A CT image segmentation method, terminal and storage medium suitable for bladder cancer

Abstract

The present invention discloses a CT image segmentation method, terminal and storage medium applicable to bladder cancer. The method comprises: obtaining a target CT image, processing the target CT image into a preset format; inputting the preprocessed target CT image into a target image segmentation model, and obtaining a lesion segmentation result output by the target image segmentation model. The CT image segmentation method applicable to bladder cancer provided by the present invention obtains an image segmentation model by training a training set including sample CT images and lesion segmentation results corresponding to the sample CT images, and performs CT image segmentation applicable to bladder cancer by an image segmentation module, without the need for manual segmentation, thus shortening the image segmentation time, and expanding the training set by selecting samples from the test set, thereby improving the training effect of the image segmentation model and ensuring the accuracy of image segmentation.

Description

Technical Field

[0001] The present invention relates to the field of magnetic resonance technology, and in particular to a CT image segmentation method, a terminal and a storage medium suitable for bladder cancer. Background Art

[0002] In clinical practice, cancer lesions are usually identified by computer tomography (CT) scanning, which is then processed to identify the lesions. Currently, the identification and segmentation of lesions rely on manual segmentation by radiologists, which is very time-consuming and unstable.

[0003] Therefore, the prior art still needs to be improved and enhanced. Summary of the invention

[0004] In view of the above-mentioned defects of the prior art, a CT image segmentation method, terminal and storage medium suitable for bladder cancer are provided, aiming to solve the problem of time-consuming manual segmentation of lesions in the prior art.

[0005] A first aspect of the present invention provides a CT image segmentation method applicable to bladder cancer, comprising:

[0006] Acquire a target CT image, and process the target CT image into a preset format;

[0007] Inputting the preprocessed target CT image into a target image segmentation model to obtain a lesion segmentation result output by the target image segmentation model;

[0008] Wherein, the target image segmentation model is obtained by training based on a first training set and a second training set, the first training set includes multiple groups of training data, each group of training data includes a first sample CT image and a first sample lesion segmentation result corresponding to the first sample CT image, the second training set includes multiple groups of training data, each group of training data includes a second sample CT image and a second sample lesion segmentation result corresponding to the second sample CT image, the second sample CT image is selected from a test set used to test the target image segmentation model, and the test set includes a third sample CT image and a third sample lesion segmentation result corresponding to the third sample CT image.

[0009] The CT image segmentation method applicable to bladder cancer, wherein, before acquiring the target CT image, the method further comprises:

[0010] Constructing an initial image segmentation model, acquiring multiple CT images to annotate lesion segmentation results, and generating the first training set and the test set;

[0011] Using the first training set to train the initial image segmentation model to obtain an intermediate image segmentation model;

[0012] Constructing the second training set according to the intermediate image segmentation model and the test set;

[0013] The intermediate image segmentation model is trained according to the first training set and the second training set, and the target image segmentation model is obtained after the training is completed.

[0014] The CT image segmentation method applicable to bladder cancer, wherein the second training set is constructed according to the intermediate image segmentation and combination model and the test set, comprises:

[0015] After processing the third sample CT images in the test set into the preset format, obtaining predicted lesion segmentation results of each of the third sample CT images according to the intermediate image segmentation model;

[0016] For the target third sample CT image, if the predicted lesion segmentation result of the target third sample CT image is consistent with the third sample lesion segmentation result corresponding to the target third sample CT image, the target third sample CT image is used as the second sample CT image, and the predicted lesion segmentation result corresponding to the target third sample CT image is used as the corresponding second sample lesion segmentation result, constituting a set of training data in the second training set.

[0017] The CT image segmentation method applicable to bladder cancer, wherein the target image segmentation model includes at least one decoding module, at least one encoding module, and a first connection module, the decoding module and the encoding module correspond one to one, the corresponding decoding module and the encoding module are connected through a second connection module, and the second connection module is provided with an attention gate;

[0018] Among them, each of the encoding modules, the first connection module and each of the decoding modules are connected in sequence, the input of each of the encoding modules and the first connection module is the output of the previous module, and the input of each decoding module is the output of the previous module and the output of the corresponding second connection module.

[0019] The CT image segmentation method suitable for bladder cancer, wherein each of the encoding modules includes a convolution layer, a batch normalization layer, an activation function and a downsampling layer, wherein the batch normalization layer includes a residual connection operation.

[0020] The CT image segmentation method suitable for bladder cancer is characterized in that the second connection module is connected to the batch normalization layer, and the output result of the batch normalization layer is activated by the attention gate and then input into the corresponding decoding module, and the attention gate in the second connection module is used to comprehensively calculate the self-attention by combining the parameters of the corresponding encoding module, the corresponding decoding module and the first module, wherein the first module is the previous module of the corresponding encoding module.

[0021] The CT image segmentation method suitable for bladder cancer, wherein each of the decoding modules includes an upsampling layer, a convolution layer, a batch normalization layer and an activation function.

[0022] The CT image segmentation method applicable to bladder cancer, wherein, after obtaining the lesion segmentation result output by the image segmentation model, the method further comprises:

[0023] The features of the lesion are extracted according to the lesion segmentation result by using a pre-built feature extractor.

[0024] A second aspect of the present invention provides a terminal, comprising: a processor, a storage medium communicatively connected to the processor, the storage medium being suitable for storing a plurality of instructions, and the processor being suitable for calling the instructions in the storage medium to execute the steps of implementing any one of the above-mentioned CT image segmentation methods applicable to bladder cancer.

[0025] A third aspect of the present invention provides a storage medium, wherein the storage medium stores one or more programs, and the one or more programs can be executed by one or more processors to implement the steps of any of the above-mentioned CT image segmentation methods applicable to bladder cancer.

[0026] Beneficial effects: Compared with the prior art, the present invention provides a CT image segmentation method, terminal and storage medium suitable for bladder cancer. The CT image segmentation method suitable for bladder cancer provided by the present invention obtains an image segmentation model by training a training set including sample CT images and lesion segmentation results corresponding to the sample CT images. The CT image segmentation suitable for bladder cancer is performed through the image segmentation module, and no manual segmentation is required, which shortens the image segmentation time. The training set is expanded by selecting samples from the test set, thereby improving the training effect of the image segmentation model and ensuring the accuracy of image segmentation. BRIEF DESCRIPTION OF THE DRAWINGS

[0027] Figure 1 A flow chart of an embodiment of a CT image segmentation method for bladder cancer provided by the present invention;

[0028] Figure 2A schematic diagram of a process for obtaining a first training set in an embodiment of a CT image segmentation method applicable to bladder cancer provided by the present invention;

[0029] Figure 3 A schematic diagram of a process for obtaining a second training set in an embodiment of a CT image segmentation method applicable to bladder cancer provided by the present invention;

[0030] Figure 4 A schematic diagram of the model training process in an embodiment of the CT image segmentation method for bladder cancer provided by the present invention;

[0031] Figure 5 A schematic diagram of the structure of a target image segmentation model in an embodiment of a CT image segmentation method applicable to bladder cancer provided by the present invention;

[0032] Figure 6 A structural principle diagram of an embodiment of a terminal provided by the present invention. DETAILED DESCRIPTION

[0033] In order to make the purpose, technical solution and effect of the present invention clearer and more specific, the present invention is further described in detail with reference to the accompanying drawings and examples. It should be understood that the specific embodiments described herein are only used to explain the present invention and are not used to limit the present invention.

[0034] The present invention provides a CT image segmentation method applicable to bladder cancer, which can be applied in a terminal, and the terminal can be but not limited to various personal computers, laptops, mobile phones, tablet computers, etc.

[0035] Embodiment 1

[0036] like Figure 1 As shown, the CT image segmentation method for bladder cancer provided by the present invention comprises the following steps:

[0037] S100: Acquire a target CT image, and process the target CT image into a preset format.

[0038] The target CT image is a CT image to be segmented, and the target CT image can be obtained by CT scanning. After the target CT image is acquired, since different conditions and different scanning objects correspond to different CT images, in order to facilitate processing, in this embodiment, the CT image to be segmented is processed into a unified format, that is, the preset format.

[0039] Specifically, processing the target CT image into a preset format includes preprocessing and data enhancement. The preprocessing includes adjusting the anatomical coordinate system of the CT image to RAS, adjusting the image voxels to a preset spatial size, and adjusting the image thickness to a preset thickness. Adjusting the image thickness to a preset thickness can be achieved by interpolation. Data enhancement includes randomly sampling the preprocessed CT image N times through a preset image reading window, where N is a positive integer. The size of the image reading window and the value of N can be set according to the actual application. For example, the size of the image reading window can be set to (96, 96, 96), and N is 8. Data enhancement of the target CT image can increase data randomness and improve segmentation accuracy.

[0040] S200: Input the preprocessed target CT image into a target image segmentation model to obtain a lesion segmentation result output by the target image segmentation model.

[0041] The target image segmentation model is obtained by training based on a first training set and a second training set, wherein the first training set includes multiple groups of training data, each group of training data includes a first sample CT image and a first sample lesion segmentation result corresponding to the first sample CT image, and the second training set includes multiple groups of training data, each group of training data includes a second sample CT image and a second sample lesion segmentation result corresponding to the second sample CT image, and the second sample CT image is selected from a test set used to test the target image segmentation model, and the test set includes a third sample CT image and a third sample lesion segmentation result corresponding to the third sample CT image.

[0042] In this embodiment, before acquiring the target CT image, the target image segmentation model is first trained, which specifically includes:

[0043] S001, constructing an initial image segmentation model, acquiring multiple CT images to annotate lesion segmentation results, and generating the first training set and the test set;

[0044] S002, using the first training set to train the initial image segmentation model, and obtaining an intermediate image segmentation model after the training is completed;

[0045] S003, constructing the second training set according to the intermediate image segmentation model and the first training set;

[0046] S004. Train the intermediate image segmentation model according to the first training set and the second training set, and obtain the target image segmentation model after the training is completed.

[0047] Specifically, the initial image segmentation model is a neural network model, and the process of training the initial image segmentation model to obtain an intermediate image segmentation model and training the intermediate image segmentation model to obtain the target image segmentation model is a process of modifying the parameters of the neural network model through training.

[0048] The first sample lesion segmentation result corresponding to the first sample CT image and the third sample lesion segmentation result corresponding to the third sample CT image in the data of the first training set and the test set can be obtained by manual outlining and annotation. The generation process of the first training set can be as follows: Figure 2 As shown, firstly, a CT image is obtained as a first sample CT image and stored in a database. The lesion segmentation result corresponding to each first sample CT image is obtained by annotating by two groups of doctors. It is judged whether the annotated results of the two groups are consistent. If they are consistent, the annotated result is used as the first sample lesion segmentation result corresponding to the first sample CT image and saved in the database; if they are inconsistent, the two groups of doctors jointly judge and re-annotate. The generation method of the test set is consistent with the production method of the first training set.

[0049] In the prior art, the neural network model is trained once according to a pre-set training set, and then the training result is tested using a test set. If a model with higher accuracy is to be obtained, a large amount of sample data is required, which requires a large amount of manpower to implement annotation, and the efficiency is low. In this embodiment, in order to improve the efficiency of model training, the initial image segmentation model is first trained using the first training set, and an intermediate image segmentation model is obtained after the training is completed. A second training set is generated based on the intermediate image segmentation model and the training set to expand the training set, and then the expanded training set is used for a second training to finally generate the target image segmentation model.

[0050] It is worth noting that when the image segmentation model is trained by the first sample CT image and the second sample CT image in the first training set and the second training set, the first sample CT image and the second sample CT image are first processed into the preset format. The specific processing method can be seen in the description of step S100 in the previous text. Whether the training is completed can be determined by the loss function. If the loss function converges, the training is judged to be completed, the iterative training is stopped, and the model parameters at this time are saved. The flowchart of model training can be as follows Figure 4 shown.

[0051] Specifically, constructing the second training set according to the intermediate image segmentation model and the test set comprises the steps of:

[0052] S031, after processing the third sample CT images in the test set into the preset format, obtaining predicted lesion segmentation results of each of the third sample CT images according to the intermediate image segmentation model;

[0053] S032. For the target third sample CT image, if the predicted lesion segmentation result of the target third sample CT image is consistent with the third sample lesion segmentation result corresponding to the target third sample CT image, then the target third sample CT image is used as the second sample CT image, and the predicted lesion segmentation result corresponding to the target third sample CT image is used as the corresponding second sample lesion segmentation result, to constitute a set of training data in the second training set.

[0054] The intermediate image segmentation model obtained by training the first training set is used to obtain the predicted lesion segmentation results of each of the third sample CT images in the test set, that is, the intermediate image segmentation model is used to predict the lesion segmentation results corresponding to each of the third sample CT images. Specifically, the third sample CT image can be processed into the preset format and then input into the intermediate image segmentation model, and the lesion segmentation result output by the intermediate image segmentation model is obtained as the predicted lesion segmentation result of the third sample CT image. For the target third sample CT image, if the predicted lesion segmentation result of the target third sample CT image is consistent with the third sample lesion segmentation result corresponding to the target third sample CT image, then the target third sample CT image can be used as a training set to continue training the intermediate image segmentation model. Specifically, the target third sample CT image is used as the second sample CT image, and the predicted lesion segmentation result corresponding to the target third sample CT image is used as the corresponding second sample lesion segmentation result to form a set of training data in the second training set.

[0055] The process diagram of constructing the second training set can be as follows: Figure 3 shown.

[0056] After the second training set is regenerated, the first training set and the second training set are used as training sets to iteratively train the intermediate image segmentation model, and the target image segmentation model is obtained after the training is completed. After the target image segmentation model is obtained, the performance of the target image segmentation model can be tested through the test set. Specifically, the lesion segmentation results of each third sample CT image in the test set are obtained according to the target image segmentation model and compared with the data in the test set, and the dice coefficient is calculated to test the prediction effect of the model. It is worth noting that in order to improve the accuracy of the test results of the target image segmentation model, the third sample CT image and the first sample CT image are different images.

[0057] The structural diagram of the target image segmentation model can be as follows Figure 5 As shown, it includes at least one decoding module, at least one encoding module, and a first connection module, the decoding modules correspond to the encoding modules one by one, and the corresponding decoding modules and encoding modules are connected through a second connection module;

[0058] The encoding modules, the first connection modules and the decoding modules are connected in sequence, the input of each encoding module and the first connection module is the output of the previous module, and the input of each decoding module is the output of the previous module and the output of the second connection module corresponding to the encoding module.

[0059] The number of the encoding modules and the decoding modules can be set according to actual applications. The first connection module can be a plurality of groups of convolutions. In the present embodiment, there are 4 encoding modules and 4 decoding modules respectively, and the first connection module includes 2 groups of 3*3 convolutions.

[0060] Each of the encoding modules includes a convolution layer, a batch normalization layer, an activation function, and a downsampling layer. The convolution layer includes two convolution operations, and the convolution kernel size can be 3*3*3. The batch normalization layer includes a residual connection operation, that is, a short-cut connection is added to prevent gradient diffusion. Specifically, the number of layers spanned by the short-cut can be a single layer or multiple layers. The activation function can use common activation functions such as ReLU. In this embodiment, the downsampling layer uses a 2*2*2 maximum pooling (max_pooling) method for downsampling, and the step size stride=2. Of course, those skilled in the art can also use other methods for downsampling.

[0061] The second connection module is connected to the batch normalization layer, and the output result of the batch normalization layer is activated by the attention gate and then input into the corresponding decoding module. The second connection module is provided with an attention gate, and the attention gate is used to determine the weight of the output of the batch normalization layer in the corresponding encoding module in the input of the decoding module. The attention gate calculates the self-attention by comprehensively considering the parameters of the corresponding encoding module, the corresponding decoding module and the first module. The first module is the previous module of the corresponding encoding module, that is, for the attention gate in the second connection module connecting the encoding module A and the decoding module B, the attention is calculated by summing the parameters of the encoding module A, the parameters of the decoding module B and the parameters of the previous encoding module C of the encoding module A. The self-attention gate can automatically learn parameters to adjust the activation value in the process of training to obtain the target image segmentation model (including the process of training the initial image segmentation model to obtain the intermediate image segmentation model and training the intermediate image segmentation model to obtain the target image segmentation model), that is, adjusting the weight of the output of the batch normalization layer of the encoding module in the input of the corresponding decoding module.

[0062] Each of the decoding modules includes an upsampling layer, a batch normalization layer, a convolution layer and an activation function. In this embodiment, the upsampling layer can be upsampled by deconvolution, the convolution kernel size can be 2*2*2, and the step size stride=2. The convolution layer includes two normal convolution operations, and the convolution kernel size can be 3*3*3.

[0063] It is understandable that the model parameters in the above-mentioned target image segmentation model, including convolution kernel size, step size, sampling method, etc., can be adjusted or reset by technical personnel in this field according to experiments, and the present invention does not make specific limitations on this.

[0064] After obtaining the lesion segmentation result output by the target image segmentation model, the CT image segmentation method for bladder cancer provided in this embodiment further includes the following steps:

[0065] S300 , extracting features of the lesion according to the lesion segmentation result using a pre-built feature extractor.

[0066] In this embodiment, a feature extractor is pre-constructed, and the feature extractor is used to extract features of the lesion according to the lesion segmentation result, including first-order image features, morphological features, gray level co-occurrence matrix features, gray level connected region matrix features, etc. Specifically, the first-order image features describe the distribution of pixel intensities within the image ROI, including image energy, entropy, abundance, uniformity, etc. The morphological features include information describing the shape, volume, compactness, etc. of the lesion site. The gray level co-occurrence matrix (GLCM) feature is extracted by counting the probability of occurrence of pixel pairs in different directions and step sizes to obtain the gray level co-occurrence matrix, and then the distribution of the gray level co-occurrence matrix is ​​quantified to obtain information describing the complexity of the lesion site, the level change, and the coarseness of the texture. The gray level size zone matrix (GLCM) feature is extracted. The GLSZM (GLSZM) feature is a process of obtaining a grayscale connected region matrix by counting the number of pixels with the same adjacent grayscale value, and then quantifying the distribution of the grayscale connected region matrix to obtain information describing the complexity of the lesion site, the level change, and the coarseness and fineness of the texture.

[0067] In summary, this embodiment provides a CT image segmentation method suitable for bladder cancer. An image segmentation model is obtained by training a training set including sample CT images and lesion segmentation results corresponding to the sample CT images. The image segmentation module is used to perform CT image segmentation suitable for bladder cancer. Manual segmentation is not required, which shortens the image segmentation time. The training set is expanded by selecting samples from the test set, thereby improving the training effect of the image segmentation model and ensuring the accuracy of image segmentation.

[0068] It should be understood that, although the various steps in the flowcharts given in the accompanying drawings of the present invention are displayed in sequence according to the indications of the arrows, these steps are not necessarily executed in sequence in the order indicated by the arrows. Unless there is a clear description in this article, the execution of the steps in the present invention is not strictly limited in order, and these steps can be executed in other orders. Moreover, at least a part of the steps of the present invention may include multiple sub-steps or multiple stages, and these sub-steps or stages are not necessarily executed at the same time, but can be executed at different times, and the execution order of these sub-steps or stages is not necessarily to be carried out in sequence, but can be executed in turn or alternately with other steps or at least a part of the sub-steps or stages of other steps.

[0069] Those skilled in the art can understand that all or part of the processes in the above-mentioned embodiment methods can be completed by instructing the relevant hardware through a computer program, and the computer program can be stored in a non-volatile computer-readable storage medium. When the computer program is executed, it can include the processes of the embodiments of the above-mentioned methods. Among them, any reference to memory, storage, database or other media used in the embodiments provided by the present invention can include non-volatile and / or volatile memory. Non-volatile memory can include read-only memory (ROM), programmable ROM (PROM), electrically programmable ROM (EPROM), electrically erasable programmable ROM (EEPROM) or flash memory. Volatile memory can include random access memory (RAM) or external cache memory. As an illustration and not limitation, RAM is available in many forms, such as static RAM (SRAM), dynamic RAM (DRAM), synchronous DRAM (SDRAM), double data rate SDRAM (DDRSDRAM), enhanced SDRAM (ESDRAM), synchronous link (Synchlink) DRAM (SLDRAM), memory bus (Rambus) direct RAM (RDRAM), direct memory bus dynamic RAM (DRDRAM), and memory bus dynamic RAM (RDRAM).

[0070] Embodiment 2

[0071] Based on the above embodiment, the present invention further provides a terminal, whose principle block diagram can be shown as follows: Figure 6 The terminal includes a memory 10 and a processor 20. The memory 10 stores a computer program. When the processor 10 executes the computer program, at least the following steps can be implemented:

[0072] Acquire a target CT image, and process the target CT image into a preset format;

[0073] Inputting the preprocessed target CT image into a target image segmentation model to obtain a lesion segmentation result output by the target image segmentation model;

[0074] Wherein, the target image segmentation model is obtained by training based on a first training set and a second training set, the first training set includes multiple groups of training data, each group of training data includes a first sample CT image and a first sample lesion segmentation result corresponding to the first sample CT image, the second training set includes multiple groups of training data, each group of training data includes a second sample CT image and a second sample lesion segmentation result corresponding to the second sample CT image, the second sample CT image is selected from a test set used to test the target image segmentation model, and the test set includes a third sample CT image and a third sample lesion segmentation result corresponding to the third sample CT image.

[0075] Before acquiring the target CT image, the method further includes:

[0076] Constructing an initial image segmentation model, acquiring multiple CT images to annotate lesion segmentation results, and generating the first training set and the test set;

[0077] Using the first training set to train the initial image segmentation model to obtain an intermediate image segmentation model;

[0078] Constructing the second training set according to the intermediate image segmentation model and the test set;

[0079] The intermediate image segmentation model is trained according to the first training set and the second training set, and the target image segmentation model is obtained after the training is completed.

[0080] Wherein, constructing the second training set according to the intermediate image segmentation and combination model and the test set includes:

[0081] After processing the third sample CT images in the test set into the preset format, obtaining predicted lesion segmentation results of each of the third sample CT images according to the intermediate image segmentation model;

[0082] For the target third sample CT image, if the predicted lesion segmentation result of the target third sample CT image is consistent with the third sample lesion segmentation result corresponding to the target third sample CT image, the target third sample CT image is used as the second sample CT image, and the predicted lesion segmentation result corresponding to the target third sample CT image is used as the corresponding second sample lesion segmentation result, constituting a set of training data in the second training set.

[0083] The target image segmentation model includes at least one decoding module, at least one encoding module, and a first connection module, wherein the decoding module corresponds to the encoding module one by one, and the corresponding decoding module and the encoding module are connected through a second connection module, and the second connection module is provided with an attention gate;

[0084] Among them, each of the encoding modules, the first connection module and each of the decoding modules are connected in sequence, the input of each of the encoding modules and the first connection module is the output of the previous module, and the input of each decoding module is the output of the previous module and the output of the corresponding second connection module.

[0085] Each of the encoding modules includes a convolutional layer, a batch normalization layer, an activation function and a downsampling layer, wherein the batch normalization layer includes a residual connection operation.

[0086] Among them, the second connection module is connected to the batch normalization layer, and the output result of the batch normalization layer is activated by the attention gate and input into the corresponding decoding module. The attention gate in the second connection module is used to comprehensively calculate the self-attention by combining the parameters of the corresponding encoding module, the corresponding decoding module and the first module, wherein the first module is the previous module of the corresponding encoding module.

[0087] Each of the decoding modules includes an upsampling layer, a convolution layer, a batch normalization layer and an activation function.

[0088] Wherein, after obtaining the lesion segmentation result output by the image segmentation model, the method further includes:

[0089] The features of the lesion are extracted according to the lesion segmentation result by using a pre-built feature extractor.

[0090] Embodiment 3

[0091] The present invention also provides a storage medium storing one or more programs, which can be executed by one or more processors to implement the steps of the CT image segmentation method for bladder cancer described in the above embodiment.

[0092] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit it. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the aforementioned embodiments, or make equivalent replacements for some of the technical features therein. However, these modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A CT image segmentation method suitable for bladder cancer, characterized in that: include: Construct an initial image segmentation model, obtain multiple CT images to annotate lesion segmentation results, and generate the first training set and test set; Using the first training set to train the initial image segmentation model to obtain an intermediate image segmentation model; Constructing a second training set according to the intermediate image segmentation model and the test set; Training the intermediate image segmentation model according to the first training set and the second training set, and obtaining a target image segmentation model after the training is completed; The step of constructing the second training set according to the intermediate image segmentation model and the test set includes: After processing the third sample CT images in the test set into a preset format, obtaining predicted lesion segmentation results of each of the third sample CT images according to the intermediate image segmentation model; For the target third sample image, if the predicted lesion segmentation result of the target third sample image is consistent with the third sample lesion segmentation result corresponding to the target third sample image, the target third sample image is used as the second sample CT image, and the predicted lesion segmentation result corresponding to the target third sample image is used as the corresponding second sample lesion segmentation result to constitute a set of training data in the second training set; The first sample CT image and the third sample CT image in the first training set are different images; Acquire a target CT image, and process the target CT image into a preset format; Inputting the preprocessed target CT image into a target image segmentation model to obtain a lesion segmentation result output by the target image segmentation model; According to the target image segmentation model, obtaining the lesion segmentation results of each third sample CT image in the test set, and performing corresponding comparison with the data in the test set to test the performance of the target image segmentation model; Wherein, the target image segmentation model is obtained by training based on a first training set and a second training set, the first training set includes multiple groups of training data, each group of training data includes a first sample CT image and a first sample lesion segmentation result corresponding to the first sample CT image, the second training set includes multiple groups of training data, each group of training data includes a second sample CT image and a second sample lesion segmentation result corresponding to the second sample CT image, the second sample CT image is selected from a test set used to test the target image segmentation model, the test set includes multiple groups of data, each group of data includes a third sample CT image and a third sample lesion segmentation result corresponding to the third sample CT image.

2. The CT image segmentation method for bladder cancer according to claim 1, characterized in that: The target image segmentation model includes at least one decoding module, at least one encoding module, and a first connection module, wherein the decoding modules correspond to the encoding modules one by one, and the corresponding decoding modules and encoding modules are connected via a second connection module, wherein the second connection module is provided with an attention gate; Among them, each of the encoding modules, the first connection module and each of the decoding modules are connected in sequence, the input of each of the encoding modules and the first connection module is the output of the previous module, and the input of each decoding module is the output of the previous module and the output of the corresponding second connection module.

3. The CT image segmentation method for bladder cancer according to claim 2, characterized in that: Each of the encoding modules includes a convolutional layer, a batch normalization layer, an activation function and a downsampling layer, wherein the batch normalization layer includes a residual connection operation.

4. The CT image segmentation method for bladder cancer according to claim 3, characterized in that: The second connection module is connected to the batch normalization layer, and the output result of the batch normalization layer is activated by the attention gate and then input into the corresponding decoding module. The attention gate in the second connection module is used to activate the corresponding encoding module, the corresponding decoding module and the parameter calculation self-attention of the first module, wherein the first module is the previous module of the corresponding encoding module.

5. The CT image segmentation method for bladder cancer according to claim 2, characterized in that: Each of the decoding modules includes an upsampling layer, a convolution layer, a batch normalization layer and an activation function.

6. The CT image segmentation method for bladder cancer according to claim 1, characterized in that: After obtaining the lesion segmentation result output by the image segmentation model, the method further includes: The features of the lesion are extracted according to the lesion segmentation result by using a pre-built feature extractor.

7. A terminal, characterized in that: The terminal includes: a processor, a storage medium communicatively connected to the processor, the storage medium being suitable for storing a plurality of instructions, and the processor being suitable for calling the instructions in the storage medium to execute the steps of the CT image segmentation method applicable to bladder cancer as described in any one of claims 1-6.

8. A storage medium, characterized in that: The storage medium stores one or more programs, and the one or more programs can be executed by one or more processors to implement the steps of the CT image segmentation method applicable to bladder cancer as described in any one of claims 1-6.

Images