WO2020238044A1

WO2020238044A1 - Method and device for constructing 3d unet network model for tumor detection, and storage medium

Info

Publication number: WO2020238044A1
Application number: PCT/CN2019/117323
Authority: WO
Inventors: 齐守良; 曾威; 姚育东; 钱唯; 郑斌; 高伟明; 葛新科; 张红治
Original assignee: 深圳市前海安测信息技术有限公司
Priority date: 2019-05-31
Filing date: 2019-11-12
Publication date: 2020-12-03
Also published as: CN112102221A

Abstract

The invention provides a method and device for constructing a 3D UNet network model for tumor detection, and a computer-readable storage medium. The method comprises: acquiring an LIDC data set from a medical image database, wherein the LIDC data set comprises a tumor CT image and a tumor region annotation in an xml format; converting the tumor region annotation in the xml format into a tumor region annotation in a mask format; dividing the converted LIDC data set into a training data set and a verification data set; preprocessing the training data set and the verification data set, and normalizing pixel values of the tumor CT image; constructing a 3D Unet network model on the basis of a keras framework; using the training data set to train the 3D Unet network model so as to acquire a weight of the 3D Unet network model; and using the verification data set to verify the validity of the 3D Unet network model. In the invention, a 3D Unet network is constructed, thereby quickly and accurately detecting a tumor focus region, and improving the efficiency and accuracy of tumor detection.

Description

3D UNet network model construction method, device and storage medium for detecting tumor

Technical field

The present invention relates to the technical field of tumor image processing, in particular to a method, device and computer readable storage medium for constructing a 3D UNet network model for tumor detection.

Background technique

With the popularization of CT applications, it provides convenience for early tumor screening. Statistics in recent years have found that the incidence of lung cancer is getting higher and higher, and it is also the primary cause of cancer mortality. The tumor lesion area is accurately segmented from the CT image, and the plan for preoperative neoadjuvant radiotherapy and chemotherapy, and postoperative Evaluation of the effect of radiotherapy and chemotherapy is of great significance. However, manually delineating the tumor area is a time-consuming and labor-intensive task. In addition, the delineation results of different radiologists on the tumor area are affected by their subjective experience, environment and many other factors, and their delineation results are not repeatable. In addition, due to differences in lung morphology of different individuals, tumor CT images are blurred, and it is difficult to find all tumor regions in all tumor CT images. The accuracy and low efficiency of tumor detection make it difficult to locate the tumor. Therefore, it is urgent to realize the automatic detection of the tumor area in clinical practice, accurately locate the tumor location, and provide guidance for the diagnosis and treatment of the tumor has become the research focus of the industry.

technical problem

The main purpose of the present invention is to provide a method, device and computer readable storage medium for constructing a 3D UNet network model for tumor detection, aiming to solve the problem that the existing tumor detection methods are limited by individual lung morphological differences, resulting in low tumor detection efficiency And the technical problem of low accuracy.

Technical solutions

In order to achieve the above objective, the present invention provides a 3D UNet network model construction method for detecting tumors. The method includes the following steps: obtaining LIDC data sets from a medical imaging database, the LIDC data sets including tumor CT images and tumors in xml format Region annotation; convert the tumor area annotation in xml format into mask tumor area annotation; divide the transformed LIDC data set into training data set and validation data set; preprocess the training data set and the validation data set, and perform CT of the tumor Image pixel value normalization processing; build 3D based on keras framework Unet network model; use the training data set to train the constructed 3D Unet network model to obtain the weight of the 3D Unet network model; use the validation data set to construct the 3D The validity of the Unet network model is verified.

Preferably, the step of preprocessing the training data set and the verification data set includes the following steps: set the interval between the pixels in the training data set and the verification data set to 1, so that the data input to the 3D Unet network model has Uniform interval; for the tumor CT images and mask tumor area annotations in the training data set, the following steps are performed: obtain the center point of the mask tumor area annotation, and use the center point as the center point of the 96×96×32 matrix according to 96×96 ×32 size Randomly crop the tumor area, randomly zoom in and out, randomly rotate the angle, randomly flip up and down, and generate diversified training data; perform the following steps for the tumor CT image and mask tumor area labeling in the validation data set: Get mask The center point of the tumor area, using the center point as the center point of the 96×96×32 matrix, crop the tumor area according to the size of 96×96×32, and save the cropped tumor CT image and mask the tumor area.

Preferably, the step of normalizing the pixel value of the tumor CT image includes the following steps: setting the pixel value greater than 0 in the tumor CT image to 0, and setting the pixel value less than -1200 in the tumor CT image to -1200 , Leave the other pixels unchanged, and then normalize the pixel value of the tumor CT image to the value range of [0, -1200] to exclude the non-tumor area in the tumor CT image.

Preferably, the size of the annotation of the mask tumor area is the same as the size of the tumor CT image, and the pixel value of the tumor area in the tumor CT image is set to 1, and the pixel value of the non-tumor area is set to 0, thereby forming the mask matrix format Mask tumor area marking.

Preferably, the LIDC data set is divided into multiple training data sets and multiple verification data sets according to the ratio of the training data set to the verification data set of 9:1, and each training data set and each verification data set includes one Tumor CT image and a corresponding mask tumor area annotation.

Preferably, the 3D Unet network model is composed of an input layer, an output layer, a 3D convolutional layer, a batch regularization layer, an activation layer, a deconvolution layer, and a maximum pooling layer, wherein the input layer size is 96× 96×32, the model maximum pooling layer is composed of 3 layers of downsampling, the deconvolution layer is composed of 3 layers of upsampling, and the size of the output layer is 96×96×32.

Preferably, the method for constructing a 3D UNet network model for tumor detection further includes: using an Adam optimizer to optimize the parameters of each layer of the 3D Unet network model; and using a DiceLoss loss function to evaluate the loss generated by the 3D Unet network model.

Preferably, the method for constructing a 3D UNet network model for detecting tumors further includes: acquiring CT images to be detected from an image scanning device; and inputting the CT images to be detected into the 3D Unet network model for detection to detect various defects. Regular tumor lesion area, and display the tumor lesion area on the monitor.

On the other hand, the present invention also provides a 3D UNet network model construction device for detecting tumors, which includes a processor suitable for implementing various computer program instructions and a memory suitable for storing multiple computer program instructions. The instructions are loaded by the processor and executed as described above for 3D tumor detection The method steps of the UNet network model construction method.

In another aspect, the present invention also provides a computer-readable storage medium that stores a plurality of computer program instructions, wherein the computer program instructions are loaded by the processor of the computer device and executed as described above. 3D for tumor detection The method steps of the UNet network model construction method.

Beneficial effect

Compared with the prior art, the 3D UNet network model construction method, device and computer-readable storage medium for detecting tumors of the present invention can construct 3D accurate tumor segmentation UNet network model, through the 3D UNet network model to effectively segment various irregular tumor regions, improve the accuracy and speed of tumor detection, and the effectiveness of tumor detection is not limited to individual lung morphological differences, so as to quickly and accurately locate Tumor location provides medical guidance for doctors on tumor diagnosis and treatment.

Description of the drawings

Fig. 1 is a schematic block diagram of a preferred embodiment of a 3D UNet network model construction device for tumor detection according to the present invention;

Fig. 2 is a method flowchart of a preferred embodiment of a method for constructing a 3D UNet network model for tumor detection according to the present invention.

The realization of the objectives, functional characteristics and advantages of the present invention will be further described in conjunction with the embodiments and with reference to the accompanying drawings.

Embodiments of the invention

In order to further explain the technical means and effects of the present invention to achieve the intended purpose of the invention, the specific implementation, structure, features and effects of the present invention will be described in detail below with reference to the drawings and preferred embodiments. It should be understood that the specific embodiments described here are only used to explain the present invention, but not to limit the present invention.

Referring to Fig. 1, Fig. 1 is a schematic structural diagram of a preferred embodiment of a 3D UNet network model construction device for detecting tumors of the present invention. In this embodiment, the 3D UNet network model construction device 1 for tumor detection includes, but is not limited to, a memory 11 suitable for storing various computer program instructions, a processor 12 for executing various computer program instructions, and a display 13. Both the memory 11 and the display 13 are electrically connected to the processor 12 through an electrical connection line, and are connected to the processor 12 through a data bus for data transmission. The processor 12 can call the 3D UNet network model construction program 10 for detecting tumors stored in the memory 11, and execute the 3D UNet network model construction program 10 to input tumor CT images from the image scanning device 3, and Using UNet network to segment lung lobes based on tumor CT image data. The 3D UNet network model construction device 1 may be a personal computer, a notebook computer, a server, and other computer devices installed with the 3D UNet network model construction program 10 of the present invention.

In this embodiment, the 3D UNet network model construction device 1 is connected to a medical image database 2 and an image scanning device 3. The medical imaging database 2 stores LIDC data sets of multiple tumor cases as samples. For example, the medical imaging database 2 stores 1000 LIDC data sets, and each LIDC data set includes tumor CT images and tumor area annotations in xml format. The image scanning device 3 may be a CT scanner, which can scan the lungs of patients to obtain CT images of tumors. The 3D The UNet network model construction device 1 executes the 3D UNet network model construction program 10 through the processor 12 to obtain multiple LIDC data sets from the medical imaging database 2, construct a 3D UNet network model from the LIDC data set, and obtain the patient’s tumor from the image scanning device 3 The CT images are input to the 3D UNet network model, and the 3D UNet network model is used to quickly and accurately detect the tumor lesion area on the input tumor CT image.

In this embodiment, the memory 11 includes at least one type of readable storage medium. The readable storage medium includes flash memory, hard disk, multimedia card, card-type memory (for example, SD or DX memory, etc.), and magnetic memory. , Disks, CDs, etc. In some embodiments, the memory 11 may be an internal storage unit of the 3D UNet network model construction device 1 for tumor detection, such as the hard disk and read-only memory of the 3D UNet network model construction device 1 for tumor detection. ROM, random access memory RAM, electrically erasable memory EEPROM, flash memory FLASH or optical disc, etc. In other embodiments, the memory 11 may also be 3D for detecting tumors. The external storage device of the UNet network model construction device 1, such as the 3D The plug-in hard disk, smart memory card (Smart memory card) equipped on the UNet network model building device 1 Media Card, SMC), Secure Digital (Secure Digital, SD) card, flash card (Flash Card), etc. Further, the memory 11 may also include both an internal storage unit of the 3D UNet network model construction device 1 and an external storage device. The memory 11 can not only be used to store the 3D Application software and various data of UNet network model construction device 1, such as storage for 3D The program code of the UNet network model construction program 10, etc., can also be used to temporarily store the tumor lesion area that has been output or will be output.

In this embodiment, the processor 12 may be a central processing unit (Central Processing Unit) in some embodiments. Processing Unit (CPU), controller, microcontroller, microprocessor or other data processing chip, used to call and run the program code or processing data stored in the memory 11, for example, execute the 3D UNet network model construction program 10, etc. The display 13 may be a touch display screen or a general LED display screen, which can display the detected tumor lesion area.

Optionally, in other embodiments, the 3D UNet network model construction program 10 for tumor detection can also be divided into one or more modules, one or more modules are stored in the memory 11, and are composed of one Or multiple processors (processor 12 in this embodiment) are executed to complete the present invention. The module referred to in the present invention refers to a series of computer program instruction segments that can complete specific functions, and is used to describe the 3D UNet network model construction program 10 The execution process in the UNet network construction device 1.

In this embodiment, the 3D UNet network model building program 10 for tumor detection is composed of program modules composed of multiple computer program instructions, including but not limited to, an image data acquisition module 101, an image data processing module 102, The network model construction module 103, the network model training module 104, and the tumor detection module 105. The module referred to in the present invention refers to a series of computer program instruction segments that can be executed by the processor 12 of the 3D UNet network model construction device 1 and can complete fixed functions, and are stored in the memory 11.

The image data acquisition module 101 is used to acquire a LIDC data set from the medical image database 2, and the LIDC data set includes tumor CT images and tumor region annotations in xml format. In this embodiment, the medical imaging database 2 stores LIDC data sets of multiple tumor cases. For example, the medical imaging database 2 stores LIDC data sets of 1000 tumor cases.

The image data processing module 102 is used to convert the tumor area annotation in xml format into a mask tumor area annotation; in this embodiment, the image data processing module 102 converts the tumor area annotation in xml format into a tumor area in mask format Labeling (referred to as mask tumor area labeling); where the size of the mask tumor area label is the same as the size of the tumor CT image, the pixel value of the tumor area in the tumor CT image is set to 1, and the pixel value of the non-tumor area is set to 0. This constitutes the mask tumor area label in the mask matrix format.

The image data processing module 102 is also used to divide the transformed LIDC data set into a training data set and a verification data set; in this embodiment, each training data set and each verification data set includes a tumor CT image And a corresponding mask tumor area label. Assuming that a total of 1000 LIDC data sets are input, the LIDC data set is divided according to the ratio of the training data set to the verification data set of 9:1, the training data set has 900 cases, and the verification data set has 100 cases.

The image data processing module 102 is also used to preprocess the training data set and the verification data set, and normalize the pixel values of the tumor CT images. In this embodiment, the preprocessing of the training data set and the verification data set includes the following steps: the interval between the pixels in the training data set and the verification data set is set to 1, so that the data of the 3D Unet network model is input Have a uniform interval; perform the following steps for the tumor CT image and mask tumor area annotation in the training data set: obtain the center point of the mask tumor area annotation, and use the center point as the center point of the 96×96×32 matrix according to 96×96 ×32 size Randomly crop the tumor area, randomly zoom in and out, randomly rotate a certain angle, the angle can be 90, 180 or 270 degrees, randomly flip up and down, and generate diversified training data to enhance the robustness of the 3D Unet network model Great; perform the following steps for the tumor CT image and mask tumor area of the validation data set: Obtain the center point of the mask tumor area, and use the center point as the center point of the 96×96×32 matrix, according to the size of 96×96×32 The tumor area is cropped, and the cropped tumor CT image and mask tumor area data are saved for subsequent verification of the effectiveness of the 3D Unet network model. In this embodiment, the normalization processing of the pixel value of the tumor CT image includes the steps: setting the pixel value greater than 0 in the tumor CT image to 0, and setting the pixel value less than -1200 in the tumor CT image to -1200 , Other pixels remain unchanged, that is, the pixel values of tumor CT images are normalized to the numerical range of [0, -1200] to exclude most non-tumor areas.

The network model construction module 103 is used to construct a 3D Unet network model based on the keras framework. Those skilled in the art can know that the Keras framework is a Python API based on deep learning neural networks, written in Python language, and is a highly modular neural network library. In this embodiment, the 3D Unet network model is composed of an input layer, an output layer, a 3D convolution layer, a batch regularization layer, an activation layer, a deconvolution layer, and a maximum pooling layer. The input layer size is 96×96 ×32, the model has 3 layers of downsampling (maximum pooling layer), 3 layers of upsampling (deconvolution layer), and the output layer size is 96×96×32. The network model building module 103 is also used to optimize the parameters of each layer of the 3D Unet network model by using the Adam optimizer, and to use the DiceLoss loss function to optimize the 3D Unet network model to assess the loss. Those skilled in the art can know that the Adam optimizer is an extension of the stochastic gradient descent optimization algorithm, and is widely used in deep learning neural network applications, especially tasks such as computer vision and natural language processing. The DiceLoss loss function is a network model that uses the dice coefficient as the loss function to evaluate the loss when using deep learning for medical image segmentation.

The network model training module 104 is used to train the 3D Unet network model by using the training data set to obtain the weight of the 3D Unet network model. In this embodiment, the network model training module 104 inputs the tumor CT images and mask tumor region annotations of the training data set into the 3D Unet network model for training according to a preset amount of data. For example, the input data amount of each round (batchsize ) Set to 12 batches, the total number of rounds (epoch) is set to 200 rounds; each round of training the 3D Unet network model will generate weights, save the weight of the lowest loss (loss) of the 3D Unet network model, and get the most optimized 3D Unet Network model.

The network model training module 104 is also used to verify the effectiveness of the 3D Unet network model using the verification data set to verify whether the constructed 3D Unet network model can be used as a subsequent effective and accurate detection of the tumor area. In this embodiment, the tumor CT image of the verification data set is input into the 3D Unet network model to output the mask label of the tumor area, and the output mask label of the tumor area is compared with the mask tumor area label in the verification data set. If both Basically the same, it indicates that the constructed 3D Unet network model is effective and can be used as follow-up tumor detection; if the difference between the two is large, it indicates that the constructed 3D The Unet network model is invalid, and a valid 3D Unet network model needs to be reconstructed.

The tumor detection module 105 is used to obtain CT images to be detected from the image scanning device 3, and input the CT images to be detected into the 3D Unet network model for detection, and then detect various irregular tumor focus areas, and The tumor lesion area is displayed on the display 13, so as to help the doctor provide more comprehensive guidance for the planning of adjuvant radiotherapy and chemotherapy before the tumor operation and the evaluation of the effect of postoperative radiotherapy and chemotherapy.

Referring to FIG. 2, it is a flowchart of a preferred embodiment of a method for constructing a 3D UNet network model for tumor detection in the present invention. In this embodiment, the various method steps of the 3D UNet network model construction method are implemented by a computer software program, and the computer software program is stored in a computer-readable storage medium (such as the memory 11) in the form of computer program instructions, The computer program instructions can be loaded by a processor (for example, the processor 12) and execute the following steps:

In step S21, a LIDC data set is obtained from the medical image database 2. The LIDC data set includes tumor CT images and tumor region annotations in xml format. In this embodiment, the medical imaging database 2 stores LIDC data sets of multiple tumor cases. For example, the medical imaging database 2 stores LIDC data sets of 1000 tumor cases.

In step S22, the tumor area annotation in xml format is converted into a mask tumor area annotation; in this embodiment, the tumor area annotation in xml format is converted into a tumor area annotation in mask format (referred to as mask tumor area annotation); wherein, The size of the mask tumor area annotation is the same as that of the tumor CT image. The pixel value of the tumor area in the tumor CT image is set to 1, and the pixel value of the non-tumor area is set to 0, thereby forming the mask tumor area annotation in the mask matrix format.

Step S23: Divide the transformed LIDC data set into a training data set and a verification data set; in this embodiment, each training data set and each verification data set includes a tumor CT image and a corresponding mask tumor area Label. Assuming that a total of 1000 LIDC data sets are input, the LIDC data set is divided according to the ratio of the training data set to the verification data set of 9:1, the training data set has 900 cases, and the verification data set has 100 cases.

In step S24, the training data set and the verification data set are preprocessed, and the pixel value of the tumor CT image is normalized. In this embodiment, the preprocessing of the training data set and the verification data set includes the following steps: the interval between the pixels in the training data set and the verification data set is set to 1, so that the data of the 3D Unet network model is input Have a uniform interval; perform the following steps for the tumor CT image and mask tumor area annotation in the training data set: obtain the center point of the mask tumor area annotation, and use the center point as the center point of the 96×96×32 matrix according to 96×96 ×32 size Randomly crop the tumor area, randomly zoom in and out, randomly rotate a certain angle, the angle can be 90, 180 or 270 degrees, randomly flip up and down, and generate diversified training data to enhance the robustness of the 3D Unet network model Great; perform the following steps for the tumor CT image and mask tumor area annotation of the validation data set: obtain the center point of the mask tumor area, and use the center point as the center point of the 96×96×32 matrix, according to the size of 96×96×32 The tumor area is cropped, and the cropped tumor CT image and mask tumor area data are saved for subsequent verification of the effectiveness of the 3D Unet network model. In this embodiment, the normalization processing of the pixel value of the tumor CT image includes the steps: setting the pixel value greater than 0 in the tumor CT image to 0, and setting the pixel value less than -1200 in the tumor CT image to -1200 , Other pixels remain unchanged, that is, the pixel values of tumor CT images are normalized to the numerical range of [0, -1200] to exclude most of the non-tumor areas in tumor CT images.

Step S25, build a 3D Unet network model based on the keras framework. Those skilled in the art can know that the Keras framework is a Python API based on deep learning neural networks, written in Python language, and is a highly modular neural network library. In this embodiment, the 3D Unet network model is composed of an input layer, an output layer, a 3D convolution layer, a batch regularization layer, an activation layer, a deconvolution layer, and a maximum pooling layer. The input layer size is 96×96 ×32, the model has 3 layers of downsampling (maximum pooling layer), 3 layers of upsampling (deconvolution layer), and the output layer size is 96×96×32.

In step S26, the Adam optimizer is used to optimize the parameters of each layer of the 3D Unet network model, and the DiceLoss loss function is used to evaluate the loss generated by the 3D Unet network model. Those skilled in the art can know that the Adam optimizer is an extension of the stochastic gradient descent optimization algorithm, and is widely used in deep learning neural network applications, especially tasks such as computer vision and natural language processing. The DiceLoss loss function is a network model that uses the dice coefficient as the loss function to evaluate the loss when using deep learning for medical image segmentation.

Step S27: Use the training data set to train the 3D Unet network model to obtain the weight of the 3D Unet network model. In this embodiment, the tumor CT images and mask tumor area annotations of the training data set are input into 3D according to the preset data amount. Unet network model training, for example, set the input data size (batchsize) of each round to 12 batches, and the total number of rounds (epoch) is set to 200 rounds; each round of training 3D Unet network model will generate weights and save 3D The weight of the lowest loss (loss) of the Unet network model obtains the optimized 3D Unet network model.

In step S28, the validity of the 3D Unet network model is verified by using the verification data set to verify whether the constructed 3D Unet network model can be used as a follow-up effective and accurate detection of the tumor area. In this embodiment, the tumor CT image of the verification data set is input into the 3D Unet network model to output the mask label of the tumor area, and the output mask label of the tumor area is compared with the mask tumor area label in the verification data set. If both Basically the same, it indicates that the constructed 3D Unet network model is effective and can be used as follow-up tumor detection; if the difference between the two is large, it indicates that the constructed 3D The Unet network model is invalid, and a valid 3D Unet network model needs to be reconstructed.

Step S29: Obtain the CT image to be detected from the image scanning device 3, and input the CT image to be detected into the 3D Unet network model for detection, and then detect various irregular tumor lesion areas, and display the tumor lesion area in The display 13 is thus helpful for doctors to provide more comprehensive medical guidance for the formulation of preoperative adjuvant radiotherapy and chemotherapy programs for tumors and the evaluation of the effects of postoperative radiotherapy and chemotherapy.

The present invention also provides a computer-readable storage medium that stores a plurality of computer program instructions, and the computer program instructions are loaded by a processor of a computer device and execute the 3D method for detecting tumors of the present invention. The steps of the UNet network model construction method. Those skilled in the art can understand that all or part of the steps of the various methods in the foregoing embodiments can be completed by related program instructions. The program can be stored in a computer-readable storage medium. The storage medium may include: read-only memory, random access memory, Disk or CD, etc.

The 3D UNet network model construction method, device and computer-readable storage medium for detecting tumors of the present invention can construct 3D precise tumor segmentation UNet network model, through the 3D UNet network model to effectively segment various irregular tumor regions, improve the accuracy and speed of tumor detection, and the effectiveness of tumor detection is not limited to individual lung morphological differences, so as to quickly and accurately locate Tumor location provides medical guidance for doctors on tumor diagnosis and treatment.

The above are only preferred embodiments of the present invention, and do not limit the scope of the present invention. Any equivalent structure or equivalent process transformation made by using the contents of the description and drawings of the present invention, or directly or indirectly applied to other related technologies In the same way, all fields are included in the scope of patent protection of the present invention.

Industrial applicability

Claims

A method for constructing a 3D UNet network model for tumor detection is characterized in that the method includes the following steps:

Obtain the LIDC data set from the medical imaging database, the LIDC data set includes tumor CT images and tumor area annotations in xml format;

Convert the tumor area annotation in xml format into mask tumor area annotation;

Divide the transformed LIDC data set into training data set and validation data set;

Preprocess the training data set and validation data set, and normalize the pixel values of tumor CT images;

Build a 3D Unet network model based on the keras framework;

3D constructed using training data set pairs The Unet network model is trained to obtain the weight of the 3D Unet network model;

The 3D constructed using the verification data set The validity of the Unet network model is verified.
The 3D UNet network model construction method for tumor detection according to claim 1, wherein the step of preprocessing the training data set and the verification data set comprises:

Set the interval between the pixels in the training data set and the verification data set to 1, so that the data input to the 3D Unet network model has a uniform interval;

For the tumor CT images and mask tumor region annotations in the training data set, the following steps are performed: Obtain the center point of the mask tumor region annotation, and use the center point as the center point of the 96×96×32 matrix, according to the size of 96×96×32 Randomly crop the tumor area, randomly zoom in and out, randomly rotate the angle, and randomly flip up and down to generate diverse training data;

For the tumor CT image and mask tumor area annotation of the validation data set, the following steps are performed: Obtain the center point of the mask tumor area, and use the center point as the center point of the 96×96×32 matrix, and the tumor according to the size of 96×96×32 The area is cropped, and the cropped tumor CT image and mask tumor area are saved.
The method for constructing a 3D UNet network model for detecting tumors according to claim 2, wherein the step of normalizing the pixel values of CT images of tumors comprises:

Set the pixel value greater than 0 in the tumor CT image to 0, set the pixel value less than -1200 in the tumor CT image to -1200, and leave the other pixels unchanged, and then normalize the pixel value of the tumor CT image to [0, -1200] This numerical interval is used to exclude non-tumor areas in tumor CT images.
The method for constructing a 3D UNet network model for detecting tumors according to claim 1, wherein the size of the mask tumor area annotation is the same as the tumor CT image size, and the pixel value of the tumor area in the tumor CT image Set to 1, the pixel value of the non-tumor area is set to 0, thereby forming the mask tumor area label in the mask matrix format.
The method for constructing a 3D UNet network model for tumor detection according to claim 1, wherein the LIDC data set is divided into a plurality of training data sets and a training data set according to the ratio of the training data set to the verification data set of 9:1. Multiple verification data sets, each training data set and each verification data set include a tumor CT image and a corresponding mask tumor area annotation.
The method for constructing a 3D UNet network model for tumor detection according to claim 1, wherein the 3D Unet network model is composed of an input layer, an output layer, a 3D convolution layer, a batch regularization layer, an activation layer, and an inverse layer. The convolutional layer and the maximum pooling layer are composed of, wherein the input layer size is 96×96×32, the model maximum pooling layer is composed of 3 layers of downsampling, and the deconvolution layer is composed of 3 layers of upsampling , The size of the output layer is 96×96×32.
The method for constructing a 3D UNet network model for tumor detection according to claim 1, wherein the method further comprises the following steps:

Use Adam optimizer to optimize the parameters of each layer of the 3D Unet network model;

The DiceLoss loss function is used to evaluate the loss generated by the 3D Unet network model.
The method for constructing a 3D UNet network model for tumor detection according to claim 1, wherein the method further comprises the following steps:

Obtain CT images to be detected from image scanning equipment;

Input the CT image to be detected into the 3D Unet network model for detection to detect various irregular tumor lesion areas, and display the tumor lesion area on the display.
A 3D UNet network model construction device for detecting tumors, comprising a processor suitable for realizing various computer program instructions and a memory suitable for storing multiple computer program instructions, characterized in that the computer program instructions are executed by the processor Load and execute each method step of the 3D UNet network model construction method for tumor detection according to any one of claims 1 to 8.
A computer-readable storage medium storing a plurality of computer program instructions, wherein the computer program instructions are loaded by a processor of a computer device and executed as described in any one of claims 1 to 8 Various method steps of the 3D UNet network model construction method for detecting tumors.