WO2021051497A1

WO2021051497A1 - Pulmonary tuberculosis determination method and apparatus, computer device, and storage medium

Info

Publication number: WO2021051497A1
Application number: PCT/CN2019/115946
Authority: WO
Inventors: 任嘉祥; 王健宗
Original assignee: 平安科技（深圳）有限公司
Priority date: 2019-09-16
Filing date: 2019-11-06
Publication date: 2021-03-25
Also published as: CN110738235B; CN110738235A

Abstract

A pulmonary tuberculosis determination method and apparatus, a computer device, and a storage medium, relating to the technical field of artificial intelligence. The pulmonary tuberculosis determination method comprises: acquiring a chest X-ray image to be classified; converting, according to pre-configured image processing steps, the chest X-ray image into a target image to be classified, the resolution and the dimension of the target image are the same as those of a target image sample for training a pulmonary tuberculosis classification model; and inputting the target image into the pulmonary tuberculosis classification model to acquire a predicted probability, the predicted probability being the probability acquired by means of prediction that a test for pulmonary tuberculosis is positive, and when the predicted probability is greater than a pre-configured threshold, determining that pulmonary tuberculosis exists in the chest X-ray image corresponding to the target image. The pulmonary tuberculosis determination method ensures high accuracy, and achieves rapid determination of pulmonary tuberculosis.

Description

Tuberculosis determination method, device, computer equipment and storage medium

This application is based on the Chinese invention patent application filed on September 16, 2019 with the application number 201910869773.6 and titled "Method, Apparatus, Computer Equipment and Storage Medium for Tuberculosis Determination", and claims its priority.

【Technical Field】

This application relates to the field of artificial intelligence technology, and in particular to a method, device, computer equipment and storage medium for determining tuberculosis.

【Background technique】

Tuberculosis is a disease that affects many people and requires accurate diagnosis before it can be treated. At present, hospitals usually have X-ray machines, but some related staff lack radiology expertise to accurately evaluate images, resulting in poor diagnosis; some related staff can manually check X-rays, but the task is time-consuming and screening costs Larger. At present, it is impossible to achieve rapid determination of tuberculosis under the premise of ensuring a high accuracy rate.

[Summary of the invention]

In view of this, the embodiments of the present application provide a tuberculosis determination method, device, computer equipment, and storage medium to solve the current problem that the rapid determination of tuberculosis cannot be achieved under the premise of ensuring a high accuracy rate.

In the first aspect, an embodiment of the present application provides a method for determining tuberculosis, including:

Acquire X-ray images of the chest to be classified;

Converting the chest X-ray image to be classified into a target image to be classified according to a preset image processing step, wherein the resolution and dimension of the target image to be classified are the same as the target image sample for training the tuberculosis classification model;

The target image to be classified is input into the tuberculosis classification model to obtain a prediction probability. The prediction probability is the probability that the tuberculosis is predicted to be positive. When the prediction probability is greater than a preset threshold, it is determined to be the same as the target to be classified. There is tuberculosis in the chest X-ray image to be classified corresponding to the image, wherein the model training steps adopted by the tuberculosis classification model include:

Construct a training sample set, wherein the training sample set includes a target image sample for model training and a target image sample for model testing, the target image sample for model training and a target image sample for model testing There is no same target image sample between them;

Adopting the ResNet-50 network as the training deep neural network, and using the pre-trained weights as the initial weights of the ResNet-50 network;

Input the target image sample used for model training into the ResNet-50 network for training, and the output result is the predicted probability of pulmonary tuberculosis positive;

Update the weight of the ResNet-50 network according to the predicted probability, and stop training until the updated change amount is less than the first preset threshold to obtain the model to be tested;

Use the target image sample for model testing to test the model to be tested, and when the accuracy rate of the test result output by the model to be tested is greater than the preset accuracy rate, use the model to be tested as the tuberculosis classification model .

In the second aspect, an embodiment of the present application provides an apparatus for determining tuberculosis, including:

The first acquisition module is used to acquire a chest X-ray image to be classified;

The second acquisition module is configured to convert the chest X-ray image to be classified into a target image to be classified according to preset image processing steps, wherein the resolution and dimension of the target image to be classified are the same as the training tuberculosis classification model The target image samples are the same;

The judging module is used to input the target image to be classified into the tuberculosis classification model to obtain the predicted probability. The predicted probability is the probability that the predicted tuberculosis is positive. When the predicted probability is greater than a preset threshold, the Tuberculosis is present in the chest X-ray image to be classified corresponding to the target image to be classified, wherein the tuberculosis classification model is obtained through a construction module, an initialization module, a training module, an update module, and a third acquisition module:

The construction module is used to construct a training sample set, wherein the training sample set includes a target image sample used for model training and a target image sample used for model testing, and the target image sample used for model training is used for model training. There is no identical target image sample among the tested target image samples;

The initialization module is used to use the ResNet-50 network as a deep neural network for training, and use the weights obtained by pre-training as the initial weights of the ResNet-50 network;

A training module, configured to input the target image samples used for model training into the ResNet-50 network for training, and the output result is the predicted probability of pulmonary tuberculosis positive;

An update module, configured to update the weight of the ResNet-50 network according to the predicted probability, and stop training until the updated change amount is less than a first preset threshold to obtain the model to be tested;

The third acquisition module is configured to use the target image sample for model testing to test the model to be tested, and when the accuracy of the test result output by the model to be tested is greater than the preset accuracy, the The model is used as the pulmonary tuberculosis classification model.

In a third aspect, a computer device includes a memory, a processor, and computer-readable instructions that are stored in the memory and can run on the processor, and the processor implements the foregoing when the computer-readable instructions are executed. The steps of the tuberculosis determination method.

In a fourth aspect, an embodiment of the present application provides a computer non-volatile readable storage medium, including: computer readable instructions, which when executed by a processor, implement the steps of the above method for determining tuberculosis.

In the embodiment of the present application, the tuberculosis classification model is used to realize the tuberculosis determination of the chest X-ray image to be classified. The tuberculosis classification model uses the ResNet-50 network as the deep neural network for training, so that the trained tuberculosis classification model has strong feature extraction capabilities and high classification accuracy; in addition, the training tuberculosis classification model also uses migration learning Method, the weights obtained by pre-training are used as the initial weights of the ResNet-50 network, which can speed up model training and improve the accuracy of model classification. The embodiment of the application uses the tuberculosis classification model trained for tuberculosis determination. After inputting the target image to be classified converted from the chest X-ray image to be classified, the tuberculosis determination can be achieved according to the predicted probability output by the tuberculosis classification model. , Under the premise of ensuring high accuracy, the rapid determination of tuberculosis can be realized.

【Explanation of the drawings】

In order to explain the technical solutions of the embodiments of the present application more clearly, the following will briefly introduce the drawings needed in the embodiments. Obviously, the drawings in the following description are only some embodiments of the present application. For those of ordinary skill in the art, without creative labor, other drawings can be obtained from these drawings.

Fig. 1 is a flowchart of a method for determining tuberculosis in an embodiment of the present application;

Figure 2 is a schematic diagram of a tuberculosis determination device in an embodiment of the present application;

Fig. 3 is a schematic diagram of a computer device in an embodiment of the present application.

【detailed description】

In order to better understand the technical solutions of the present application, the embodiments of the present application will be described in detail below with reference to the accompanying drawings.

It should be clear that the described embodiments are only a part of the embodiments of the present application, rather than all of the embodiments. Based on the embodiments in this application, all other embodiments obtained by those of ordinary skill in the art without creative work shall fall within the protection scope of this application.

The terms used in the embodiments of the present application are only for the purpose of describing specific embodiments, and are not intended to limit the present application. The singular forms of "a", "said" and "the" used in the embodiments of the present application and the appended claims are also intended to include plural forms, unless the context clearly indicates other meanings.

It should be understood that the term "and/or" used herein is only a description of the same field of the associated object, indicating that there can be three relationships. For example, A and/or B can mean that A exists alone and A exists at the same time. And B, there are three cases of B alone. In addition, the character "/" in this text generally indicates that the associated objects before and after are in an "or" relationship.

It should be understood that, although the terms first, second, third, etc. may be used in the embodiments of the present application to describe the preset range, etc., these preset ranges should not be limited to these terms. These terms are only used to distinguish the preset ranges from each other. For example, without departing from the scope of the embodiments of the present application, the first preset range may also be referred to as the second preset range, and similarly, the second preset range may also be referred to as the first preset range.

Depending on the context, the word "if" as used herein can be interpreted as "when" or "when" or "in response to determination" or "in response to detection". Similarly, depending on the context, the phrase "if determined" or "if detected (statement or event)" can be interpreted as "when determined" or "in response to determination" or "when detected (statement or event) )" or "in response to detection (statement or event)".

Fig. 1 shows a flow chart of the method for determining tuberculosis in this embodiment. The tuberculosis determination method can be applied to the tuberculosis determination system, and the tuberculosis determination system can be used to determine the tuberculosis of the chest X-ray image. The pulmonary tuberculosis determination system can be specifically applied to a computer device, where the computer device is a device that can perform human-computer interaction with a user, including but not limited to devices such as computers, smart phones, and tablets. As shown in Figure 1, the method for determining tuberculosis includes the following steps:

S1: Obtain a chest X-ray image to be classified.

Understandably, before the tuberculosis determination is performed, the chest X-ray image that the user wants to determine for tuberculosis is the chest X-ray image to be classified.

S2: Convert the chest X-ray image to be classified into the target image to be classified according to the preset image processing steps, where the resolution and dimension of the target image to be classified are the same as the target image sample for training the tuberculosis classification model.

Understandably, the directly obtained chest X-ray image to be classified needs to be converted into the target image to be classified before it can be input into the tuberculosis determination model for determination, so that its resolution and dimension are the same as the target image for training the tuberculosis classification model. The samples are the same, thereby improving the accuracy of tuberculosis determination.

Understandably, the resolution and dimension of the target image samples used for training the tuberculosis classification model are different from those of chest X-ray images. After preset image processing, the ability to extract features of the tuberculosis determination model can be improved, thereby improving the accuracy of tuberculosis determination. rate. The preset image processing is the processing of the image samples of the chest X-ray film when constructing the training sample set in the following steps S11-S15.

S3: Input the target image to be classified into the tuberculosis classification model to obtain the predicted probability. The predicted probability is the probability that the predicted tuberculosis is positive. When the predicted probability is greater than the preset threshold, determine the chest X-ray to be classified corresponding to the target image to be classified The film image has tuberculosis.

Understandably, according to the predicted probability, the chest X-ray image to be classified can be classified according to whether there is tuberculosis, and the determination of tuberculosis is also completed during the classification.

Among them, the preset threshold can be specifically set to 0.5, and the accuracy of tuberculosis determination under the preset threshold is relatively high.

Among them, the model training steps adopted by the tuberculosis classification model include:

S10: Construct a training sample set, where the training sample set includes a target image sample used for model training and a target image sample used for model testing, between the target image sample used for model training and the target image sample used for model testing There are no identical target image samples.

Among them, the absence of the same target image sample between the target image sample used for model training and the target image sample used for model testing can improve the generalization ability of the tuberculosis classification model, and can deal with more tuberculosis judgments in different scenarios.

Further, the step of constructing the training sample set specifically includes:

S11: Obtain the image sample of the chest X-ray film and the label of the image sample. When the image sample is positive for tuberculosis, the label is 1, and when the image sample is negative for tuberculosis, the label is 0;

S12: Process the image sample into an image sample with a preset resolution, where, for an image sample with a resolution higher than the preset resolution, the down-sampling method is used to down-sample the resolution of the image sample to the preset resolution. For image samples whose resolution is lower than the preset resolution, use bilinear interpolation to upsample the resolution of the image samples to the preset resolution;

Among them, the preset resolution may be 512*512. The tuberculosis classification model trained at this resolution has a faster calculation speed and a higher classification accuracy.

It is understandable that the resolution of the directly acquired image samples of the chest X-ray film may be too high or too low. The image samples may be processed to a resolution that is beneficial for model training to ensure the accuracy of the model.

S13: Normalize the value of each pixel of the image sample of the preset resolution to the interval [-1, 1];

Understandably, normalizing the pixel value can compress the sample space and improve the computational efficiency.

Specifically, when the number of pixel colors is 256, the normalized expression is specifically (x-127.5)/127.5.

S14: Copy the normalized image sample, expand the dimension of the image sample, and obtain the target image sample;

In one embodiment, if the preset resolution is 512*512, the image sample is being copied, and the target image sample obtained by expanding the dimension of the image sample will be expressed as 512*512*N, where N is the number of times of copying. Copying the image sample and expanding the dimension of the image sample can increase the number of input samples, help the model to fully train, and improve the accuracy of tuberculosis determination.

S15: Use the target image samples to construct a training sample set, where the ratio of the target image samples used for model training and the target image samples used for model testing in the training sample set is 5:1.

Among them, the model training and model configuration can be completed well under this ratio, which is a better ratio.

In steps S11-S15, a specific implementation manner for constructing a training sample set is provided, which can effectively process the image samples of the original chest X-ray film, so that the classification effect of the tuberculosis classification model is more accurate.

S20: Use the ResNet-50 network as the trained deep neural network, and use the pre-trained weights as the initial weights of the ResNet-50 network.

Among them, the ResNet-50 network contains a total of 49 convolutional layers, a standardized layer and a fully connected layer.

The classification effect of the ResNet-50 network is better. In this embodiment, the ResNet-50 network is used as the original model for training the tuberculosis classification model, and the migration learning method is adopted, and the weights obtained by pre-training are used as the initial weights of the ResNet-50 network. The weight value obtained by the pre-training may specifically be the initial weight value used by the developer when dealing with other projects. Among them, the content of the project or the principle of function realization are more relevant to the tuberculosis classification, the better.

In one embodiment, by using the ResNet-50 network as the training deep neural network, and using the pre-trained weights as the initial weights of the ResNet-50 network, the speed of model training can be accelerated, and the accuracy of model classification can be improved. rate.

S30: Input the target image sample used for model training into the ResNet-50 network for training, and the output result is the predicted probability of pulmonary tuberculosis positive.

Specifically, the input dimension of the ResNet-50 network can be specifically set to 512x512x3.

In an embodiment, the target image samples used for model training are input into the ResNet-50 network for training, and a 256x256x64 feature map is obtained after a 7x7x64 convolutional layer and a 3x3 maximum pooling layer with a step size of 2; Then after 4 groups of residual modules, the output changes to 128x128x256, 64x64x512, 32x32x1024, 16x16x2048 feature maps (wherein, for feature maps with different dimensions, first use a 1x1 convolutional layer to adjust the dimensions of the input features to match the desired The dimension of the feature map is added, and then the elements of the corresponding position are added), and finally through the standardized layer and the fully connected layer of dimension 1, the output result is the predicted probability of pulmonary tuberculosis positive.

S40: Update the weight of the ResNet-50 network according to the predicted probability, and stop training until the updated change amount is less than the first preset threshold to obtain the model to be tested.

Further, the step of updating the weight of the ResNet-50 network according to the predicted probability specifically includes:

S41: Use the cross-entropy loss function to calculate the loss value generated during the training process, where the cross-entropy loss function is expressed as:

Represents the label of the target image sample used for training, and y represents the predicted probability;

S42: Use a backpropagation algorithm to return the loss value generated during the training process to the ResNet-50 network, and update the weight of the ResNet-50 network according to the loss value returned during each training.

In steps S41-S42, a specific implementation manner for updating the weights of the ResNet-50 network according to the predicted probability is provided, and the network parameters can be updated under supervised learning.

S50: Use the target image sample for model testing to test the model to be tested, and when the accuracy rate of the test result output by the model to be tested is greater than the preset accuracy rate, the model to be tested is used as a tuberculosis classification model.

Further, the ResNet-50 network includes a convolutional layer, a standardization layer, and a fully connected layer. The tuberculosis classification model first performs a preset number of passes on the convolutional layer in the ResNet-50 network when updating the weights of the ResNet-50 network. Update, after the preset number of training passes, freeze the weights of the convolutional layer in the ResNet-50 network, and use a learning rate of 0.001 to train the standardized layer and the fully connected layer in the ResNet-50 network for 1000 times. The weights of the standardized layer and the fully connected layer are updated, where freezing means that the weights of the convolutional layer in the ResNet-50 network are not updated.

Understandably, the weights of the convolutional layer contain key features for distinguishing the target image samples, but the features reflected in the pre-training weights come from the sample training of other items and cannot be used to distinguish tuberculosis samples. Here, transfer learning is used to first train the convolutional layer to reflect the characteristics of judging tuberculosis disease; then freeze the convolutional layer, train the standardized layer and the fully connected layer, which can further improve the feature extraction ability of the model, and then improve the model The accuracy rate.

Further, when the tuberculosis classification model updates the weights of the ResNet-50 network, it first updates the convolutional layer in the ResNet-50 network within a preset number of passes. The ResNet-50 network includes a convolutional layer, and the preset number of passes is specific It can be 3000 times. When the ResNet-50 network updates the weights of the convolutional layer, the training process is to train the ResNet-50 network 3000 times with a learning rate of 0.0001, where each pass includes 10 target image samples for training.

Using the specific parameters mentioned in the above model training and parameter update process can improve the feature extraction ability of the tuberculosis classification model and the accuracy of the tuberculosis classification model.

It should be understood that the size of the sequence number of each step in the foregoing embodiment does not mean the order of execution. The execution sequence of each process should be determined by its function and internal logic, and should not constitute any limitation on the implementation process of the embodiment of the present application.

Based on the pulmonary tuberculosis determination method provided in the embodiments, the embodiments of the present application further provide device embodiments that implement the steps and methods in the foregoing method embodiments.

Fig. 2 shows a principle block diagram of a tuberculosis determination device corresponding to the tuberculosis determination method in the embodiment one-to-one. As shown in FIG. 2, the tuberculosis determination device includes a first acquisition module 10, a second acquisition module 20, a determination module 30, a construction module 40, an initialization module 50, a training module 60, an update module 70 and a third acquisition module 80. Among them, the implementation functions of the first acquisition module 10, the second acquisition module 20, the determination module 30, the construction module 40, the initialization module 50, the training module 60, the update module 70, and the third acquisition module 80 correspond to the tuberculosis determination method in the embodiment The steps of are one-to-one correspondence, in order to avoid repetition, this embodiment will not describe them one by one.

The first acquisition module 10 is used to acquire a chest X-ray image to be classified.

The second acquisition module 20 is used to convert the chest X-ray image to be classified into a target image to be classified according to the preset image processing steps, wherein the resolution and dimension of the target image to be classified are the same as the target image for training the tuberculosis classification model The samples are the same.

The determination module 30 is used to input the target image to be classified into the tuberculosis classification model to obtain the predicted probability. The predicted probability is the probability that the tuberculosis is predicted to be positive. When the predicted probability is greater than a preset threshold, determine the target image to be classified. The classification of chest X-ray images has tuberculosis, and the tuberculosis classification model is obtained through the construction module, the initialization module, the training module, the update module, and the third acquisition module:

The construction module 40 is used to construct a training sample set, where the training sample set includes a target image sample used for model training and a target image sample used for model testing, a target image sample used for model training and a target used for model testing The same target image sample does not exist between the image samples.

The initialization module 50 is used to use the ResNet-50 network as a training deep neural network, and use the weights obtained by pre-training as the initial weights of the ResNet-50 network.

The training module 60 is used to input the target image samples used for model training into the ResNet-50 network for training, and the output result is the predicted probability of pulmonary tuberculosis positive.

The update module 70 is configured to update the weight of the ResNet-50 network according to the predicted probability, and stop training until the updated change amount is less than the first preset threshold to obtain the model to be tested.

The third acquisition module 80 is used to test the model to be tested using the target image sample for model testing, and when the accuracy of the test result output by the model to be tested is greater than the preset accuracy, the model to be tested is used as a tuberculosis classification model.

Optionally, the building module 40 is specifically used for:

Obtain the image sample of the chest X-ray film and the label of the image sample. When the image sample is positive for tuberculosis, the label is 1, and when the image sample is negative for tuberculosis, the label is 0;

The image samples are processed into image samples with a preset resolution. For image samples with a higher resolution than the preset resolution, the down-sampling method is used to down-sample the resolution of the image samples to the preset resolution. For image samples lower than the preset resolution, use bilinear interpolation to upsample the resolution of the image samples to the preset resolution;

Normalize the value of each pixel of the image sample of the preset resolution to the interval of [-1, 1];

Copy the normalized image sample, expand the dimension of the image sample, and obtain the target image sample;

The target image samples are used to construct a training sample set, where the ratio of the target image samples used for model training and the target image samples used for model testing in the training sample set is 5:1.

Optionally, the ResNet-50 network includes a convolutional layer. When the ResNet-50 network updates the weights of the convolutional layer, the training process is to use a learning rate of 0.0001 to train 3000 times the ResNet-50 network, where each training pass includes 10 A sample of the target image used for training.

Optionally, the ResNet-50 network includes a convolutional layer, a standardization layer, and a fully connected layer. The tuberculosis classification model first performs a preset number of passes to the convolutional layer in the ResNet-50 network when updating the weights of the ResNet-50 network After updating, the preset number of passes is trained, the weights of the convolutional layer in the ResNet-50 network are frozen, and the learning rate of 0.001 is used to train the standardized layer and the fully connected layer in the ResNet-50 network 1000 times. For the ResNet-50 network The weights of the standardized layer and the fully connected layer are updated, where freezing means that the weights of the convolutional layer in the ResNet-50 network are not updated.

Optionally, the update module 70 is specifically used for:

The cross entropy loss function is used to calculate the loss value generated during the training process, where the cross entropy loss function is expressed as:

The backpropagation algorithm is used to return the loss value generated during the training process to the ResNet-50 network, and the weight of the ResNet-50 network is updated according to the loss value returned during each training.

This embodiment provides a computer non-volatile readable storage medium, the computer non-volatile readable storage medium stores computer readable instructions, and when the computer readable instructions are executed by a processor, the method for determining tuberculosis in the embodiment is implemented To avoid repetition, I won’t repeat them here. Alternatively, the computer-readable instructions realize the functions of the various modules/units in the tuberculosis determination device in the embodiment when being executed by the processor. In order to avoid repetition, details are not repeated here.

Fig. 3 is a schematic diagram of a computer device provided by an embodiment of the present application. As shown in FIG. 3, the computer device 90 of this embodiment includes: a processor 91, a memory 92, and computer-readable instructions 93 stored in the memory 92 and running on the processor 91, and the computer-readable instructions 93 are processed. The method for determining tuberculosis in the embodiment is implemented when the device 91 is executed. In order to avoid repetition, it will not be repeated here. Alternatively, when the computer-readable instruction 93 is executed by the processor 91, the function of each model/unit in the tuberculosis determination device in the embodiment is realized. In order to avoid repetition, it will not be repeated here.

The computer device 90 may be a computing device such as a desktop computer, a notebook, a palmtop computer, and a cloud server. The computer device 90 may include, but is not limited to, a processor 91 and a memory 92. Those skilled in the art can understand that FIG. 3 is only an example of the computer device 90, and does not constitute a limitation on the computer device 90. It may include more or less components than those shown in the figure, or a combination of certain components, or different components. For example, computer equipment may also include input and output devices, network access devices, buses, and so on.

The so-called processor 91 may be a central processing unit (Central Processing Unit, CPU), other general-purpose processors, digital signal processors (Digital Signal Processor, DSP), application specific integrated circuits (Application Specific Integrated Circuit, ASIC), Field-Programmable Gate Array (FPGA) or other programmable logic devices, discrete gates or transistor logic devices, discrete hardware components, etc. The general-purpose processor may be a microprocessor or the processor may also be any conventional processor or the like.

The memory 92 may be an internal storage unit of the computer device 90, such as a hard disk or a memory of the computer device 90. The memory 92 may also be an external storage device of the computer device 90, such as a plug-in hard disk equipped on the computer device 90, a smart media card (SMC), a secure digital (SD) card, and a flash memory card (Flash). Card) and so on. Further, the memory 92 may also include both an internal storage unit of the computer device 90 and an external storage device. The memory 92 is used to store computer readable instructions and other programs and data required by the computer equipment. The memory 92 can also be used to temporarily store data that has been output or will be output.

Those skilled in the art can clearly understand that for the convenience and conciseness of description, only the division of the above functional units and modules is used as an example. In practical applications, the above functions can be allocated to different functional units and modules as required. Module completion, that is, the internal structure of the device is divided into different functional units or modules to complete all or part of the functions described above.

The above embodiments are only used to illustrate the technical solutions of the present application, not to limit them; although the present application has been described in detail with reference to the foregoing embodiments, those of ordinary skill in the art should understand that they can still compare the previous embodiments. The recorded technical solutions are modified, or some of the technical features are equivalently replaced; and these modifications or replacements do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the application, and shall be included in the application Within the scope of protection.

Claims

A method for determining tuberculosis, characterized in that the method includes:

Acquire X-ray images of the chest to be classified;

According to a preset image processing step, converting the chest X-ray image to be classified into a target image to be classified, wherein the resolution and dimension of the target image to be classified are the same as the target image sample for training the tuberculosis classification model;

Input the target image to be classified into the pulmonary tuberculosis classification model to obtain the predicted probability. The predicted probability is the probability that the predicted pulmonary tuberculosis is positive. There is tuberculosis in the chest X-ray image to be classified corresponding to the image, wherein the model training steps adopted by the tuberculosis classification model include:

Construct a training sample set, wherein the training sample set includes a target image sample for model training and a target image sample for model testing, the target image sample for model training and a target image sample for model testing There is no same target image sample between them;

Adopting the ResNet-50 network as the training deep neural network, and using the pre-trained weights as the initial weights of the ResNet-50 network;

Input the target image sample used for model training into the ResNet-50 network for training, and the output result is the predicted probability of pulmonary tuberculosis positive;

Update the weight of the ResNet-50 network according to the predicted probability, and stop training until the updated change amount is less than the first preset threshold to obtain the model to be tested;

Use the target image sample for model testing to test the model to be tested, and when the accuracy rate of the test result output by the model to be tested is greater than the preset accuracy rate, use the model to be tested as the tuberculosis classification model .
The method according to claim 1, wherein said constructing a training sample set comprises:

Acquiring an image sample of a chest X-ray film and a label of the image sample, wherein when the image sample is positive for tuberculosis, the label is 1, and when the image sample is negative for tuberculosis, the label is 0;

The image sample is processed into an image sample with a preset resolution, where, for an image sample with a resolution higher than the preset resolution, a down-sampling method is used to down-sample the resolution of the image sample to the preset Resolution, for an image sample with a resolution lower than the preset resolution, up-sampling the resolution of the image sample to the preset resolution by using a bilinear interpolation method;

Normalize the value of each pixel of the image sample of the preset resolution to the interval [-1, 1];

Copying the normalized image sample to expand the dimension of the image sample to obtain a target image sample;

The target image samples are used to construct the training sample set, wherein the ratio of the target image samples used for model training to the target image samples used for model testing in the training sample set is 5:1.
The method according to claim 1, wherein the ResNet-50 network includes a convolutional layer, and when the ResNet-50 network updates the weights of the convolutional layer, the training process adopts a learning rate of 0.0001 The ResNet-50 network is trained 3000 times, wherein each training pass includes 10 target image samples for training.
The method according to claim 1, wherein the ResNet-50 network includes a convolutional layer, a standardization layer, and a fully connected layer, and the tuberculosis classification model first updates the weights of the ResNet-50 network. The convolutional layer in the ResNet-50 network is updated within the preset number of passes. After the preset number of passes is trained, the weight of the convolutional layer in the ResNet-50 network is frozen, and the learning rate of 0.001 is used to train the The standardized layer and the fully connected layer in the ResNet-50 network are updated 1,000 times, and the weights of the standardized layer and the fully connected layer in the ResNet-50 network are updated, where the freezing means that the ResNet-50 network is not updated. The weight of the middle convolutional layer is updated.
The method according to any one of claims 1 to 4, wherein the updating the weight of the ResNet-50 network according to the predicted probability comprises:

The cross-entropy loss function is used to calculate the loss value generated during the training process, where the cross-entropy loss function is expressed as:
Represents the label of the target image sample used for training, and y represents the predicted probability;

A back propagation algorithm is used to return the loss value generated during the training process to the ResNet-50 network, and the weight value of the ResNet-50 network is updated according to the loss value returned during each training.
A device for determining tuberculosis, characterized in that the device comprises:

The first acquisition module is used to acquire a chest X-ray image to be classified;

The second acquisition module is configured to convert the chest X-ray image to be classified into a target image to be classified according to preset image processing steps, wherein the resolution and dimension of the target image to be classified are the same as the training tuberculosis classification model The target image samples are the same;

The judging module is used to input the target image to be classified into the tuberculosis classification model to obtain the predicted probability. The predicted probability is the probability that the predicted tuberculosis is positive. When the predicted probability is greater than a preset threshold, the Tuberculosis is present in the chest X-ray image to be classified corresponding to the target image to be classified, wherein the tuberculosis classification model is obtained through a construction module, an initialization module, a training module, an update module, and a third acquisition module:

The construction module is used to construct a training sample set, wherein the training sample set includes a target image sample used for model training and a target image sample used for model testing, and the target image sample used for model training is used for model training. There is no identical target image sample among the tested target image samples;

The initialization module is used to use the ResNet-50 network as a deep neural network for training, and use the weights obtained by pre-training as the initial weights of the ResNet-50 network;

A training module, configured to input the target image samples used for model training into the ResNet-50 network for training, and the output result is the predicted probability of pulmonary tuberculosis positive;

An update module, configured to update the weight of the ResNet-50 network according to the predicted probability, and stop training until the updated change amount is less than a first preset threshold to obtain the model to be tested;

The third acquisition module is configured to use the target image sample for model testing to test the model to be tested, and when the accuracy of the test result output by the model to be tested is greater than the preset accuracy, the The model is used as the tuberculosis classification model.
The device according to claim 6, wherein the building module is specifically configured to:

Acquiring an image sample of a chest X-ray film and a label of the image sample, wherein when the image sample is positive for tuberculosis, the label is 1, and when the image sample is negative for tuberculosis, the label is 0;

The image sample is processed into an image sample with a preset resolution, where, for an image sample with a resolution higher than the preset resolution, a down-sampling method is used to down-sample the resolution of the image sample to the preset Resolution, for an image sample with a resolution lower than the preset resolution, up-sampling the resolution of the image sample to the preset resolution by using a bilinear interpolation method;

Normalize the value of each pixel of the image sample of the preset resolution to the interval [-1, 1];

Copying the normalized image sample to expand the dimension of the image sample to obtain a target image sample;

The target image samples are used to construct the training sample set, wherein the ratio of the target image samples used for model training to the target image samples used for model testing in the training sample set is 5:1.
The device according to claim 6, wherein the ResNet-50 network includes a convolutional layer, and when the ResNet-50 network updates the weights of the convolutional layer, the training process adopts a learning rate of 0.0001 The ResNet-50 network is trained 3000 times, wherein each training pass includes 10 target image samples for training.
The device according to claim 6, wherein the ResNet-50 network includes a convolutional layer, a standardization layer, and a fully connected layer, and the tuberculosis classification model is first used when updating the weights of the ResNet-50 network. The convolutional layer in the ResNet-50 network is updated within the preset number of passes. After the preset number of passes is trained, the weight of the convolutional layer in the ResNet-50 network is frozen, and the learning rate of 0.001 is used to train the The standardized layer and the fully connected layer in the ResNet-50 network are updated 1,000 times, and the weights of the standardized layer and the fully connected layer in the ResNet-50 network are updated, where the freezing means that the ResNet-50 network is not updated. The weight of the middle convolutional layer is updated.
The device according to any one of claims 6-9, wherein the update module is specifically configured to:

The cross-entropy loss function is used to calculate the loss value generated during the training process, where the cross-entropy loss function is expressed as:
Represents the label of the target image sample used for training, and y represents the predicted probability;

A back propagation algorithm is used to return the loss value generated during the training process to the ResNet-50 network, and the weight value of the ResNet-50 network is updated according to the loss value returned during each training.
A computer device includes a memory, a processor, and computer-readable instructions stored in the memory and capable of running on the processor, wherein the processor executes the computer-readable instructions as follows step:

Acquire X-ray images of the chest to be classified;

Converting the chest X-ray image to be classified into a target image to be classified according to a preset image processing step, wherein the resolution and dimension of the target image to be classified are the same as the target image sample for training the tuberculosis classification model;

The target image to be classified is input into the tuberculosis classification model to obtain a prediction probability. The prediction probability is the probability that the tuberculosis is predicted to be positive. When the prediction probability is greater than a preset threshold, it is determined to be the same as the target to be classified. There is tuberculosis in the chest X-ray image to be classified corresponding to the image, wherein the model training steps adopted by the tuberculosis classification model include:

Construct a training sample set, wherein the training sample set includes a target image sample for model training and a target image sample for model testing, the target image sample for model training and a target image sample for model testing There is no same target image sample between them;

Adopting the ResNet-50 network as the training deep neural network, and using the pre-trained weights as the initial weights of the ResNet-50 network;

Input the target image sample used for model training into the ResNet-50 network for training, and the output result is the predicted probability of pulmonary tuberculosis positive;

Update the weight of the ResNet-50 network according to the predicted probability, and stop training until the updated change amount is less than the first preset threshold to obtain the model to be tested;

Use the target image sample for model testing to test the model to be tested, and when the accuracy rate of the test result output by the model to be tested is greater than the preset accuracy rate, use the model to be tested as the tuberculosis classification model .
The computer device according to claim 11, wherein when the processor executes the computer-readable instructions to construct a training sample set, it comprises the following steps:

Acquiring an image sample of a chest X-ray film and a label of the image sample, wherein when the image sample is positive for tuberculosis, the label is 1, and when the image sample is negative for tuberculosis, the label is 0;

The image sample is processed into an image sample with a preset resolution, where, for an image sample with a resolution higher than the preset resolution, the resolution of the image sample is downsampled to the preset Resolution, for image samples with a resolution lower than the preset resolution, up-sampling the resolution of the image samples to the preset resolution by using a bilinear interpolation method;

Normalize the value of each pixel of the image sample of the preset resolution to the interval [-1, 1];

Copying the normalized image sample to expand the dimension of the image sample to obtain a target image sample;

The target image samples are used to construct the training sample set, wherein the ratio of the target image samples used for model training to the target image samples used for model testing in the training sample set is 5:1.
The computer device according to claim 11, wherein the ResNet-50 network includes a convolutional layer, and when the ResNet-50 network updates the weights of the convolutional layer, the training process is to use 0.0001 learning The ResNet-50 network is trained 3000 times at a rate, wherein each training pass includes 10 target image samples for training.
The computer device according to claim 11, wherein the ResNet-50 network includes a convolutional layer, a standardization layer, and a fully connected layer, and the tuberculosis classification model first updates the weights of the ResNet-50 network. The convolutional layer in the ResNet-50 network is updated within the preset number of passes. After the preset number of passes is trained, the weights of the convolutional layer in the ResNet-50 network are frozen, and the learning rate of 0.001 is used to train the The standardized layer and the fully connected layer in the ResNet-50 network are 1000 times, and the weights of the standardized layer and the fully connected layer in the ResNet-50 network are updated. The weight of the convolutional layer in the network is updated.
The computer device according to any one of claims 11-14, wherein when the processor executes the computer-readable instructions to update the weight of the ResNet-50 network according to the predicted probability, it includes the following step:

The cross-entropy loss function is used to calculate the loss value generated during the training process, where the cross-entropy loss function is expressed as:
Represents the label of the target image sample used for training, and y represents the predicted probability;

A back propagation algorithm is used to return the loss value generated during the training process to the ResNet-50 network, and the weight value of the ResNet-50 network is updated according to the loss value returned during each training.
A computer non-volatile readable storage medium, the computer non-volatile readable storage medium storing computer readable instructions, wherein the computer readable instructions are executed by a processor to implement the following steps:

Acquire X-ray images of the chest to be classified;

Converting the chest X-ray image to be classified into a target image to be classified according to a preset image processing step, wherein the resolution and dimension of the target image to be classified are the same as the target image sample for training the tuberculosis classification model;

Input the target image to be classified into the pulmonary tuberculosis classification model to obtain the predicted probability. The predicted probability is the probability that the predicted pulmonary tuberculosis is positive. When the predicted probability is greater than a preset threshold, it is determined that the target There is tuberculosis in the chest X-ray image to be classified corresponding to the image, wherein the model training steps adopted by the tuberculosis classification model include:

Construct a training sample set, wherein the training sample set includes a target image sample for model training and a target image sample for model testing, the target image sample for model training and a target image sample for model testing There is no same target image sample between them;

Adopting the ResNet-50 network as the training deep neural network, and using the pre-trained weights as the initial weights of the ResNet-50 network;

Input the target image sample used for model training into the ResNet-50 network for training, and the output result is the predicted probability of pulmonary tuberculosis positive;

Update the weight of the ResNet-50 network according to the predicted probability, and stop training until the updated change amount is less than the first preset threshold to obtain the model to be tested;

Use the target image sample for model testing to test the model to be tested, and when the accuracy rate of the test result output by the model to be tested is greater than the preset accuracy rate, use the model to be tested as the tuberculosis classification model .
The computer non-volatile readable storage medium according to claim 16, wherein when the computer readable instructions are executed by one or more processors to construct a training sample set, the method comprises the following steps:

Acquiring an image sample of a chest X-ray film and a label of the image sample, wherein when the image sample is positive for tuberculosis, the label is 1, and when the image sample is negative for tuberculosis, the label is 0;

The image sample is processed into an image sample with a preset resolution, where, for an image sample with a resolution higher than the preset resolution, a down-sampling method is used to down-sample the resolution of the image sample to the preset Resolution, for an image sample with a resolution lower than the preset resolution, up-sampling the resolution of the image sample to the preset resolution by using a bilinear interpolation method;

Normalize the value of each pixel of the image sample of the preset resolution to the interval [-1, 1];

Copying the normalized image sample to expand the dimension of the image sample to obtain a target image sample;

The target image samples are used to construct the training sample set, wherein the ratio of the target image samples used for model training to the target image samples used for model testing in the training sample set is 5:1.
The computer non-volatile readable storage medium according to claim 16, wherein the ResNet-50 network comprises a convolutional layer, and when the ResNet-50 network updates the weight of the convolutional layer, The training process is to train the ResNet-50 network 3000 times with a learning rate of 0.0001, wherein each training pass includes 10 target image samples for training.
The computer non-volatile readable storage medium according to claim 16, wherein the ResNet-50 network includes a convolutional layer, a standardized layer, and a fully connected layer, and the tuberculosis classification model is updating the ResNet-50 network. When the weight of the 50 network is used, the convolutional layer in the ResNet-50 network is first updated within the preset number of passes. After the preset number of passes is trained, the weight of the convolutional layer in the ResNet-50 network is frozen, Use a learning rate of 0.001 to train the standardized layer and the fully connected layer in the ResNet-50 network 1000 times, and update the weights of the standardized layer and the fully connected layer in the ResNet-50 network, wherein the freezing is It means that the weight of the convolutional layer in the ResNet-50 network is not updated.
The computer non-volatile readable storage medium according to any one of claims 16-19, wherein the computer readable instructions are executed by one or more processors to update the ResNet according to the predicted probability -50 network weight, including the following steps:

The cross-entropy loss function is used to calculate the loss value generated during the training process, where the cross-entropy loss function is expressed as:
Represents the label of the target image sample used for training, and y represents the predicted probability;

A back propagation algorithm is used to return the loss value generated during the training process to the ResNet-50 network, and the weight value of the ResNet-50 network is updated according to the loss value returned during each training.