Training Method, Device, Electronic Device, Storage Medium of Gastric Cancer Lesion Segmentation Model, and Gastric Cancer Lesion Segmentation Method

Abstract

The present invention discloses a training method, device, electronic device, storage medium and gastric cancer lesion segmentation method for a gastric cancer lesion segmentation model. The present invention relates to the technical field of image processing. The training method of the gastric cancer lesion segmentation model includes: inputting a three-dimensional computed tomography sample image into a semantic segmentation model to be trained to obtain a gastric lesion segmentation result, wherein the semantic segmentation model includes at least one residual encoding module, at least one bottleneck module and at least one decoding module, and the residual encoding module is jump-connected to the decoding module; determining a target model loss based on the gastric lesion segmentation result and the gastric lesion segmentation label, and updating the model parameters of the semantic segmentation model based on the target model loss until the model training stop condition is satisfied, so as to obtain a trained gastric cancer lesion segmentation model. The above technical solution trains a model capable of automatically segmenting gastric cancer lesions, thereby efficiently and accurately segmenting the gastric cancer lesion area.

Description

Technical Field

[0001] The present invention relates to the technical field of image processing, and particularly to a training method, device, electronic device, storage medium and gastric cancer lesion segmentation method for a gastric cancer lesion segmentation model. Background Art

[0002] The early diagnosis and precise treatment of gastric cancer have become the research focus in the global medical field. Surgery is the main means of treating gastric cancer, and preoperative examination is crucial for the success of the surgery.

[0003] Computed Tomography (CT) can provide comprehensive preoperative image data, clearly showing the size and distribution of gastric cancer lesions, and providing an important basis for doctors to formulate surgical plans.

[0004] Although CT images play an important role in the diagnosis of gastric cancer, most of the current film reading work still relies on the manual analysis of doctors. This method is affected by doctors' experience, and the accuracy of the results is unstable. In addition, the manual recognition process is time-consuming and laborious, and it is difficult to meet the needs of rapid diagnosis. Summary of the Invention

[0005] The present invention provides a training method, device, electronic device, storage medium and gastric cancer lesion segmentation method for a gastric cancer lesion segmentation model, so as to realize the automatic segmentation of gastric cancer lesions and efficiently and accurately segment the gastric cancer lesion area.

[0006] According to one aspect of the present invention, a training method for a gastric cancer lesion segmentation model is provided, including:

[0007] Obtaining three-dimensional computed tomography sample images and gastric lesion segmentation labels corresponding to the three-dimensional computed tomography sample images;

[0008] Inputting the three-dimensional computed tomography sample images into a semantic segmentation model to be trained to obtain a gastric lesion segmentation result, where the semantic segmentation model includes at least one residual encoding module, at least one bottleneck module and at least one decoding module, and the residual encoding module is jump-connected to the decoding module;

[0009] Determining a target model loss based on the gastric lesion segmentation result and the gastric lesion segmentation label, and updating the model parameters of the semantic segmentation model based on the target model loss until a model training stop condition is satisfied, to obtain a trained gastric cancer lesion segmentation model.

[0010] According to another aspect of the present invention, a training device for a gastric cancer lesion segmentation model is provided, including:

[0011] A training sample acquisition module, configured to acquire three-dimensional computed tomography (CT) sample images and gastric lesion segmentation labels corresponding to the three-dimensional CT sample images;

[0012] A gastric lesion segmentation result prediction module, configured to input the three-dimensional CT sample images into a semantic segmentation model to be trained to obtain gastric lesion segmentation results, wherein the semantic segmentation model includes at least one residual encoding module, at least one bottleneck module, and at least one decoding module, and the residual encoding module is skip-connected to the decoding module;

[0013] A model parameter update module, configured to determine a target model loss based on the gastric lesion segmentation results and the gastric lesion segmentation labels, and update the model parameters of the semantic segmentation model based on the target model loss until a model training stop condition is satisfied, thereby obtaining a trained gastric cancer lesion segmentation model.

[0014] According to another aspect of the present invention, there is provided a method for segmenting gastric cancer lesions, including:

[0015] Acquiring a computed tomography image to be segmented;

[0016] Inputting the computed tomography image to be segmented into a trained gastric cancer lesion segmentation model to obtain an initial gastric cancer lesion segmentation result, wherein the gastric cancer lesion segmentation model is obtained by training according to the training method of the gastric cancer lesion segmentation model described in any embodiment of the present invention;

[0017] Performing morphological processing and / or extracting the largest connected component on the initial gastric cancer lesion segmentation result to obtain a target gastric cancer lesion segmentation result.

[0018] According to another aspect of the present invention, there is provided a device for segmenting gastric cancer lesions, including:

[0019] A computed tomography image acquisition module, configured to acquire a computed tomography image to be segmented;

[0020] An initial gastric cancer lesion segmentation result prediction module, configured to input the computed tomography image to be segmented into a trained gastric cancer lesion segmentation model to obtain an initial gastric cancer lesion segmentation result, wherein the gastric cancer lesion segmentation model is obtained by training according to the training method of the gastric cancer lesion segmentation model described in any embodiment of the present invention;

[0021] A target gastric cancer lesion segmentation result determination module, configured to perform morphological processing and / or extract the largest connected component on the initial gastric cancer lesion segmentation result to obtain a target gastric cancer lesion segmentation result.

[0022] According to another aspect of the present invention, there is provided an electronic device, which includes:

[0023] at least one processor;

[0024] and a memory communicatively connected to the at least one processor;

[0025] wherein, the memory stores a computer program executable by the at least one processor, and when the computer program is executed by the at least one processor, the at least one processor is enabled to execute the training method of the gastric cancer lesion segmentation model or the gastric cancer lesion segmentation method according to any embodiment of the present invention.

[0026] According to another aspect of the present invention, there is provided a computer-readable storage medium storing computer instructions for implementing the training method of the gastric cancer lesion segmentation model or the gastric cancer lesion segmentation method according to any embodiment of the present invention when executed by a processor.

[0027] In the technical solution of the embodiment of the present invention, by obtaining three-dimensional computed tomography sample images and gastric lesion segmentation labels corresponding to the three-dimensional computed tomography sample images, and then inputting the three-dimensional computed tomography sample images into a semantic segmentation model to be trained to obtain gastric lesion segmentation results. The semantic segmentation model includes at least one residual encoding module, at least one bottleneck module, and at least one decoding module, and the residual encoding module is jump-connected to the decoding module. Then, based on the gastric lesion segmentation results and the gastric lesion segmentation labels, the target model loss is determined, and based on the target model loss, the model parameters of the semantic segmentation model are updated until the model training stop condition is satisfied, and a trained gastric cancer lesion segmentation model is obtained. The above technical solution optimizes the feature extraction ability of the model through the residual encoding module, effectively improves the accuracy of the gastric cancer lesion segmentation model, and realizes the automatic segmentation of gastric cancer lesions through the gastric cancer lesion segmentation model, efficiently and accurately segmenting the gastric cancer lesion area.

[0028] It should be understood that the content described in this part is not intended to identify the key or important features of the embodiments of the present invention, nor is it used to limit the scope of the present invention. Other features of the present invention will become easily understood through the following description. BRIEF DESCRIPTION OF THE DRAWINGS

[0029] In order to more clearly illustrate the technical solutions in the embodiments of the present invention, the following will briefly introduce the drawings required for the description of the embodiments. Obviously, the drawings in the following description are only some embodiments of the present invention, and those of ordinary skill in the art can obtain other drawings without creative efforts based on these drawings.

[0030] Figure 1 It is a flowchart of a method for training a gastric cancer lesion segmentation model provided in Embodiment 1 of the present invention;

[0031] Figure 2 It is a flowchart of a method for training a gastric cancer lesion segmentation model provided in Embodiment 2 of the present invention;

[0032] Figure 3 It is a flowchart of a method for training a gastric cancer lesion segmentation model provided in Embodiment 3 of the present invention;

[0033] Figure 4 It is a schematic structural diagram of a gastric cancer lesion segmentation model provided in an embodiment of the present invention;

[0034] Figure 5 It is a schematic structural diagram of a first basic residual module provided in an embodiment of the present invention;

[0035] Figure 6 It is a schematic structural diagram of a second basic residual module provided in an embodiment of the present invention;

[0036] Figure 7 It is a flowchart of a method for segmenting gastric cancer lesions provided in Embodiment 4 of the present invention;

[0037] Figure 8 It is a flowchart of a method for segmenting gastric cancer lesions provided in an embodiment of the present invention;

[0038] Figure 9 It is a schematic structural diagram of a training device for a gastric cancer lesion segmentation model provided in Embodiment 5 of the present invention;

[0039] Figure 10 It is a schematic structural diagram of a gastric cancer lesion segmentation device provided in Embodiment 6 of the present invention;

[0040] Figure 11 It is a schematic structural diagram of an electronic device for implementing the method for training a gastric cancer lesion segmentation model in an embodiment of the present invention. Detailed implementation manners

[0041] In order to enable those skilled in the art to better understand the solution of the present invention, the technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are only a part of the embodiments of the present invention, rather than all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative efforts shall fall within the protection scope of the present invention.

[0042] It should be noted that the terms "first", "second", etc. in the description, claims and above-mentioned drawings of the present invention are used to distinguish similar objects, and do not necessarily need to describe a specific order or sequence. It should be understood that the data used in this way can be interchanged under appropriate circumstances, so that the embodiments of the present invention described here can be implemented in an order other than those illustrated or described here. In addition, the terms "comprising" and "having" and any variations thereof are intended to cover non-exclusive inclusion. For example, a process, method, system, product or device comprising a series of steps or units does not necessarily have to be limited to those steps or units clearly listed, but may include other steps or units not clearly listed or inherent to these processes, methods, products or devices. The acquisition, storage, use, processing, etc. of the data in the technical solution of this application all comply with the relevant provisions of national laws and regulations.

[0043] Embodiment 1

[0044] Figure 1 The figure is a flowchart of a training method for a gastric cancer lesion segmentation model provided in Embodiment 1 of the present invention. This embodiment is applicable to the situation of identifying gastric cancer lesions from abdominal CT images. This method can be executed by a training device for a gastric cancer lesion segmentation model. The training device for the gastric cancer lesion segmentation model can be implemented in the form of hardware and / or software, and the training device for the gastric cancer lesion segmentation model can be configured in electronic devices such as terminals and / or servers. As Figure 1 shown, the method includes:

[0045] S110. Obtain three-dimensional computed tomography sample images and gastric lesion segmentation labels corresponding to the three-dimensional computed tomography sample images.

[0046] In the embodiment of the present invention, the three-dimensional computed tomography sample image refers to a three-dimensional CT image used for model training. For example, the three-dimensional computed tomography sample image can be a three-dimensional CT image of the abdominal region or other regions that can cover the gastric region. The gastric lesion segmentation label refers to a segmentation label of the gastric region used for model training. For example, the gastric lesion segmentation label can be a segmentation label of the entire stomach or the gastric lesion region, etc.

[0047] Exemplarily, multiple three-dimensional computed tomography sample images and gastric lesion segmentation labels corresponding to each three-dimensional computed tomography sample image can be read from a preset storage path of an electronic device, and multiple three-dimensional computed tomography sample images and gastric lesion segmentation labels corresponding to each three-dimensional computed tomography sample image can also be obtained from the cloud or other devices communicatively connected to the electronic device. Specific limitations are not made here.

[0048] S120. Input the three-dimensional computed tomography sample image into the semantic segmentation model to be trained to obtain the gastric lesion segmentation result. The semantic segmentation model includes at least one residual encoding module, at least one bottleneck module, and at least one decoding module, and the residual encoding module is jump-connected to the decoding module.

[0049] Among them, the semantic segmentation model is a deep learning model based on a residual encoder, including at least one residual encoding module, at least one bottleneck module, and at least one decoding module. The residual encoding module is jump-connected to the decoding module. The residual encoding module is a residual encoder, which can be composed of one or more residual basic modules. The bottleneck module is a bottleneck layer (Bottleneck) for compressing information and extracting the most representative features. The decoding module is a decoder for reconstructing the compressed data to restore the integrity and accuracy of the original data.

[0050] It should be noted that by introducing the residual encoding module and adding jump connections, the network depth and feature extraction ability of the semantic segmentation model are improved, and the gradient disappearance problem is alleviated at the same time.

[0051] In some optional embodiments, the semantic segmentation model can be a deep learning model improved from nnU-Net. Specifically, the encoder in nnU-Net can be replaced with a residual encoder, and other modules remain unchanged, so as to obtain the semantic segmentation model.

[0052] In the embodiment of the present invention, the gastric lesion segmentation result is the segmentation result of the gastric cancer lesion predicted by the semantic segmentation model, which is the gastric cancer lesion area in the three-dimensional computed tomography sample image.

[0053] S130. Determine the target model loss based on the gastric lesion segmentation result and the gastric lesion segmentation label, and update the model parameters of the semantic segmentation model based on the target model loss until the model training stop condition is satisfied, and obtain the trained gastric cancer lesion segmentation model.

[0054] In the embodiment of the present invention, the target model loss refers to the loss value calculated by the loss function, which can be used to update the model parameters of the semantic segmentation model, realize the supervised training of the semantic segmentation model, and thus obtain the gastric cancer lesion segmentation model.

[0055] Specifically, the target model loss can be calculated by a single loss function or a mixed loss function, which is not specifically limited here. The loss function can include but is not limited to the mean square error loss function, Dice loss function, and cross-entropy loss function, etc.

[0056] In some alternative embodiments, a five-fold cross-validation method can be adopted for model training to ensure a comprehensive evaluation of the performance of the gastric cancer lesion segmentation model.

[0057] In some alternative embodiments, to enhance the global feature extraction ability of the gastric cancer lesion segmentation model, a transfer learning strategy with shared parameters is introduced. Specifically, for convolutional kernels of different sizes, during weight initialization, the pre-trained convolutional kernel parameters are transformed into the corresponding sizes for initialization. For example, the pre-trained parameters of a convolutional kernel with an original size of 3×3 are transformed into those of a 5×5 size using trilinear interpolation to help the large convolutional kernel network overcome the performance saturation problem of limited data common in medical image segmentation.

[0058] The technical solution of the embodiment of the present invention is to obtain a three-dimensional computed tomography sample image and a gastric lesion segmentation label corresponding to the three-dimensional computed tomography sample image, and then input the three-dimensional computed tomography sample image into the semantic segmentation model to be trained to obtain a gastric lesion segmentation result. Among them, the semantic segmentation model includes at least one residual encoding module, at least one bottleneck module, and at least one decoding module. The residual encoding module is jump-connected to the decoding module. Then, based on the gastric lesion segmentation result and the gastric lesion segmentation label, the target model loss is determined, and the model parameters of the semantic segmentation model are updated based on the target model loss until the model training stop condition is met, and a trained gastric cancer lesion segmentation model is obtained. The above technical solution optimizes the feature extraction ability of the model through the residual encoding module, effectively improving the accuracy of the gastric cancer lesion segmentation model. The automatic segmentation of gastric cancer lesions is realized through the gastric cancer lesion segmentation model, and the gastric cancer lesion area is segmented efficiently and accurately.

[0059] Embodiment 2

[0060] Figure 2The figure is a flowchart of a method for training a gastric cancer lesion segmentation model provided in the second embodiment of the present invention. The method in this embodiment can be combined with each optional solution in the method for training the gastric cancer lesion segmentation model provided in the above embodiment. The method for training the gastric cancer lesion segmentation model provided in this embodiment is further optimized. Optionally, the obtaining of the three-dimensional computed tomography sample image and the gastric lesion segmentation label corresponding to the three-dimensional computed tomography sample image includes: obtaining a three-dimensional computed tomography image, performing a preprocessing operation on the three-dimensional computed tomography image to obtain a preprocessed three-dimensional computed tomography image, where the preprocessing operation includes: performing spatial redirection on the three-dimensional computed tomography image; performing resampling on the three-dimensional computed tomography image; performing foreground segmentation on the three-dimensional computed tomography image; inputting the preprocessed three-dimensional computed tomography image into a pre-trained gastric lesion segmentation model to obtain a gastric lesion segmentation label; obtaining a pre-designed computed tomography value range interval, and cropping the preprocessed three-dimensional computed tomography image based on the pre-designed computed tomography value range interval to obtain a cropped three-dimensional computed tomography image; performing voxel intensity normalization on the cropped three-dimensional computed tomography image to obtain a three-dimensional computed tomography sample image.

[0061] As Figure 2 shown, the method includes:

[0062] S210. Obtain a three-dimensional computed tomography image, perform a preprocessing operation on the three-dimensional computed tomography image to obtain a preprocessed three-dimensional computed tomography image, where the preprocessing operation includes: performing spatial redirection on the three-dimensional computed tomography image; performing resampling on the three-dimensional computed tomography image; performing foreground segmentation on the three-dimensional computed tomography image.

[0063] Among them, the three-dimensional computed tomography image can be a three-dimensional CT image that meets the inclusion criteria.

[0064] It should be noted that spatial redirection is used to adjust the direction of the three-dimensional CT image to ensure the consistency of multiple three-dimensional CT images. Resampling is used to adjust the spatial resolution of the three-dimensional CT image, thereby ensuring the consistency of multiple three-dimensional CT images. Foreground segmentation is used to extract the foreground region of the stomach and surrounding tissues in the three-dimensional CT image, thereby reducing the interference of irrelevant backgrounds on model training.

[0065] S220. Input the preprocessed three-dimensional computed tomography image into a pre-trained gastric lesion segmentation model to obtain a gastric lesion segmentation label.

[0066] Among them, the pre-trained gastric lesion segmentation model can be a pre-trained model obtained by self-supervised training with a large amount of data in an open-source project. For example, the pre-trained gastric lesion segmentation model can be a U-Net or nnU-Net model, etc., which is not specifically limited here.

[0067] In some alternative embodiments, after obtaining the gastric lesion segmentation labels, the gastric lesion segmentation labels can be inspected and corrected to ensure the accuracy and consistency of the gastric lesion segmentation labels.

[0068] S230. Obtain a pre-designed computed tomography value range interval, and crop the pre-processed three-dimensional computed tomography image based on the pre-designed computed tomography value range interval to obtain a cropped three-dimensional computed tomography image.

[0069] Among them, the pre-designed computed tomography value range interval is a pre-set effective CT value range.

[0070] It should be noted that by cropping the pre-processed three-dimensional computed tomography image with the pre-designed computed tomography value range interval, the effective CT value range of the pre-processed three-dimensional CT image can be intercepted, abnormal CT values and artifacts can be removed, and the enhancement of the three-dimensional CT image is realized.

[0071] S240. Perform voxel intensity normalization on the cropped three-dimensional computed tomography image to obtain a three-dimensional computed tomography sample image.

[0072] Specifically, mapping the voxel values of the cropped three-dimensional CT image to a unified range can eliminate the intensity differences between three-dimensional CT images, thereby providing a standardized and high-quality three-dimensional computed tomography sample image for model training.

[0073] S250. Input the three-dimensional computed tomography sample image into the semantic segmentation model to be trained to obtain a gastric lesion segmentation result. Among them, the semantic segmentation model includes at least one residual encoding module, at least one bottleneck module, and at least one decoding module, and the residual encoding module is jump-connected to the decoding module.

[0074] S260. Determine the target model loss based on the gastric lesion segmentation result and the gastric lesion segmentation label, and update the model parameters of the semantic segmentation model based on the target model loss until the model training stop condition is met, and obtain a trained gastric cancer lesion segmentation model.

[0075] In the technical solution of the embodiment of the present invention, by acquiring a three-dimensional computed tomography (CT) image, performing a preprocessing operation on the three-dimensional CT image to obtain a preprocessed three-dimensional CT image, then inputting the preprocessed three-dimensional CT image into a pre-trained gastric lesion segmentation model to obtain a gastric lesion segmentation label, further obtaining a pre-designed CT value range interval, cropping the preprocessed three-dimensional CT image based on the pre-designed CT value range interval to obtain a cropped three-dimensional CT image, and then performing voxel intensity normalization on the cropped three-dimensional CT image to obtain a three-dimensional CT sample image. The above technical solution provides a high-quality data set for model training through operations such as data acquisition, preprocessing, image annotation, image enhancement, and image standardization.

[0076] Embodiment III

[0077] Figure 3The flowchart of a training method for a gastric cancer lesion segmentation model provided in Embodiment 3 of the present invention. The method of this embodiment can be combined with each optional solution in the training method of the gastric cancer lesion segmentation model provided in the above embodiments. The training method of the gastric cancer lesion segmentation model provided in this embodiment is further optimized. Optionally, the step of inputting the three-dimensional computed tomography sample image into the semantic segmentation model to be trained to obtain the gastric lesion segmentation result includes: inputting the three-dimensional computed tomography sample image into the first residual encoding module of the semantic segmentation model to be trained to obtain the first computed tomography encoded feature image; inputting the first computed tomography encoded feature image into the second residual encoding module of the semantic segmentation model to be trained to obtain the second computed tomography encoded feature image; inputting the second computed tomography encoded feature image into the third residual encoding module of the semantic segmentation model to be trained to obtain the third computed tomography encoded feature image; inputting the third computed tomography encoded feature image into the fourth residual encoding module of the semantic segmentation model to be trained to obtain the fourth computed tomography encoded feature image; inputting the fourth computed tomography encoded feature image into at least one bottleneck module of the semantic segmentation model to be trained to obtain the computed tomography bottleneck feature image; inputting the computed tomography bottleneck feature image and the fourth computed tomography encoded feature image into the fourth decoding module of the semantic segmentation model to be trained to obtain the fourth computed tomography decoded feature image; inputting the fourth computed tomography decoded feature image and the third computed tomography encoded feature image into the third decoding module of the semantic segmentation model to be trained to obtain the third computed tomography decoded feature image; inputting the third computed tomography decoded feature image and the second computed tomography encoded feature image into the second decoding module of the semantic segmentation model to be trained to obtain the second computed tomography decoded feature image; inputting the second computed tomography decoded feature image and the first computed tomography encoded feature image into the first decoding module of the semantic segmentation model to be trained to obtain the gastric lesion segmentation result.

[0078] As Figure 3 shown, the method includes:

[0079] S301. Obtain a three-dimensional computed tomography sample image and the gastric lesion segmentation label corresponding to the three-dimensional computed tomography sample image.

[0080] S302. Input the three-dimensional computed tomography sample image into the first residual encoding module of the semantic segmentation model to be trained to obtain the first computed tomography encoded feature image.

[0081] S303. Input the first computed tomography encoded feature image into the second residual encoding module of the semantic segmentation model to be trained, and obtain a second computed tomography encoded feature image.

[0082] S304. Input the second computed tomography encoded feature image into the third residual encoding module of the semantic segmentation model to be trained, and obtain a third computed tomography encoded feature image.

[0083] S305. Input the third computed tomography encoded feature image into the fourth residual encoding module of the semantic segmentation model to be trained, and obtain a fourth computed tomography encoded feature image.

[0084] S306. Input the fourth computed tomography encoded feature image into at least one bottleneck module of the semantic segmentation model to be trained, and obtain a computed tomography bottleneck feature image.

[0085] S307. Input the computed tomography bottleneck feature image and the fourth computed tomography encoded feature image into the fourth decoding module of the semantic segmentation model to be trained, and obtain a fourth computed tomography decoded feature image.

[0086] S308. Input the fourth computed tomography decoded feature image and the third computed tomography encoded feature image into the third decoding module of the semantic segmentation model to be trained, and obtain a third computed tomography decoded feature image.

[0087] S309. Input the third computed tomography decoded feature image and the second computed tomography encoded feature image into the second decoding module of the semantic segmentation model to be trained, and obtain a second computed tomography decoded feature image.

[0088] S310. Input the second computed tomography decoded feature image and the first computed tomography encoded feature image into the first decoding module of the semantic segmentation model to be trained, and obtain a gastric lesion segmentation result.

[0089] S311. Determine a target model loss based on the gastric lesion segmentation result and the gastric lesion segmentation label, and update the model parameters of the semantic segmentation model based on the target model loss until a model training stop condition is satisfied, and obtain a trained gastric cancer lesion segmentation model.

[0090] In the embodiment of the present invention, Figure 4It is a schematic structural diagram of a gastric cancer lesion segmentation model provided according to an embodiment of the present invention. Among them, the first residual encoding module is skip-connected to the first decoding module, the second residual encoding module is skip-connected to the second decoding module, the third residual encoding module is skip-connected to the third decoding module, and the fourth residual encoding module is skip-connected to the fourth decoding module. The number of bottleneck modules can be one or more, and no specific limitation is made here.

[0091] It should be noted that through the four residual encoding modules, the bottleneck module, and the four decoding modules skip-connected to the residual encoding modules, the network depth and feature extraction ability are improved, the gradient disappearance problem is alleviated, and the segmentation accuracy of the gastric cancer lesion segmentation model is effectively improved.

[0092] Optionally, the residual encoding module includes: a first basic residual module and at least one second basic residual module. The first basic residual module and the second basic residual module are concatenated. The first basic residual module includes a first convolutional layer, a second convolutional layer, a third convolutional layer, and a downsampling layer. The second basic residual module includes a fourth convolutional layer, a fifth convolutional layer, and a sixth convolutional layer.

[0093] In the embodiment of the present invention, Figure 5 is a schematic structural diagram of a first basic residual module provided according to an embodiment of the present invention; Figure 6 is a schematic structural diagram of a second basic residual module provided according to an embodiment of the present invention.

[0094] Exemplarily, the first convolutional layer can be a 1×1 convolution, the second convolutional layer can be a 3×3 convolution, the third convolutional layer can be a 1×1 convolution, and the downsampling layer can be a 1×1 convolution. The fourth convolutional layer can be a 1×1 convolution, the fifth convolutional layer can be a 3×3 convolution, and the sixth convolutional layer can be a 1×1 convolution. One or more second basic residual modules can be concatenated after the first basic residual module, and normalization processing and non-linear processing can also be included after each layer of the first basic residual module and the second basic residual module.

[0095] Optionally, determining the target model loss based on the gastric lesion segmentation result and the gastric lesion segmentation label includes: determining the segmentation task loss based on the gastric lesion segmentation result and the gastric lesion segmentation label; determining the classification task loss based on the fourth computed tomography encoded feature image and the gastric lesion segmentation label; determining the target depth supervision loss based on the computed tomography bottleneck feature image, the fourth computed tomography decoded feature image, the third computed tomography decoded feature image, the second computed tomography decoded feature image, and the gastric lesion segmentation label; and determining the target model loss based on the segmentation task loss, the classification task loss, and the target depth supervision loss.

[0096] In the embodiments of the present invention, the segmentation task loss includes Dice loss and cross-entropy loss, and the calculation formula of the segmentation task loss can be as follows:

[0097] ;

[0098] ;

[0099] ;

[0100] ;

[0101] ;

[0102] wherein, represents the segmentation result of the gastric lesion corresponding to the i-th three-dimensional computed tomography sample image, represents the segmentation label of the gastric lesion corresponding to the i-th three-dimensional computed tomography sample image; is a hyperparameter, which can be 1 or other values, represents the Dice loss, represents the cross-entropy loss, represents the segmentation task loss, and N represents the number of three-dimensional computed tomography sample images. It should be noted that by mixing the Dice loss and the cross-entropy loss, the integrity of the region segmentation and the classification accuracy are taken into account.

[0103] The calculation formula of the classification task loss can be as follows:

[0104] ;

[0105] wherein, represents the fourth computed tomography encoded feature image corresponding to the i-th three-dimensional computed tomography sample image, represents the classification task loss.

[0106] The target depth supervision loss is the loss calculated according to the depth supervision strategy. The calculation formula of the target model loss can be:

[0107] ;

[0108] wherein, represents the target model loss, represents the target depth supervision loss.

[0109] Optionally, determining the target depth supervision loss based on the computed tomography bottleneck feature image, the fourth computed tomography decoding feature image, the third computed tomography decoding feature image, the second computed tomography decoding feature image, and the gastric lesion segmentation label includes: determining a first depth supervision loss based on the computed tomography bottleneck feature image and the gastric lesion segmentation label; determining a second depth supervision loss based on the fourth computed tomography decoding feature image and the gastric lesion segmentation label; determining a third depth supervision loss based on the third computed tomography decoding feature image and the gastric lesion segmentation label; determining a fourth depth supervision loss based on the second computed tomography decoding feature image and the gastric lesion segmentation label; determining the target depth supervision loss based on the first depth supervision loss, the second depth supervision loss, the third depth supervision loss, and the fourth depth supervision loss.

[0110] In an embodiment of the present invention, the calculation formula of the target depth supervision loss may be as follows:

[0111] ;

[0112] ;

[0113] ;

[0114] ;

[0115] ;

[0116] wherein, represents the first depth supervision loss, represents the second depth supervision loss, represents the third depth supervision loss, represents the fourth depth supervision loss, represents the computed tomography bottleneck feature image corresponding to the i-th three-dimensional computed tomography sample image, represents the fourth computed tomography decoding feature image corresponding to the i-th three-dimensional computed tomography sample image, represents the third computed tomography decoding feature image corresponding to the i-th three-dimensional computed tomography sample image, represents the second computed tomography decoding feature image corresponding to the i-th three-dimensional computed tomography sample image.

[0117] It should be noted that by adopting a deep supervision strategy, supervision is imposed on each layer in the upsampling stage, and the loss is calculated layer by layer, enabling the model to learn more diverse feature representations at different resolutions and scales. This not only improves the discriminability of intermediate layer features but also enhances the diversity of global features. In addition, it also enhances the adaptability of the model to changes in input data and shows higher robustness to uncertainties such as noise and data loss.

[0118] The technical solution of the embodiment of the present invention improves the network depth and feature extraction ability through four residual encoding modules, a bottleneck module, and four decoding modules that are skip-connected to the residual encoding modules, alleviates the problem of gradient disappearance, and effectively improves the accuracy of the gastric cancer lesion segmentation model.

[0119] Embodiment 4

[0120] Figure 7 FIG. is a flowchart of a gastric cancer lesion segmentation method provided by Embodiment 4 of the present invention. This embodiment is applicable to the situation of identifying gastric cancer lesions in abdominal CT images. This method can be executed by a gastric cancer lesion segmentation device, which can be implemented in the form of hardware and / or software, and the gastric cancer lesion segmentation device can be configured in a computer terminal. As Figure 7 shown, the method includes:

[0121] S410. Obtain a computer tomography image to be segmented.

[0122] Among them, the computer tomography image to be segmented can be a CT image of a new patient.

[0123] S420. Input the computer tomography image to be segmented into the trained gastric cancer lesion segmentation model to obtain an initial gastric cancer lesion segmentation result, where the gastric cancer lesion segmentation model is obtained after being trained according to the training method of the gastric cancer lesion segmentation model described in any embodiment of the present invention.

[0124] S430. Perform morphological processing and / or maximum connected component extraction on the initial gastric cancer lesion segmentation result to obtain a target gastric cancer lesion segmentation result.

[0125] Among them, morphological processing can fill the holes in the initial gastric cancer lesion segmentation result through closing operation and remove artifacts through opening operation to optimize the coherence of the segmentation area. Maximum connected component extraction can retain the largest connected area and filter out isolated areas.

[0126] Exemplarily, Figure 8 is a flowchart of a gastric cancer lesion segmentation method provided by an embodiment of the present invention. The gastric cancer lesion segmentation method includes:

[0127] First, preprocess the CT images that meet the inclusion criteria to obtain the preprocessed CT images. The preprocessing includes spatial redirection, resampling, foreground segmentation, etc.

[0128] Second, use the pre-trained gastric lesion segmentation model to automatically label the preprocessed CT images to obtain the gastric lesion segmentation labels.

[0129] Third, determine whether the labeling is incorrect. In the case of incorrect labeling, manually correct the gastric lesion segmentation labels; in the case of correct labeling, proceed to the fourth step.

[0130] Fourth, construct a semantic segmentation model based on the residual encoding module. The semantic segmentation model includes at least one residual encoding module, at least one bottleneck module, and at least one decoding module. The residual encoding module is skip-connected to the decoding module.

[0131] Fifth, perform multi-task deep supervision training. Specifically, by adopting a deep supervision strategy, supervision is imposed on each layer in the upsampling stage, and the loss is calculated layer by layer, enabling the model to learn more diverse feature representations at different resolutions and scales. This not only improves the discriminability of the intermediate layer features but also enhances the diversity of the global features. In addition, it enhances the adaptability of the model to changes in the input data and shows higher robustness to uncertainties such as noise and data loss.

[0132] Sixth, model application and post-processing. Input the computer tomography image to be segmented into the trained gastric cancer lesion segmentation model to obtain the initial gastric cancer lesion segmentation result, and then perform post-processing operations such as morphological processing and / or maximum connected component extraction on the initial gastric cancer lesion segmentation result to obtain the high-precision target gastric cancer lesion segmentation result.

[0133] The technical solution of the embodiment of the present invention realizes the automatic segmentation of gastric cancer lesions through the gastric cancer lesion segmentation model, thereby efficiently and accurately segmenting the gastric cancer lesion area.

[0134] Embodiment Five

[0135] Figure 9 It is a schematic structural diagram of a training device for a gastric cancer lesion segmentation model provided in Embodiment Five of the present invention. As Figure 9 shown, the device includes:

[0136] A training sample acquisition module 510, configured to acquire three-dimensional computer tomography sample images and the corresponding gastric lesion segmentation labels of the three-dimensional computer tomography sample images;

[0137] The gastric lesion segmentation result prediction module 520 is configured to input the three-dimensional computed tomography sample image into a semantic segmentation model to be trained, and obtain a gastric lesion segmentation result. The semantic segmentation model includes at least one residual encoding module, at least one bottleneck module, and at least one decoding module, and the residual encoding module is jump-connected to the decoding module;

[0138] The model parameter update module 530 is configured to determine a target model loss based on the gastric lesion segmentation result and the gastric lesion segmentation label, and update the model parameters of the semantic segmentation model based on the target model loss until a model training stop condition is satisfied, so as to obtain a trained gastric cancer lesion segmentation model.

[0139] The technical solution of the embodiment of the present invention is to obtain a three-dimensional computed tomography sample image and a gastric lesion segmentation label corresponding to the three-dimensional computed tomography sample image, and then input the three-dimensional computed tomography sample image into a semantic segmentation model to be trained to obtain a gastric lesion segmentation result. The semantic segmentation model includes at least one residual encoding module, at least one bottleneck module, and at least one decoding module, and the residual encoding module is jump-connected to the decoding module. Then, a target model loss is determined based on the gastric lesion segmentation result and the gastric lesion segmentation label, and the model parameters of the semantic segmentation model are updated based on the target model loss until a model training stop condition is satisfied, so as to obtain a trained gastric cancer lesion segmentation model. Through the above technical solution, the feature extraction ability of the model is optimized by the residual encoding module, effectively improving the accuracy of the gastric cancer lesion segmentation model. The automatic segmentation of gastric cancer lesions is realized through the gastric cancer lesion segmentation model, and the gastric cancer lesion area is segmented efficiently and accurately.

[0140] In some alternative embodiments, the training sample acquisition module 510 may specifically be configured to:

[0141] Obtain a three-dimensional computed tomography image, perform preprocessing operations on the three-dimensional computed tomography image to obtain a preprocessed three-dimensional computed tomography image. The preprocessing operations include: performing spatial redirection on the three-dimensional computed tomography image; performing resampling on the three-dimensional computed tomography image; performing foreground segmentation on the three-dimensional computed tomography image;

[0142] Input the preprocessed three-dimensional computed tomography image into a pre-trained gastric lesion segmentation model to obtain a gastric lesion segmentation label;

[0143] Obtain a pre-designed computed tomography value range interval, and crop the preprocessed three-dimensional computed tomography image based on the pre-designed computed tomography value range interval to obtain a cropped three-dimensional computed tomography image;

[0144] Perform voxel intensity normalization on the cropped three-dimensional computed tomography (CT) image to obtain a three-dimensional CT sample image.

[0145] In some alternative embodiments, the gastric lesion segmentation result prediction module 520 may specifically be configured to:

[0146] Input the three-dimensional CT sample image into the first residual encoding module of the semantic segmentation model to be trained, and obtain a first CT encoding feature image;

[0147] Input the first CT encoding feature image into the second residual encoding module of the semantic segmentation model to be trained, and obtain a second CT encoding feature image;

[0148] Input the second CT encoding feature image into the third residual encoding module of the semantic segmentation model to be trained, and obtain a third CT encoding feature image;

[0149] Input the third CT encoding feature image into the fourth residual encoding module of the semantic segmentation model to be trained, and obtain a fourth CT encoding feature image;

[0150] Input the fourth CT encoding feature image into at least one bottleneck module of the semantic segmentation model to be trained, and obtain a CT bottleneck feature image;

[0151] Input the CT bottleneck feature image and the fourth CT encoding feature image into the fourth decoding module of the semantic segmentation model to be trained, and obtain a fourth CT decoding feature image;

[0152] Input the fourth CT decoding feature image and the third CT encoding feature image into the third decoding module of the semantic segmentation model to be trained, and obtain a third CT decoding feature image;

[0153] Input the third CT decoding feature image and the second CT encoding feature image into the second decoding module of the semantic segmentation model to be trained, and obtain a second CT decoding feature image;

[0154] Input the second CT decoding feature image and the first CT encoding feature image into the first decoding module of the semantic segmentation model to be trained, and obtain the gastric lesion segmentation result.

[0155] In some alternative embodiments, the residual encoding module includes: a first basic residual module and at least one second basic residual module, the first basic residual module and the second basic residual module are concatenated, the first basic residual module includes a first convolutional layer, a second convolutional layer, a third convolutional layer and a downsampling layer, and the second basic residual module includes a fourth convolutional layer, a fifth convolutional layer and a sixth convolutional layer.

[0156] In some alternative embodiments, the model parameter update module 530 includes:

[0157] A segmentation task loss determination unit, configured to determine a segmentation task loss based on the gastric lesion segmentation result and the gastric lesion segmentation label;

[0158] A classification task loss determination unit, configured to determine a classification task loss based on the fourth computed tomography encoded feature image and the gastric lesion segmentation label;

[0159] A depth supervision loss determination unit, configured to determine a target depth supervision loss based on the computed tomography bottleneck feature image, the fourth computed tomography decoded feature image, the third computed tomography decoded feature image, the second computed tomography decoded feature image, and the gastric lesion segmentation label;

[0160] A target model loss determination unit, configured to determine a target model loss based on the segmentation task loss, the classification task loss, and the target depth supervision loss.

[0161] In some alternative embodiments, the depth supervision loss determination unit may specifically be configured to:

[0162] Determine a first depth supervision loss based on the computed tomography bottleneck feature image and the gastric lesion segmentation label;

[0163] Determine a second depth supervision loss based on the fourth computed tomography decoded feature image and the gastric lesion segmentation label;

[0164] Determine a third depth supervision loss based on the third computed tomography decoded feature image and the gastric lesion segmentation label;

[0165] Determine a fourth depth supervision loss based on the second computed tomography decoded feature image and the gastric lesion segmentation label;

[0166] Determine a target depth supervision loss based on the first depth supervision loss, the second depth supervision loss, the third depth supervision loss, and the fourth depth supervision loss.

[0167] The training device for the gastric cancer lesion segmentation model provided by the embodiments of the present invention can execute the training method for the gastric cancer lesion segmentation model provided by any embodiment of the present invention, and has the corresponding functional modules and beneficial effects for executing the method.

[0168] Embodiment Six

[0169] Figure 10 FIG. is a schematic structural diagram of a gastric cancer lesion segmentation device provided by Embodiment Six of the present invention. As Figure 10 shown, the device includes:

[0170] A computed tomography (CT) image acquisition module 610, configured to acquire a CT image to be segmented;

[0171] An initial gastric cancer lesion segmentation result prediction module 620, configured to input the CT image to be segmented into a trained gastric cancer lesion segmentation model to obtain an initial gastric cancer lesion segmentation result, where the gastric cancer lesion segmentation model is obtained by training according to the training method for the gastric cancer lesion segmentation model described in any embodiment of the present invention;

[0172] A target gastric cancer lesion segmentation result determination module 630, configured to perform morphological processing and / or maximum connected component extraction on the initial gastric cancer lesion segmentation result to obtain a target gastric cancer lesion segmentation result.

[0173] The technical solution of the embodiment of the present invention realizes the automatic segmentation of gastric cancer lesions through the gastric cancer lesion segmentation model, thereby efficiently and accurately segmenting the gastric cancer lesion area.

[0174] The gastric cancer lesion segmentation device provided by the embodiments of the present invention can execute the gastric cancer lesion segmentation method provided by any embodiment of the present invention, and has the corresponding functional modules and beneficial effects for executing the method.

[0175] Embodiment Seven

[0176] Figure 11 FIG. shows a schematic structural diagram of an electronic device 10 that can be used to implement the embodiments of the present invention. The electronic device is intended to represent various forms of digital computers, such as, a laptop computer, a desktop computer, a workbench, a personal digital assistant, a server, a blade server, a mainframe computer, and other suitable computers. The electronic device can also represent various forms of mobile devices, such as, a personal digital assistant, a cellular phone, a smart phone, a wearable device (such as a helmet, glasses, a watch, etc.) and other similar computing devices. The components shown herein, their connections and relationships, and their functions are merely examples and are not intended to limit the implementation of the present invention described and / or claimed herein.

[0177] As Figure 11As shown, the electronic device 10 includes at least one processor 11 and a memory communicatively connected to the at least one processor 11, such as read-only memory (ROM) 12, random access memory (RAM) 13, etc. Among them, the memory stores a computer program executable by the at least one processor. The processor 11 can execute various appropriate actions and processes according to the computer program stored in the read-only memory (ROM) 12 or the computer program loaded from the storage unit 18 into the random access memory (RAM) 13. In the RAM 13, various programs and data required for the operation of the electronic device 10 can also be stored. The processor 11, ROM 12, and RAM 13 are connected to each other through a bus 14. The I / O interface 15 is also connected to the bus 14.

[0178] Multiple components in the electronic device 10 are connected to the I / O interface 15, including: an input unit 16, such as a keyboard, a mouse, etc.; an output unit 17, such as various types of displays, speakers, etc.; a storage unit 18, such as a disk, an optical disc, etc.; and a communication unit 19, such as a network card, a modem, a wireless communication transceiver, etc. The communication unit 19 allows the electronic device 10 to exchange information / data with other devices through a computer network such as the Internet and / or various telecommunication networks.

[0179] The processor 11 can be various general-purpose and / or special-purpose processing components with processing and computing capabilities. Some examples of the processor 11 include but are not limited to a central processing unit (CPU), a graphics processing unit (GPU), various dedicated artificial intelligence (AI) computing chips, various processors running machine learning model algorithms, a digital signal processor (DSP), and any appropriate processor, controller, microcontroller, etc. The processor 11 executes the various methods and processes described above, such as the training method of the gastric cancer lesion segmentation model, and the method includes:

[0180] Obtain a three-dimensional computed tomography sample image and a gastric lesion segmentation label corresponding to the three-dimensional computed tomography sample image;

[0181] Input the three-dimensional computed tomography sample image into a semantic segmentation model to be trained to obtain a gastric lesion segmentation result, where the semantic segmentation model includes at least one residual encoding module, at least one bottleneck module, and at least one decoding module, and the residual encoding module is jump-connected to the decoding module;

[0182] Determine a target model loss based on the gastric lesion segmentation result and the gastric lesion segmentation label, and update the model parameters of the semantic segmentation model based on the target model loss until a model training stop condition is satisfied to obtain a trained gastric cancer lesion segmentation model.

[0183] In some embodiments, the training method or the gastric cancer lesion segmentation method of the gastric cancer lesion segmentation model can be implemented as a computer program, which is tangibly contained in a computer-readable storage medium, such as storage unit 18. In some embodiments, part or all of the computer program can be loaded and / or installed onto the electronic device 10 via the ROM 12 and / or the communication unit 19. When the computer program is loaded into the RAM 13 and executed by the processor 11, one or more steps of the training method or the gastric cancer lesion segmentation method described above can be executed. Alternatively, in other embodiments, the processor 11 can be configured to execute the training method or the gastric cancer lesion segmentation method by any other suitable means (e.g., by means of firmware).

[0184] The various embodiments of the systems and techniques described above in this document can be implemented in digital electronic circuitry, integrated circuit systems, field-programmable gate arrays (FPGA), application-specific integrated circuits (ASIC), application-specific standard products (ASSP), system-on-a-chip (SOC), complex programmable logic devices (CPLD), computer hardware, firmware, software, and / or combinations thereof. These various embodiments can include: being implemented in one or more computer programs that can be executed and / or interpreted on a programmable system including at least one programmable processor, which can be a dedicated or general-purpose programmable processor that receives data and instructions from a storage system, at least one input device, and at least one output device, and transmits the data and instructions to the storage system, the at least one input device, and the at least one output device.

[0185] The computer programs for implementing the methods of the present invention can be written in any combination of one or more programming languages. These computer programs can be provided to the processor of a general-purpose computer, a dedicated computer, or other programmable data processing devices, such that when the computer programs are executed by the processor, the functions / operations specified in the flowcharts and / or block diagrams are implemented. The computer programs can be executed entirely on the machine, partially on the machine, as a stand-alone software package partially on the machine and partially on a remote machine, or entirely on a remote machine or server.

[0186] In the context of the present invention, a computer-readable storage medium can be a tangible medium that can contain or store a computer program for use by or in connection with an instruction execution system, apparatus, or device. The computer-readable storage medium can include, but is not limited to, electronic, magnetic, optical, electromagnetic, infrared, or semiconductor systems, apparatus, or devices, or any suitable combination of the foregoing. Alternatively, the computer-readable storage medium can be a machine-readable signal medium. More specific examples of the machine-readable storage medium would include an electrical connection based on one or more wires, a portable computer disk, a hard disk, a random access memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or Flash memory), an optical fiber, a portable compact disc read-only memory (CD-ROM), an optical storage device, a magnetic storage device, or any suitable combination of the foregoing.

[0187] To provide for interaction with a user, the systems and techniques described herein can be implemented on an electronic device having: a display device (e.g., a CRT (cathode ray tube) or LCD (liquid crystal display) monitor) for displaying information to the user; and a keyboard and a pointing device (e.g., a mouse or a trackball) by which the user can provide input to the electronic device. Other kinds of devices can also be used to provide for interaction with the user; for example, feedback provided to the user can be any form of sensory feedback (e.g., visual feedback, auditory feedback, or tactile feedback); and input from the user can be received in any form (including acoustic input, voice input, or tactile input).

[0188] The systems and techniques described herein can be implemented in a computing system that includes backend components (e.g., as a data server), or a computing system that includes middleware components (e.g., an application server), or a computing system that includes frontend components (e.g., a user computer having a graphical user interface or a web browser through which the user can interact with an implementation of the systems and techniques described herein), or a computing system that includes any combination of such backend components, middleware components, or frontend components. The components of the system can be interconnected by any form or medium of digital data communication (e.g., a communication network). Examples of communication networks include: a local area network (LAN), a wide area network (WAN), a blockchain network, and the Internet.

[0189] A computing system may include a client and a server. The client and the server are generally far from each other and usually interact via a communication network. The client-server relationship is created by computer programs running on respective computers and having a client-server relationship with each other. The server may be a cloud server, also known as a cloud computing server or a cloud host, which is a host product in the cloud computing service system, solving the defects of difficult management and weak business scalability existing in traditional physical hosts and VPS services.

[0190] It should be understood that various forms of the processes shown above can be used, steps can be reordered, added or deleted. For example, the steps described in the present invention can be executed in parallel, sequentially or in different orders, as long as the desired results of the technical solution of the present invention can be achieved, and no limitations are imposed herein.

[0191] The above specific embodiments do not constitute a limitation on the protection scope of the present invention. Those skilled in the art should understand that various modifications, combinations, sub-combinations and substitutions can be made according to design requirements and other factors. Any modifications, equivalent substitutions and improvements made within the spirit and principle of the present invention shall be included within the protection scope of the present invention.

Claims

1. A training method for a gastric cancer lesion segmentation model, characterized in that: Applicable to the identification of gastric cancer lesions in abdominal CT images, including: Acquire a three-dimensional computed tomography sample image and a gastric lesion segmentation label corresponding to the three-dimensional computed tomography sample image; During weight initialization, a 3×3 convolution kernel parameter is converted into a 5×5 pre-trained parameter using a three-way linear interpolation method; the three-dimensional computed tomography sample image is input into a semantic segmentation model to be trained to obtain a gastric lesion segmentation result, wherein the semantic segmentation model includes at least one residual encoding module, at least one bottleneck module and at least one decoding module, and the residual encoding module is jump-connected with the decoding module; Determining a target model loss based on the gastric lesion segmentation result and the gastric lesion segmentation label, and updating a model parameter of the semantic segmentation model based on the target model loss until a model training stop condition is met, thereby obtaining a trained gastric cancer lesion segmentation model; The step of inputting the three-dimensional computer tomography sample image into the semantic segmentation model to be trained to obtain a gastric lesion segmentation result includes: Inputting the three-dimensional computed tomography sample image into a first residual coding module of a semantic segmentation model to be trained to obtain a first computed tomography coding feature image; Inputting the first computed tomography encoding feature image into a second residual coding module of a semantic segmentation model to be trained to obtain a second computed tomography encoding feature image; Inputting the second computed tomography encoding feature image into a third residual coding module of the semantic segmentation model to be trained to obtain a third computed tomography encoding feature image; Inputting the third computed tomography encoding feature image into a fourth residual coding module of the semantic segmentation model to be trained to obtain a fourth computed tomography encoding feature image; Inputting the fourth computed tomography encoding feature image into at least one bottleneck module of the semantic segmentation model to be trained to obtain a computed tomography bottleneck feature image; Inputting the computed tomography bottleneck feature image and the fourth computed tomography encoding feature image into a fourth decoding module of the semantic segmentation model to be trained to obtain a fourth computed tomography decoding feature image; Inputting the fourth computed tomography decoding feature image and the third computed tomography encoding feature image into a third decoding module of the semantic segmentation model to be trained to obtain a third computed tomography decoding feature image; Inputting the third computed tomography decoding feature image and the second computed tomography encoding feature image into a second decoding module of a semantic segmentation model to be trained to obtain a second computed tomography decoding feature image; Inputting the second computed tomography decoding feature image and the first computed tomography encoding feature image into a first decoding module of a semantic segmentation model to be trained to obtain a gastric lesion segmentation result; The determining the target model loss based on the gastric lesion segmentation result and the gastric lesion segmentation label comprises: determining a segmentation task loss based on the gastric lesion segmentation result and the gastric lesion segmentation label; determining a classification task loss based on the fourth computed tomography encoded feature image and the gastric lesion segmentation label; Determining a first deep supervision loss based on the computed tomography bottleneck feature image and the gastric lesion segmentation label; determining a second deep supervision loss based on the fourth computed tomography decoded feature image and the gastric lesion segmentation label; determining a third deep supervision loss based on the third computed tomography decoded feature image and the gastric lesion segmentation label; determining a fourth deep supervision loss based on the second computed tomography decoded feature image and the gastric lesion segmentation label; Determine a target deep supervision loss based on the first deep supervision loss, the second deep supervision loss, the third deep supervision loss, and the fourth deep supervision loss; A target model loss is determined based on the segmentation task loss, the classification task loss, and the target deep supervision loss.

2. The method according to claim 1, characterized in that: The step of acquiring a three-dimensional computer tomography sample image and a gastric lesion segmentation label corresponding to the three-dimensional computer tomography sample image comprises: Acquire a three-dimensional computer tomography image, and perform a preprocessing operation on the three-dimensional computer tomography image to obtain a preprocessed three-dimensional computer tomography image, wherein the preprocessing operation includes: spatially redirecting the three-dimensional computer tomography image; resampling the three-dimensional computer tomography image; and performing foreground segmentation on the three-dimensional computer tomography image; Inputting the preprocessed three-dimensional computed tomography image into a pre-trained gastric lesion segmentation model to obtain a gastric lesion segmentation label; Acquiring a preset CT value range interval, and cropping the preprocessed three-dimensional CT image based on the preset CT value range interval to obtain a cropped three-dimensional CT image; The cropped three-dimensional computed tomography image is subjected to voxel intensity normalization to obtain a three-dimensional computed tomography sample image.

3. The method according to claim 1, characterized in that The residual coding module includes: a first basic residual module and at least one second basic residual module, the first basic residual module and the second basic residual module are spliced ​​and connected, the first basic residual module includes a first convolution layer, a second convolution layer, a third convolution layer and a downsampling layer, and the second basic residual module includes a fourth convolution layer, a fifth convolution layer and a sixth convolution layer.

4. A gastric cancer lesion segmentation method, characterized in that: include: acquiring a computed tomography image to be segmented; Inputting the computed tomography image to be segmented into a trained gastric cancer lesion segmentation model to obtain an initial gastric cancer lesion segmentation result, wherein the gastric cancer lesion segmentation model is obtained after training according to the training method for the gastric cancer lesion segmentation model according to any one of claims 1 to 3; The initial gastric cancer lesion segmentation result is subjected to morphological processing and / or maximum connected domain extraction to obtain a target gastric cancer lesion segmentation result.

5. A training device for a gastric cancer lesion segmentation model, characterized in that: Applicable to the identification of gastric cancer lesions in abdominal CT images, including: A training sample acquisition module, used to acquire a three-dimensional computer tomography sample image and a gastric lesion segmentation label corresponding to the three-dimensional computer tomography sample image; A gastric lesion segmentation result prediction module is used to convert the convolution kernel parameters of 3×3 size into pre-trained parameters of 5×5 size by using a three-way linear interpolation method when initializing the weights; input the three-dimensional computed tomography sample image into a semantic segmentation model to be trained to obtain a gastric lesion segmentation result, wherein the semantic segmentation model includes at least one residual encoding module, at least one bottleneck module and at least one decoding module, and the residual encoding module is jump-connected with the decoding module; A model parameter updating module, used to determine a target model loss based on the gastric lesion segmentation result and the gastric lesion segmentation label, and update the model parameters of the semantic segmentation model based on the target model loss until a model training stop condition is met, thereby obtaining a trained gastric cancer lesion segmentation model; Gastric lesion segmentation result prediction module is specifically used for: Inputting the three-dimensional computed tomography sample image into a first residual coding module of a semantic segmentation model to be trained to obtain a first computed tomography coding feature image; Inputting the first computed tomography encoding feature image into a second residual coding module of a semantic segmentation model to be trained to obtain a second computed tomography encoding feature image; Inputting the second computed tomography encoding feature image into a third residual coding module of the semantic segmentation model to be trained to obtain a third computed tomography encoding feature image; Inputting the third computed tomography encoding feature image into a fourth residual coding module of the semantic segmentation model to be trained to obtain a fourth computed tomography encoding feature image; Inputting the fourth computed tomography encoding feature image into at least one bottleneck module of the semantic segmentation model to be trained to obtain a computed tomography bottleneck feature image; Inputting the computed tomography bottleneck feature image and the fourth computed tomography encoding feature image into a fourth decoding module of the semantic segmentation model to be trained to obtain a fourth computed tomography decoding feature image; Inputting the fourth computed tomography decoding feature image and the third computed tomography encoding feature image into a third decoding module of the semantic segmentation model to be trained to obtain a third computed tomography decoding feature image; Inputting the third computed tomography decoding feature image and the second computed tomography encoding feature image into a second decoding module of a semantic segmentation model to be trained to obtain a second computed tomography decoding feature image; Inputting the second computed tomography decoding feature image and the first computed tomography encoding feature image into a first decoding module of a semantic segmentation model to be trained to obtain a gastric lesion segmentation result; Model parameter update module, including: a segmentation task loss determination unit, configured to determine the segmentation task loss based on the gastric lesion segmentation result and the gastric lesion segmentation label; a classification task loss determination unit, configured to determine the classification task loss based on the fourth computed tomography encoding feature image and the gastric lesion segmentation label; A depth supervision loss determination unit, configured to determine a first depth supervision loss based on the computed tomography bottleneck feature image and the gastric lesion segmentation label; determine a second depth supervision loss based on the fourth computed tomography decoding feature image and the gastric lesion segmentation label; determine a third depth supervision loss based on the third computed tomography decoding feature image and the gastric lesion segmentation label; determine a fourth depth supervision loss based on the second computed tomography decoding feature image and the gastric lesion segmentation label; and determine a target depth supervision loss based on the first depth supervision loss, the second depth supervision loss, the third depth supervision loss, and the fourth depth supervision loss; A target model loss determination unit is used to determine the target model loss based on the segmentation task loss, the classification task loss and the target deep supervision loss.

6. An electronic device, characterized in that: The electronic device comprises: at least one processor; and a memory communicatively coupled to the at least one processor; In which, the memory stores a computer program that can be executed by the at least one processor, and the computer program is executed by the at least one processor so that the at least one processor can execute the training method of the gastric cancer lesion segmentation model described in any one of claims 1-3, or execute the gastric cancer lesion segmentation method described in claim 4.

7. A computer-readable storage medium, characterized in that: The computer-readable storage medium stores computer instructions, and the computer instructions are used to enable a processor to implement the training method for a gastric cancer lesion segmentation model according to any one of claims 1 to 3, or to implement the gastric cancer lesion segmentation method according to claim 4 when executed.

Images