对所述神经网络模型进行训练，获得X光片分类模型；最后获取待分类的目标X光片，提取所述目标X光片中的所有标签信息，将所述标签信息输入所述X光片分类模型进行分类识别，获得分类结果。本申请根据X光片各标签信息之间的关联度进行分类时，以充分利用各标签信息之间的关联信息，提高分类的准确性。
在一实施例中，所述构建模块33具体被配置为：
构建第一分类器和第二分类器；其中，所述第二分类器采用MSML损失函数；
将所述标签样本集的标签信息输入第一分类器中，得到第一分类结果；
将第一分类结果及标签样本集的关联度输入第二分类器中，得到目标训练结果；
当目标训练结果满足预设要求时，得到X光片分类模型。
在一实施例中，所述构建模块33还被配置为：
利用标签样本集的关联度对所述第一分类结果进行调整，得到第二分类 结果；
将第一分类结果和第二分类器分类得到的第二分类结果进行双线性池化处理，得到双线性池化处理结果；
将双线性池化处理结果进行卷积处理，得到目标训练结果。
在一实施例中，所述双线性池化处理包括如下公式：
Figure PCTCN2019103655-appb-000010
其中，
Figure PCTCN2019103655-appb-000011
是第一分类器中池化层的局部特征，
Figure PCTCN2019103655-appb-000012
是第二分类器中池化层的局部特征，P i,j是对局部特征进行向量化的结果，Vec是向量化操作处理。
在一实施例中，所述MSML损失函数为：
Figure PCTCN2019103655-appb-000013
其中，l和k表示标签样本，Y i表示标签样本集中的正样本，
Figure PCTCN2019103655-appb-000014
表示标签样本集的负样本，|Y i|为归一化的基数,
Figure PCTCN2019103655-appb-000015
为标签样本集的特征。
在一实施例中，所述分类识别模块34还被配置为：
获取X光片验证集；
将所述X光片验证集输入所述X光片分类模型中，得到验证结果；
当验证结果满足预设条件时，生成验证成功的X光片分类模型。
在一实施例中，所述分类识别模块34还被配置为：
获取X光片测试集；
利用所述X光片测试集对所述验证成功的X光片分类模型进行测试，当测试结果合格时，得到测试合格的X光片分类模型。
关于上述实施例中的装置，其中各个模块执行操作的具体方式已经在有关该方法的实施例中进行了详细描述，此处将不做详细阐述说明。
本申请提供的一种终端，包括存储器和处理器，所述存储器中存储有计算机可读指令，所述计算机可读指令被所述处理器执行时，使得所述处理器执行如上任一项所述的X光片的分类方法的步骤。
在一实施例中，所述终端为一种计算机设备，如图4所示。本实施例所述的计算机设备可以是服务器、个人计算机以及网络设备等设备。所述计算机设备包括处理器402、存储器403、输入单元404以及显示单元405等器件。本领域技术人员可以理解，图4示出的设备结构器件并不构成对所有设备的限定，可以包括比图示更多或更少的部件，或者组合某些部件。存储器403可用于存储计算机程序401以及各功能模块，处理器402运行存储在存储器403的计算机程序401，从而执行设备的各种功能应用以及数据处理。存储器可以是内存储器或外存储器，或者包括内存储器和外存储器两者。内存储器可以包括只读存储器(ROM)、可编程ROM(PROM)、电可编程ROM(EPROM)、电可擦写可编程ROM(EEPROM)、快闪存储器、或者随机存储器。外存储器可以包括硬盘、软盘、ZIP盘、U盘、磁带等。本申请所公开的存储器包括但不限于这些类型的存储器。本申请所公开的存储器只作为例子而非作为限定。
输入单元404用于接收信号的输入，以及接收用户输入的关键字。输入单元404可包括触控面板以及其它输入设备。触控面板可收集用户在其上或附近的触摸操作(比如用户使用手指、触笔等任何适合的物体或附件在触控面板上或在触控面板附近的操作)，并根据预先设定的程序驱动相应的连接装置；其它输入设备可以包括但不限于物理键盘、功能键(比如播放控制按键、开关按键等)、轨迹球、鼠标、操作杆等中的一种或多种。显示单元405可用于显示用户输入的信息或提供给用户的信息以及计算机设备的各种菜单。显示单元405可采用液晶显示器、有机发光二极管等形式。处理器402是计算机设备的控制中心，利用各种接口和线路连接整个电脑的各个部分，通过运行或执行存储在存储器402内的软件程序和/或模块，以及调用存储在存储器内的数据，执行各种功能和处理数据。
作为一个实施例，所述计算机设备包括：一个或多个处理器402，存储器403，一个或多个计算机程序401，其中所述一个或多个计算机程序401被存储在存储器403中并被配置为由所述一个或多个处理器402执行，所述一个或多个计算机程序401配置用于执行一种X光片的分类方法其中，所述X光片的分类方法包括以下步骤：获取X光片样本集，对所述X光片样本集中的X光片进行降噪处理，通过定位方式从降噪处理后的所述X光片的关键信息区域中提取出多个标签信息；计算所述X光片中两两标签信息之间的关联度，将大于设定阈值的关联度及所述关联度对应的两个标签信息形成标签样本，多个标签样本形成标签样本集；构建X光片的神经网络模型，根据所述标签样本集对所述神经网络模型进行训练，获得X光片分类模型；获取待分类的目标X光片，提取所述目标X光片中的所有标签信息，将所述标签信息输入所述X光片分类模型进行分类识别，获得分类结果。
在一个实施例中，本申请还提出了一种存储有计算机可读指令的存储介质，该计算机可读指令被一个或多个处理器执行时，使得一个或多个处理器执行一种X光片的分类方法，其中，所述X光片的分类方法包括以下步骤：获取X光片样本集，对所述X光片样本集中的X光片进行降噪处理，通过定位方式从降噪处理后的所述X光片的关键信息区域中提取出多个标签信息；计算所述X光片中两两标签信息之间的关联度，将大于设定阈值的关联度及所述关联度对应的两个标签信息形成标签样本，多个标签样本形成标签样本集；构建X光片的神经网络模型，根据所述标签样本集对所述神经网络模型进行训练，获得X光片分类模型；获取待分类的目标X光片，提取所述目标X光片中的所有标签信息，将所述标签信息输入所述X光片分类模型进行分类识别，获得分类结果。例如，所述存储介质可以是ROM、随机存取存储器(RAM)、CD-ROM、磁带、软盘和光数据存储设备等。
本领域普通技术人员可以理解实现上述实施例方法中的全部或部分流程，是可以通过计算机程序来指令相关的硬件来完成，该计算机程序可存储于一存储介质中，该程序在执行时，可包括如上述各方法的实施例的流程。其中，前述的存储介质可为磁碟、光盘、只读存储记忆体(Read-Only Memory，ROM) 等非易失性存储介质，或随机存储记忆体(Random Access Memory，RAM)等。
综合上述实施例可知，本申请最大的有益效果在于：
本申请提供的X光片的分类方法、装置、终端及存储介质，首先通过获取X光片样本集，对所述X光片样本集中的X光片进行降噪处理，以将获取到的X光片中的干扰信息尽量去除；并通过定位方式从降噪处理后的所述X光片的关键信息区域中提取出多个标签信息；计算所述X光片中两两标签信息之间的关联度，并将大于设定阈值的关联度及所述关联度对应的两个标签信息形成标签样本，多个标签样本形成标签样本集；然后构建X光片的神经网络模型，根据所述标签样本集对所述神经网络模型进行训练，获得X光片分类模型；最后获取待分类的目标X光片，提取所述目标X光片中的所有标签信息，将所述标签信息输入所述X光片分类模型进行分类识别，获得分类结果。本申请根据X光片各标签信息之间的关联度进行分类时，以充分利用各标签信息之间的关联信息，提高分类的准确性。
以上所述实施例的各技术特征可以进行任意的组合，为使描述简洁，未对上述实施例中的各个技术特征所有可能的组合都进行描述，然而，只要这些技术特征的组合不存在矛盾，都应当认为是本说明书记载的范围。
以上所述实施例仅表达了本申请的几种实施方式，其描述较为具体和详细，但并不能因此而理解为对本申请专利范围的限制。应当指出的是，对于本领域的普通技术人员来说，在不脱离本申请构思的前提下，还可以做出若干变形和改进，这些都属于本申请的保护范围。因此，本申请专利的保护范围应以所附权利要求为准。

Claims (20)

  1. 一种X光片的分类方法，包括：
    获取X光片样本集，对所述X光片样本集中的X光片进行降噪处理，通过定位方式从降噪处理后的所述X光片的关键信息区域中提取出多个标签信息；
    计算所述X光片中两两标签信息之间的关联度，将大于设定阈值的关联度及所述关联度对应的两个标签信息形成标签样本，多个标签样本形成标签样本集；
    构建X光片的神经网络模型，根据所述标签样本集对所述神经网络模型进行训练，获得X光片分类模型；
    获取待分类的目标X光片，提取所述目标X光片中的所有标签信息，将所述标签信息输入所述X光片分类模型进行分类识别，获得分类结果。
  2. 根据权利要求1所述的X光片的分类方法，所述构建X光片的神经网络模型，根据所述标签样本集对所述神经网络模型进行训练，获得X光片分类模型的步骤，包括：
    构建第一分类器和第二分类器；其中，所述第二分类器采用MSML损失函数；
    将所述标签样本集的标签信息输入第一分类器中，得到第一分类结果；
    将第一分类结果及标签样本集的关联度输入第二分类器中，得到目标训练结果；
    当目标训练结果满足预设要求时，得到X光片分类模型。
  3. 根据权利要求2所述的X光片的分类方法，所述将第一分类结果及标签样本集的关联度输入第二分类器中，得到目标训练结果的步骤，包括：
    利用标签样本集的关联度对所述第一分类结果进行调整，得到第二分类结果；
    将第一分类结果和第二分类器分类得到的第二分类结果进行双线性池化处理，得到双线性池化处理结果；
    将双线性池化处理结果进行卷积处理，得到目标训练结果。
  4. 根据权利要求3所述的X光片的分类方法，所述双线性池化处理的步骤中，包括如下公式：
    Figure PCTCN2019103655-appb-100001
    其中，
    Figure PCTCN2019103655-appb-100002
    是第一分类器中池化层的局部特征，
    Figure PCTCN2019103655-appb-100003
    是第二分类器中池化层的局部特征，P i,j是对局部特征进行向量化的结果，Vec是向量化操作处理。
  5. 根据权利要求2所述的X光片的分类方法，所述MSML损失函数为：
    Figure PCTCN2019103655-appb-100004
    其中，l和k表示标签样本，Y i表示标签样本集中的正样本，
    Figure PCTCN2019103655-appb-100005
    表示标签样 本集的负样本，|Y i|为归一化的基数,
    Figure PCTCN2019103655-appb-100006
    为标签样本集的特征。
  6. 根据权利要求1所述的X光片的分类方法，所述获得X光片分类模型之后，还包括：
    获取X光片验证集；
    将所述X光片验证集输入所述X光片分类模型中，得到验证结果；
    当验证结果满足预设条件时，生成验证成功的X光片分类模型。
  7. 根据权利要求6所述的X光片的分类方法，所述生成验证成功的X光片分类模型之后，还包括：
    获取X光片测试集；
    利用所述X光片测试集对所述验证成功的X光片分类模型进行测试，当测试结果合格时，得到测试合格的X光片分类模型。
  8. 一种X光片的分类装置，包括：
    获取模块，用于获取X光片样本集，对所述X光片样本集中的X光片进行降噪处理，通过定位方式从降噪处理后的所述X光片的关键信息区域中提取出多个标签信息；
    计算模块，用于计算所述X光片中两两标签信息之间的关联度，将大于设定阈值的关联度及所述关联度对应的两个标签信息形成标签样本，多个标签样本形成标签样本集；
    构建模块，用于构建X光片的神经网络模型，根据所述标签样本集对所述神经网络模型进行训练，获得X光片分类模型；
    分类识别模块，用于获取待分类的目标X光片，提取所述目标X光片中的所有标签信息，将所述标签信息输入所述X光片分类模型进行分类识别，获得分类结果。
  9. 一种终端，包括存储器和处理器，所述存储器中存储有计算机可读指令，所述计算机可读指令被所述处理器执行时，使得所述处理器执行一种X光片的分类方法的步骤；
    其中，所述X光片的分类方法包括以下步骤：
    获取X光片样本集，对所述X光片样本集中的X光片进行降噪处理，通过定位方式从降噪处理后的所述X光片的关键信息区域中提取出多个标签信息；
    计算所述X光片中两两标签信息之间的关联度，将大于设定阈值的关联度及所述关联度对应的两个标签信息形成标签样本，多个标签样本形成标签样本集；
    构建X光片的神经网络模型，根据所述标签样本集对所述神经网络模型进行训练，获得X光片分类模型；
    获取待分类的目标X光片，提取所述目标X光片中的所有标签信息，将所述标签信息输入所述X光片分类模型进行分类识别，获得分类结果。
  10. 根据权利要求9所述的终端，所述构建X光片的神经网络模型，根据所述标签样本集对所述神经网络模型进行训练，获得X光片分类模型的步骤， 包括：
    构建第一分类器和第二分类器；其中，所述第二分类器采用MSML损失函数；
    将所述标签样本集的标签信息输入第一分类器中，得到第一分类结果；
    将第一分类结果及标签样本集的关联度输入第二分类器中，得到目标训练结果；
    当目标训练结果满足预设要求时，得到X光片分类模型。
  11. 根据权利要求10所述的终端，所述将第一分类结果及标签样本集的关联度输入第二分类器中，得到目标训练结果的步骤，包括：
    利用标签样本集的关联度对所述第一分类结果进行调整，得到第二分类结果；
    将第一分类结果和第二分类器分类得到的第二分类结果进行双线性池化处理，得到双线性池化处理结果；
    将双线性池化处理结果进行卷积处理，得到目标训练结果。
  12. 根据权利要求11所述的终端，所述双线性池化处理的步骤中，包括如下公式：
    Figure PCTCN2019103655-appb-100007
    其中，
    Figure PCTCN2019103655-appb-100008
    是第一分类器中池化层的局部特征，
    Figure PCTCN2019103655-appb-100009
    是第二分类器中池化层的局部特征，P i,j是对局部特征进行向量化的结果，Vec是向量化操作处理。
  13. 根据权利要求10所述的终端，所述MSML损失函数为：
    Figure PCTCN2019103655-appb-100010
    其中，l和k表示标签样本，Y i表示标签样本集中的正样本，
    Figure PCTCN2019103655-appb-100011
    表示标签样本集的负样本，|Y i|为归一化的基数,
    Figure PCTCN2019103655-appb-100012
    为标签样本集的特征。
  14. 根据权利要求9所述的终端，所述获得X光片分类模型之后，还包括：
    获取X光片验证集；
    将所述X光片验证集输入所述X光片分类模型中，得到验证结果；
    当验证结果满足预设条件时，生成验证成功的X光片分类模型。
  15. 根据权利要求14所述的X终端，所述生成验证成功的X光片分类模型之后，还包括：
    获取X光片测试集；
    利用所述X光片测试集对所述验证成功的X光片分类模型进行测试，当测试结果合格时，得到测试合格的X光片分类模型。
  16. 一种非易失性存储介质，其上存储有计算机程序，该计算机程序被处理器执行时，实现一种X光片的分类方法；
    其中，所述X光片的分类方法包括以下步骤：
    获取X光片样本集，对所述X光片样本集中的X光片进行降噪处理，通过定位方式从降噪处理后的所述X光片的关键信息区域中提取出多个标签信 息；
    计算所述X光片中两两标签信息之间的关联度，将大于设定阈值的关联度及所述关联度对应的两个标签信息形成标签样本，多个标签样本形成标签样本集；
    构建X光片的神经网络模型，根据所述标签样本集对所述神经网络模型进行训练，获得X光片分类模型；
    获取待分类的目标X光片，提取所述目标X光片中的所有标签信息，将所述标签信息输入所述X光片分类模型进行分类识别，获得分类结果。
  17. 根据权利要求16所述的非易失性存储介质，所述构建X光片的神经网络模型，根据所述标签样本集对所述神经网络模型进行训练，获得X光片分类模型的步骤，包括：
    构建第一分类器和第二分类器；其中，所述第二分类器采用MSML损失函数；
    将所述标签样本集的标签信息输入第一分类器中，得到第一分类结果；
    将第一分类结果及标签样本集的关联度输入第二分类器中，得到目标训练结果；
    当目标训练结果满足预设要求时，得到X光片分类模型。
  18. 根据权利要求17所述的非易失性存储介质，所述将第一分类结果及标签样本集的关联度输入第二分类器中，得到目标训练结果的步骤，包括：
    利用标签样本集的关联度对所述第一分类结果进行调整，得到第二分类结果；
    将第一分类结果和第二分类器分类得到的第二分类结果进行双线性池化处理，得到双线性池化处理结果；
    将双线性池化处理结果进行卷积处理，得到目标训练结果。
  19. 根据权利要求18所述的非易失性存储介质，所述双线性池化处理的步骤中，包括如下公式：
    Figure PCTCN2019103655-appb-100013
    其中，
    Figure PCTCN2019103655-appb-100014
    是第一分类器中池化层的局部特征，
    Figure PCTCN2019103655-appb-100015
    是第二分类器中池化层的局部特征，P i,j是对局部特征进行向量化的结果，Vec是向量化操作处理。
  20. 根据权利要求17所述的非易失性存储介质，所述MSML损失函数为：
    Figure PCTCN2019103655-appb-100016
    其中，l和k表示标签样本，Y i表示标签样本集中的正样本，
    Figure PCTCN2019103655-appb-100017
    表示标签样本集的负样本，|Y i|为归一化的基数,
    Figure PCTCN2019103655-appb-100018
    为标签样本集的特征。
PCT/CN2019/103655 2019-06-26 2019-08-30 X光片的分类方法、装置、终端及存储介质 WO2020258507A1 (zh)

Applications Claiming Priority (2)

Application Number Priority Date Filing Date Title
CN201910563462.7A CN110414562B (zh) 2019-06-26 2019-06-26 X光片的分类方法、装置、终端及存储介质
CN201910563462.7 2019-06-26

Publications (1)

Publication Number Publication Date
WO2020258507A1 true WO2020258507A1 (zh) 2020-12-30

Family

ID=68359749

Family Applications (1)

Application Number Title Priority Date Filing Date
PCT/CN2019/103655 WO2020258507A1 (zh) 2019-06-26 2019-08-30 X光片的分类方法、装置、终端及存储介质

Country Status (2)

Country Link
CN (1) CN110414562B (zh)
WO (1) WO2020258507A1 (zh)

Cited By (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN113505655A (zh) * 2021-06-17 2021-10-15 电子科技大学 面向数字孪生系统的轴承故障智能诊断方法
CN115797341A (zh) * 2023-01-16 2023-03-14 四川大学 一种自动即时判定头颅侧位x光片自然头位的方法

Citations (6)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN104715259A (zh) * 2015-01-22 2015-06-17 苏州工业职业技术学院 X光乳腺图像的核自适应优化分类方法
CN105574859A (zh) * 2015-12-14 2016-05-11 中国科学院深圳先进技术研究院 一种基于ct图像的肝脏肿瘤分割方法及装置
CN107871316A (zh) * 2017-10-19 2018-04-03 浙江工业大学 一种基于深度神经网络的x光片手骨兴趣区域自动提取方法
CN108830282A (zh) * 2018-05-29 2018-11-16 电子科技大学 一种乳腺x光图像的乳腺肿块信息提取及分类方法
WO2019051356A1 (en) * 2017-09-08 2019-03-14 The General Hospital Corporation SYSTEM AND METHOD FOR AUTOMATICALLY LABELING AND ANNOUNTING NON-STRUCTURED MEDICAL DATA SETS
CN109493328A (zh) * 2018-08-31 2019-03-19 上海联影智能医疗科技有限公司 医学图像显示方法、查看设备以及计算机设备

Family Cites Families (6)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN106844407B (zh) * 2015-12-07 2020-03-10 华为技术有限公司 基于数据集相关性的标签网络产生方法和系统
US10107769B2 (en) * 2016-01-11 2018-10-23 Carl Zeiss X-Ray Microscopy Inc. Multimodality mineralogy segmentation system and method
CN106960219B (zh) * 2017-03-10 2021-04-16 百度在线网络技术（北京）有限公司 图片识别方法及装置、计算机设备及计算机可读介质
CN107145898B (zh) * 2017-04-14 2020-04-10 北京航星机器制造有限公司 一种基于神经网络的射线图像分类方法
CN109300121B (zh) * 2018-09-13 2019-11-01 华南理工大学 一种心血管疾病诊断模型的构建方法、系统及该诊断装置
CN109685102B (zh) * 2018-11-13 2024-07-09 平安科技（深圳）有限公司 胸部病灶图像分类方法、装置、计算机设备及存储介质

Patent Citations (6)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN104715259A (zh) * 2015-01-22 2015-06-17 苏州工业职业技术学院 X光乳腺图像的核自适应优化分类方法
CN105574859A (zh) * 2015-12-14 2016-05-11 中国科学院深圳先进技术研究院 一种基于ct图像的肝脏肿瘤分割方法及装置
WO2019051356A1 (en) * 2017-09-08 2019-03-14 The General Hospital Corporation SYSTEM AND METHOD FOR AUTOMATICALLY LABELING AND ANNOUNTING NON-STRUCTURED MEDICAL DATA SETS
CN107871316A (zh) * 2017-10-19 2018-04-03 浙江工业大学 一种基于深度神经网络的x光片手骨兴趣区域自动提取方法
CN108830282A (zh) * 2018-05-29 2018-11-16 电子科技大学 一种乳腺x光图像的乳腺肿块信息提取及分类方法
CN109493328A (zh) * 2018-08-31 2019-03-19 上海联影智能医疗科技有限公司 医学图像显示方法、查看设备以及计算机设备

Cited By (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN113505655A (zh) * 2021-06-17 2021-10-15 电子科技大学 面向数字孪生系统的轴承故障智能诊断方法
CN113505655B (zh) * 2021-06-17 2023-10-13 电子科技大学 面向数字孪生系统的轴承故障智能诊断方法
CN115797341A (zh) * 2023-01-16 2023-03-14 四川大学 一种自动即时判定头颅侧位x光片自然头位的方法
CN115797341B (zh) * 2023-01-16 2023-04-14 四川大学 一种自动即时判定头颅侧位x光片自然头位的方法

Also Published As

Publication number Publication date
CN110414562B (zh) 2023-11-24
CN110414562A (zh) 2019-11-05

Similar Documents

Publication Publication Date Title
CN111126574B (zh) 基于内镜图像对机器学习模型进行训练的方法、装置和存储介质
Abdullah et al. Lung cancer prediction and classification based on correlation selection method using machine learning techniques
US20220036135A1 (en) Method and apparatus for determining image to be labeled and model training method and apparatus
Zhang et al. Dynamic hypergraph structure learning.
CN112580416A (zh) 基于深暹罗网络和贝叶斯优化的视频跟踪
US20120310864A1 (en) Adaptive Batch Mode Active Learning for Evolving a Classifier
US20220004906A1 (en) Learning and applying contextual similiarities between entities
CN112784778B (zh) 生成模型并识别年龄和性别的方法、装置、设备和介质
WO2022105118A1 (zh) 基于图像的健康状态识别方法、装置、设备及存储介质
Hua et al. Collaborative active visual recognition from crowds: A distributed ensemble approach
Luo et al. Orthogonally constrained matrix factorization for robust unsupervised feature selection with local preserving
Seiter et al. Discovery of activity composites using topic models: An analysis of unsupervised methods
Wang et al. Multi-branching temporal convolutional network for sepsis prediction
WO2021056765A1 (zh) 图像处理方法及相关装置
Akilandasowmya et al. Skin cancer diagnosis: Leveraging deep hidden features and ensemble classifiers for early detection and classification
Yue et al. Attention-driven cascaded network for diabetic retinopathy grading from fundus images
WO2020258507A1 (zh) X光片的分类方法、装置、终端及存储介质
Ibanez et al. Deep learning for pulmonary nodule ct image retrieval—an online assistance system for novice radiologists
JP2024020506A (ja) 視覚認識階層ベースの物体認識のための方法及び装置
Suganya et al. Prognosticating various acute covid lung disorders from COVID-19 patient using chest CT Images
Ullah et al. Severity classification of COVID-19 patients data using quantum machine learning approaches
Haghpanah et al. Real-time hand rubbing quality estimation using deep learning enhanced by separation index and feature-based confidence metric
US20240028831A1 (en) Apparatus and a method for detecting associations among datasets of different types
Zeng et al. Re-KISSME: A robust resampling scheme for distance metric learning in the presence of label noise
Yasmin et al. Impact of fuzziness for skin lesion classification with transformer-based model

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: 19935018

Country of ref document: EP

Kind code of ref document: A1

NENP Non-entry into the national phase

Ref country code: DE

122 Ep: pct application non-entry in european phase

Ref document number: 19935018

Country of ref document: EP

Kind code of ref document: A1