Medical image segmentation method and device based on hierarchical shape prior enhanced ResUNet model

Abstract

The invention provides a medical image segmentation method and device based on a hierarchical shape prior enhanced ResUNet model, and relates to the technical field of medical image segmentation. The method comprises the following steps: acquiring an initial 3D medical image; based on preset processing parameters, performing data processing on the initial 3D medical image to obtain a training 3D medical image; constructing a medical image segmentation model based on the ResUnet model structure; inputting the training 3D medical image into a medical image segmentation model for image segmentation prediction training to obtain a 3D segmented image; carrying out loss function calculation according to the preset marked 3D medical image and the 3D segmentation image to obtain a model segmentation loss; performing parameter optimization on the medical image segmentation model according to the model segmentation loss to obtain an optimized image segmentation model; obtaining a 3D medical image to be segmented; and according to the 3D medical image to be segmented, performing image segmentation by optimizing the image segmentation model. The 3D medical image segmentation method is an accurate and efficient 3D medical image segmentation method based on the hierarchical shape prior enhanced ResUNet model.

Description

Technical Field

[0001] The present invention relates to the technical field of medical image segmentation, and in particular to a medical image segmentation method and device based on a hierarchical shape prior enhanced ResUNet model. Background Art

[0002] Medical images refer to images of the internal structure of the human body taken through a variety of imaging technologies such as X-ray imaging, computed tomography (CT), computed tomography angiography (CTA) and magnetic resonance imaging (MRI), providing important assistance for modern medical care.

[0003] Medical image segmentation refers to dividing an image into several regions based on the similarities and differences between regions in a medical image; it also refers to extracting specific regions of interest based on the different imaging brightness of target organs, tissues, and cells. The segmentation results can be applied to clinical disease diagnosis, medical image visualization, planning treatment plans, and evaluating treatment effects.

[0004] One of the most traditional methods for segmenting medical images is to use manually annotated medical images, which requires medical professionals to manually draw or mark different areas in the segmented image. This requires well-trained professionals and has high labor costs when facing high-resolution images such as multi-slice CT or MAI images. When concentrating on manual operations for a long time, it is also easy to cause fatigue to the annotator, thereby increasing the possibility of errors. At the same time, for areas with complex structures or unclear boundaries due to low contrast, manual annotation may not be able to fully capture the tiny or complex details in the medical image. The various limitations of manual annotation have prompted the development of automated medical image segmentation using computers.

[0005] The earliest automated medical image segmentation methods included edge detection and threshold segmentation using various operators, as well as combining machine learning techniques with traditional image processing. Subsequently, deep learning techniques, especially convolutional neural networks (CNNs), were introduced and rapidly promoted. Among them, the UNet model and its variants became mainstream, and ResUNet, which introduced residual connections in the UNet model, effectively alleviated the gradient vanishing problem in the neural network training process, stood out among a number of variant models, and achieved excellent results in multiple segmentation challenges. However, the ResUNet model still has inherent limitations in capturing global contextual information. When dealing with low contrast between the target structure and surrounding tissues, and when the morphology and position of pathological tissues are highly ambiguous, it is difficult to achieve satisfactory results.

[0006] In the prior art, there is a lack of an accurate and efficient 3D medical image segmentation method based on a hierarchical shape prior enhanced ResUNet model. Summary of the invention

[0007] In order to solve the technical problems of the existing technology in the medical image segmentation problem where the contrast between the target structure and the surrounding tissue is low and the morphology and position of the pathological tissue are highly ambiguous, the embodiment of the present invention provides a medical image segmentation method and device based on a hierarchical shape prior enhanced ResUNet model. The technical solution is as follows:

[0008] On the one hand, a medical image segmentation method based on a hierarchical shape prior enhanced ResUNet model is provided, the method is implemented by a medical image segmentation device, and the method includes:

[0009] Acquire an initial 3D medical image; perform data processing on the initial 3D medical image based on preset processing parameters to obtain a training 3D medical image;

[0010] Constructing a medical image segmentation model based on the ResUnet model structure; the medical image segmentation model is a hierarchical shape-based prior-enhanced ResUNet model;

[0011] Inputting the training 3D medical image into the medical image segmentation model to perform image segmentation prediction training to obtain a 3D segmented image;

[0012] Constructing a loss function according to the preset 3D medical image characteristics; based on the loss function, calculating according to the preset annotated 3D medical image and the 3D segmented image to obtain the model segmentation loss; the loss function includes a Dice loss function, a cross entropy loss function and a BoundaryDoU loss function;

[0013] According to the model segmentation loss, optimizing the parameters of the medical image segmentation model to obtain an optimized image segmentation model;

[0014] Acquire a 3D medical image to be segmented; and perform image segmentation using the optimized image segmentation model according to the 3D medical image to be segmented.

[0015] On the other hand, a medical image segmentation device based on a hierarchical shape prior enhanced ResUNet model is provided, and the device is applied to a medical image segmentation method based on a hierarchical shape prior enhanced ResUNet model, and the device comprises:

[0016] An image processing module, used to obtain an initial 3D medical image; based on preset processing parameters, perform data processing on the initial 3D medical image to obtain a training 3D medical image;

[0017] A model building module, used to build a medical image segmentation model based on the ResUnet model structure; the medical image segmentation model is a hierarchical shape-based prior-enhanced ResUNet model;

[0018] A model training module, used for inputting the training 3D medical image into the medical image segmentation model to perform image segmentation prediction training to obtain a 3D segmented image;

[0019] A loss calculation module, used to construct a loss function according to the preset 3D medical image characteristics; based on the loss function, calculate according to the preset annotated 3D medical image and the 3D segmented image to obtain the model segmentation loss; the loss function includes a Dice loss function, a cross entropy loss function and a BoundaryDoU loss function;

[0020] A model optimization module, used to optimize the parameters of the medical image segmentation model according to the model segmentation loss to obtain an optimized image segmentation model;

[0021] The image segmentation module is used to obtain the 3D medical image to be segmented; and perform image segmentation based on the 3D medical image to be segmented using the optimized image segmentation model.

[0022] On the other hand, a medical image segmentation device is provided, comprising: a processor; a memory, wherein the memory stores computer-readable instructions, and when the computer-readable instructions are executed by the processor, any one of the medical image segmentation methods based on the hierarchical shape prior enhanced ResUNet model is implemented.

[0023] On the other hand, a computer-readable storage medium is provided, in which at least one instruction is stored, and the at least one instruction is loaded and executed by a processor to implement any one of the above-mentioned medical image segmentation methods based on the hierarchical shape prior enhanced ResUNet model.

[0024] The beneficial effects brought about by the technical solution provided by the embodiment of the present invention include at least:

[0025] The present invention proposes a medical image segmentation method based on a hierarchical shape prior enhanced ResUNet model; a hierarchical shape prior module composed of three layers of shape prior modules realizes the layer-by-layer transfer and update of shape prior, and enhances the model's comprehensive understanding of medical images; a comprehensive loss function composed of DiceLoss, Cross-EntropyLoss and is introduced to achieve balanced learning of the detail features and overall structure of medical images, and ensure that the model has high training efficiency when facing complex medical image segmentation scenarios such as high detail, low contrast between segmentation targets and surrounding tissues, and variable morphological positions of diseased organs. The present invention has good portability. The present invention enhances the jump connection part of the ResUNet model without modifying the encoder and decoder of the ResUNet model, so the present invention can be applied to variant models of other UNet models, and has high flexibility and practicality. The present invention is an accurate and efficient 3D medical image segmentation method based on a hierarchical shape prior enhanced ResUNet model. BRIEF DESCRIPTION OF THE DRAWINGS

[0026] In order to more clearly illustrate the technical solutions in the embodiments of the present invention, the following briefly introduces the drawings required for use in the description of the embodiments. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without creative work.

[0027] Figure 1 It is a flow chart of a medical image segmentation method based on a hierarchical shape prior enhanced ResUNet model provided by an embodiment of the present invention;

[0028] Figure 2 It is a schematic diagram of the overall structure of a ResUNet model based on hierarchical shape prior enhancement provided by an embodiment of the present invention;

[0029] Figure 3 It is a schematic diagram of the structure of a self-update module provided by an embodiment of the present invention;

[0030] Figure 4 is a schematic diagram of the overall structure of a cross update module provided by an embodiment of the present invention;

[0031] Figure 5 It is a block diagram of a medical image segmentation device based on a hierarchical shape prior enhanced ResUNet model provided by an embodiment of the present invention;

[0032] Figure 6 It is a structural schematic diagram of a medical image segmentation device provided by an embodiment of the present invention. DETAILED DESCRIPTION

[0033] The technical solution of the present invention is described below in conjunction with the accompanying drawings.

[0034] In the embodiments of the present invention, words such as "exemplarily" and "for example" are used to indicate examples, illustrations or explanations. Any embodiment or design described as "example" in the present invention should not be interpreted as being more preferred or more advantageous than other embodiments or designs. Specifically, the use of the word "example" is intended to present the concept in a specific way. In addition, in the embodiments of the present invention, the meaning expressed by "and / or" can be both, or it can be either of the two.

[0035] In the embodiments of the present invention, "image" and "picture" can sometimes be used interchangeably. It should be noted that when the difference between them is not emphasized, the meanings they intend to express are the same. "of", "corresponding, relevant" and "corresponding" can sometimes be used interchangeably. It should be noted that when the difference between them is not emphasized, the meanings they intend to express are the same.

[0036] In the embodiments of the present invention, sometimes a subscript such as W1 may be written as a non-subscript such as W1. When the difference is not emphasized, the meanings to be expressed are the same.

[0037] In order to make the technical problems, technical solutions and advantages to be solved by the present invention more clear, a detailed description will be given below with reference to the accompanying drawings and specific embodiments.

[0038] The embodiment of the present invention provides a medical image segmentation method based on a hierarchical shape prior enhanced ResUNet model, which can be implemented by a medical image segmentation device, which can be a terminal or a server. Figure 1 The flowchart of the medical image segmentation method based on the hierarchical shape prior enhanced ResUNet model is shown. The processing flow of the method may include the following steps:

[0039] S1. Acquire an initial 3D medical image; perform data processing on the initial 3D medical image based on preset processing parameters to obtain a training 3D medical image.

[0040] Optionally, based on preset processing parameters, data processing is performed on the initial 3D medical image to obtain a training 3D medical image, including:

[0041] Resampling the initial 3D medical image to obtain a sampled 3D medical image with a reduced image size;

[0042] Performing Z-Score normalization processing on the sampled 3D medical image to obtain a processed 3D medical image;

[0043] The processed 3D medical images are subjected to data augmentation operations to obtain training 3D medical images; the data augmentation operations include random pixel perturbation, random image flipping, and random image rotation.

[0044] In a feasible implementation manner, in the present invention, the initial 3D medical image is resampled to reduce the size of the original image and reduce the training pressure; and the data obtained after the resampling is Z-Score normalized to complete the data preprocessing operation; the medical image that has completed the data preprocessing operation is subjected to random pixel perturbation, random flipping and random rotation to complete the data enhancement operation and obtain an image to be segmented that can be used for training.

[0045] The initial 3D medical image is resampled to reduce the original image size and reduce the training pressure, including: using cubic spline interpolation for the initial 3D medical image; using the nearest neighbor interpolation method for label data. Before the initial 3D medical image is input into the medical image segmentation model, the size of each 3D medical image is adjusted to (128, 128, 160) by the above cubic spline interpolation method.

[0046] Random pixel perturbations include random intensity displacement and random intensity scaling; random intensity displacement randomly selects a floating point number from -0.1 to 0.1 and adds it to the image intensity; random intensity scaling randomly selects a floating point number from 0.9 to 1.1 and multiplies it to the image intensity.

[0047] Random flipping is performed on the x, y, and z axes with a probability of 50%. The random rotation angle is randomly selected from -15° to 15°, and the rotation axis is randomly selected from the x, y, and z axes.

[0048] S2. Construct a medical image segmentation model based on the ResUnet model structure; the medical image segmentation model is a hierarchical shape-based prior-enhanced ResUNet model.

[0049] Among them, the medical image segmentation model includes an encoder module, a hierarchical shape prior module and a decoder module;

[0050] The encoder module includes a first downsampling module, a second downsampling module, a third downsampling module and a fourth downsampling module; the first downsampling module includes a compressed convolution layer, a comprehensive convolution layer and an extended convolution layer; the module structure of the second downsampling module is the same as the module structure of the first downsampling module; the module structure of the third downsampling module is the same as the module structure of the first downsampling module; the module structure of the fourth downsampling module is the same as the module structure of the first downsampling module;

[0051] The hierarchical shape prior module includes a first shape prior module, a second shape prior module and a third shape prior module; the model structure of the first shape prior module includes a first self-update module and a first cross-update module; the second shape prior module includes a second self-update module and a second cross-update module; the third shape prior module includes a third self-update module and a third cross-update module;

[0052] The decoder module includes a first upsampling module, a second upsampling module, a third upsampling module and a fourth upsampling module; the first upsampling module includes a compressed convolution layer, a comprehensive convolution layer and an expanded convolution layer; the module structure of the second upsampling module is the same as the module structure of the first upsampling module; the module structure of the third upsampling module is the same as the module structure of the first upsampling module; the module structure of the fourth upsampling module is the same as the module structure of the first upsampling module.

[0053] In a feasible implementation, the first downsampling module includes a compressed convolution layer, a comprehensive convolution layer, and an expanded convolution layer connected sequentially; the convolution layers include a 3D convolution layer, a 3D normalization layer and a ReLU activation function connected sequentially; the second downsampling module, the third downsampling module and the fourth downsampling module have the same structure as the first downsampling module.

[0054] The kernel size of the 3D convolution layer of the compressed convolution layer of the encoder module is set to The convolution kernel size of the 3D convolution layer of the comprehensive convolution layer is set to The convolution kernel size of the 3D convolution layer of the expanded convolution layer is set to .

[0055] The 3D normalization layer of the encoder module uses the group normalization method (GroupNorm) to perform normalization within a feature channel group; the shape of the image to be segmented in the first downsampling module of the encoder module is (128, 128, 160). The shapes of the image after passing through the first, second, third, and fourth downsampling modules of the encoder are x1 (64, 64, 80), x2 (32, 32, 40), x3 (16, 16, 20), and x4 (8, 8, 10), respectively.

[0056] The decoder module includes a first upsampling module, a second upsampling module, a third upsampling module and a fourth upsampling module; the first upsampling module includes a compressed convolution layer, a comprehensive convolution layer, and an expanded convolution layer connected in sequence; the convolution layers include a 3D convolution layer, a 3D normalization layer and a ReLU activation function connected in sequence; the second upsampling module, the third upsampling module and the fourth upsampling module have the same structure as the first upsampling module.

[0057] The kernel size of the 3D convolution layer of the compressed convolution layer of the decoder module is set to The convolution kernel size of the 3D convolution layer of the comprehensive convolution layer is set to The convolution kernel size of the 3D convolution layer of the expanded convolution layer is set to The 3D normalization layer of the decoder module uses batch normalization (BatchNorm) to normalize between samples within a batch.

[0058] The shapes of the images after passing through the fourth, third, second, and first upsampling modules of the decoder are x5 (16, 16, 20), x6 (32, 32, 40), x7 (64, 64, 80), and x8 (128, 128, 160), respectively.

[0059] S3. Input the training 3D medical image into the medical image segmentation model for image segmentation prediction training to obtain a 3D segmented image.

[0060] Optionally, inputting the training 3D medical image into a medical image segmentation model for image segmentation prediction training to obtain a 3D segmented image includes:

[0061] Inputting the training 3D medical image into the encoder module for feature extraction to obtain a first 3D image feature, a second 3D image feature, a third 3D image feature, and a fourth 3D image feature;

[0062] Inputting the first 3D image feature, the second 3D image feature and the third 3D image feature into a hierarchical shape prior module to capture global information, thereby obtaining a first updated shape prior, a second updated shape prior and a third updated shape prior;

[0063] The first updated shape prior, the second updated shape prior, the third updated shape prior and the fourth 3D image feature are input into a decoder module to obtain a 3D segmented image.

[0064] In a feasible implementation, Figure 2 As shown, the hierarchical shape prior module includes a first shape prior module, a second shape prior block and a third shape prior module; each shape prior module includes a self-update module and a cross-update module.

[0065] The image to be segmented is input into the encoder module of the model. The downsampling module of the encoder module extracts medical image features through pooling and convolution operations and inputs them into the cross-update module of the same layer and the downsampling module of the next layer in the form of original feature maps. Among them, since the hierarchical shape prior module only includes three layers of shape prior modules, the downsampling module of the lowest layer only inputs the extracted medical image features into the decoder.

[0066] The cross-update module combines the original feature map input by the downsampling module of the same layer with the shape prior input by the self-update module of the same layer, outputs it to the upsampling module of the same layer of the decoder, and inputs the updated shape prior into the self-update module of the shape prior module of the next layer, thereby introducing the shape prior information into the decoding process and further updating and transmitting the shape prior.

[0067] The self-update module accepts the shape prior input from the cross-update module of the next layer, updates it using the self-attention mechanism, and inputs the updated result into the cross-update module of the next layer.

[0068] The upsampling module of the decoder uses the new feature map input by the cross-update module of the same layer to further restore the restored feature map input by the upsampling module of the next layer, and inputs the result into the upsampling module of the next layer; among them, since the hierarchical shape prior module only includes three layers of shape prior modules, the upsampling module of the lowest layer only accepts the original feature map input from the downsampling module of the lowest layer at the same layer; the upsampling module of the highest layer of the decoder outputs the segmented medical image.

[0069] Optionally, inputting the first 3D image feature, the second 3D image feature and the third 3D image feature into a hierarchical shape prior module to capture global information, and obtaining a first updated shape prior, a second updated shape prior and a third updated shape prior, comprises:

[0070] Inputting the first 3D image feature into a first self-update module for prior capture to obtain a first shape prior; inputting the first 3D image feature and the first shape prior into a first cross-update module for prior update to obtain a first updated shape prior;

[0071] Inputting the second 3D image feature and the first updated shape prior into the second self-update module for prior capture to obtain a second shape prior; inputting the second 3D image feature and the second shape prior into the second cross-update module for prior update to obtain a second updated shape prior;

[0072] The third 3D image feature and the second updated shape prior are input into the third self-update module for prior capture to obtain a third shape prior; the third 3D image feature and the third shape prior are input into the third cross-update module for prior update to obtain a third updated shape prior.

[0073] In a feasible implementation, Figure 3 As shown, the self-update module uses a self-attention mechanism to update the shape prior output from the cross-update module of the higher layer, enhances the ability of the shape prior to capture the global information and internal structural relationship of the medical image, and inputs the shape prior into the cross-update module of the same layer.

[0074] In the self-update module, according to the regularization principle, the shape prior is layer-normalized to reduce internal covariate shift and enhance stability; the self-attention calculation method is used to calculate the query matrix, key matrix and value matrix, and the attention probability matrix is ​​calculated based on the calculation results, and random inactivation is performed; the attention probability matrix is ​​dot-multiplied with the value matrix to derive a context vector with global information; the residual connection method is used to add the context vector to the shape prior, and the result is layer-normalized; a two-layer fully connected network is used to process the shape prior, and the GELU activation function is used for activation in the first layer of the fully connected network; the residual link method is used to add the shape prior processed by the fully connected network to the unprocessed shape prior, and the result is further layer-normalized.

[0075] like Figure 4 As shown, the cross-update module combines the shape prior output by the self-update module with the medical image features extracted by the encoder, outputs it to the upsampling module of the decoder, and inputs the updated shape prior into the self-update module of the shape prior module of the next layer, thereby introducing the shape prior information into the decoding process and further updating and transmitting the shape prior.

[0076] In the cross-update module, the shape prior and medical image features are fused based on the class feature calculation method, and the results are normalized; the normalized result is combined with the shape prior using the Einstein summation convention, and the result is added to the medical image features extracted by the encoder using the residual link method to obtain the updated medical image features and output them to the upsampling module of the decoder in a jump connection manner; the updated medical image features are processed using the residual convolution block, scaled using the cubic spline interpolation method, and added to the shape prior to complete the further update of the shape prior; the updated shape prior is input into the self-update module of the shape prior module of the next layer to complete the transfer of the shape prior.

[0077] S4. Construct a loss function according to the preset 3D medical image characteristics; based on the loss function, calculate according to the preset annotated 3D medical image and 3D segmented image to obtain the model segmentation loss; the loss function includes Dice loss function, cross entropy loss function and BoundaryDoU loss function.

[0078] Optionally, based on the loss function, calculation is performed according to preset annotated 3D medical images and 3D segmented images to obtain a model segmentation loss, including:

[0079] Based on the Dice loss function, the Dice loss is calculated according to the preset annotated 3D medical images and 3D segmented images;

[0080] Based on the cross entropy loss function, the cross entropy loss is calculated according to the preset annotated 3D medical image and the 3D segmented image;

[0081] Based on the BoundaryDoU loss function, the BoundaryDoU loss is calculated based on the preset annotated 3D medical images and 3D segmented images;

[0082] The model segmentation loss is obtained by weighted calculation based on Dice loss, cross entropy loss and BoundaryDoU loss.

[0083] In a feasible implementation, a loss function designed according to the characteristics of 3D medical images is used to calculate the model loss. The corresponding loss function values ​​are calculated according to the following loss function formulas: (1), (2), (3):

[0084] (1);

[0085] (2);

[0086] (3);

[0087] in, Represents the segmentation mask predicted by the model (Predicted mask); represents the ground truth mask; M is the number of samples; and are the true label and predicted probability of sample i belonging to category n respectively; Represents the predicted boundary (Predicted boundary); Represents the ground truth boundary; Represents the symmetric difference between the predicted boundary and the true boundary; DiceLoss represents Dice loss; Cross-EntropyLoss represents Cross-Entropy loss; BoundaryDoULoss represents BoundaryDoU loss.

[0088] The model segmentation loss is calculated based on the DiceLoss between the model segmentation result and the actual annotation, the Cross-EntropyLoss between the 3D segmentation image and the actual annotation, and the BoundaryDoULoss between the model segmentation result and the actual annotation, as shown in the following formula (4):

[0089] (4);

[0090] Among them, 0.5 is the weight of DiceLoss; 0.1 is the weight of Cross-EntropyLoss; 0.4 is the weight of BoundaryDoULoss.

[0091] S5. Optimize the parameters of the medical image segmentation model according to the model segmentation loss to obtain an optimized image segmentation model.

[0092] In one feasible implementation, the model is back-propagated based on the model segmentation loss to update the parameters in the model.

[0093] Based on the similarity between the overall segmentation result of the DiceLoss calculation model and the overall actual annotation, the model is guided to increase the overlap between the segmentation result and the actual annotation; based on the Cross-EntropyLoss calculation model, the difference between each type of segmentation result and each type of actual annotation is guided to better handle the multi-classification task including four categories of renal tissue, renal tumor, renal artery and renal vein; based on the BoundaryDoULoss calculation model, the difference between the boundary of the overall segmentation result and the boundary of the overall actual annotation. Using the loss function in the present invention, structures with subtle morphological structures but important clinical applications, such as thin blood vessels, can be avoided from being ignored.

[0094] In the process of backpropagation updating the parameters in the model, the cosine annealing strategy is used to dynamically adjust the learning rate of the updated parameters. The initial learning rate is 0.0005, and a hot restart is performed every 25 cycles to help the model jump out of the local optimum and improve the generalization ability of the model.

[0095] S6. Acquire the 3D medical image to be segmented; and perform image segmentation by optimizing the image segmentation model according to the 3D medical image to be segmented.

[0096] In a feasible implementation, the 3D medical image to be segmented is input into the optimized model to obtain the segmentation result, and the ITK-Snap software is used to construct a visual 3D medical image.

[0097] The 3D image of the segmentation target is input into the encoder of the trained model, and the segmentation result is obtained from the decoder of the trained model; the initial 3D image is used as the main image, and a new project is created in ITK-Snap; the segmentation result is added to the main image for three-dimensional reconstruction, and the final visualized 3D medical image is obtained.

[0098] In a feasible implementation manner, the present invention uses a segmentation model and a series of models based on a UNet model, as well as a basic UNet model and a basic ResUNet model. The effects on the KiPA 22 dataset and the ACDC heart segmentation dataset are shown in Table 1 (KiPA 22 dataset image segmentation effect table) and Table 2 (ACDC dataset image segmentation effect table):

[0099] Table 1

[0100]

[0101] Table 2

[0102]

[0103] UNet is the unmodified basic UNet model, ResUNet is the unmodified basic ResUNet model, Neural Network U-Net (nnUNet), Multi-layer Feature Aggregation Search Network UXNET for 3D Medical Image Segmentation and Swin Transformers for Semantic Segmentation (SwinUNTR) for 3D Medical Image Analysis are excellent series of models based on the UNet model;

[0104] The values ​​in Tables 1 and 2 include the Dice Similarity Coefficient (DSC), the larger the value, the better the segmentation effect; Hausdorff Distance (HD), where "95" means that the 95th percentile of the distance is considered in this case to reduce the impact of extreme outliers, and the smaller the value, the better the segmentation effect; Average Symmetric Surface Distance (ASSD), the smaller the value, the better the segmentation effect; the bold black values ​​in the table indicate the best segmentation performance.

[0105] Compared with the UNet model, the segmentation index DSC of the present invention on the KiPA 22 dataset is improved by 5.26%, HD95 is reduced by 9.9104mm, and ASSD is reduced by 1.9111mm; the segmentation index DSC of the present invention on the ACDC dataset is improved by 1.54%, HD95 is reduced by 0.4468mm, and ASSD is reduced by 0.1526mm.

[0106] Compared with nnU-Net, UX-NET and SwinUNTR, the segmentation index DSC of the present invention on the KiPA 22 dataset increased by 1.08%, 1.18% and 2.9% respectively; HD95 decreased by 3.5987mm, 3.5872mm and 3.9389mm respectively; ASSD decreased by 0.5964mm, 0.4949mm and 0.8713mm respectively. The segmentation index DSC on the ACDC dataset increased by 1.48%, 2.65% and 2.88% respectively; HD95 decreased by 0.4041mm, 0.7633mm and 0.9737mm respectively; ASSD decreased by 0.0926mm, 0.3422mm and 0.3382mm respectively.

[0107] Compared with the ResUNet model, the segmentation index DSC of the present invention on the KiPA 22 dataset is improved by 1.04%, HD95 is reduced by 1.1150mm, and ASSD is reduced by 0.2146mm; the segmentation index DSC of the present invention on the ACDC dataset is improved by 1.24%, HD95 is reduced by 0.2961mm, and ASSD is reduced by 0.0665mm.

[0108] From the comparison results of Table 1 and Table 2, it can be concluded that the model of the present invention significantly improves the segmentation accuracy of medical images with low contrast, pathological tissue morphology and highly blurred position compared with the existing medical 3D image segmentation method.

[0109] The present invention proposes a medical image segmentation method based on a hierarchical shape prior enhanced ResUNet model; a hierarchical shape prior module composed of three layers of shape prior modules realizes the layer-by-layer transfer and update of shape prior, and enhances the model's comprehensive understanding of medical images; a comprehensive loss function composed of DiceLoss, Cross-EntropyLoss and is introduced to achieve balanced learning of the detail features and overall structure of medical images, and ensure that the model has high training efficiency when facing complex medical image segmentation scenarios such as high detail, low contrast between segmentation targets and surrounding tissues, and variable morphological positions of diseased organs. The present invention has good portability. The present invention enhances the jump connection part of the ResUNet model without modifying the encoder and decoder of the ResUNet model, so the present invention can be applied to variant models of other UNet models, and has high flexibility and practicality. The present invention is an accurate and efficient 3D medical image segmentation method based on a hierarchical shape prior enhanced ResUNet model.

[0110] Figure 51 is a block diagram of a medical image segmentation device based on a hierarchical shape prior enhanced ResUNet model according to an exemplary embodiment, wherein the device is used for a medical image segmentation method based on a hierarchical shape prior enhanced ResUNet model. Figure 5 The device includes an image processing module 510, a model building module 520, a model training module 530, a loss calculation module 540, a model optimization module 550 and an image segmentation module 560. Among them:

[0111] The image processing module 510 is used to obtain an initial 3D medical image; based on preset processing parameters, the initial 3D medical image is processed to obtain a training 3D medical image;

[0112] A model building module 520 is used to build a medical image segmentation model based on the ResUnet model structure; the medical image segmentation model is a hierarchical shape-based prior-enhanced ResUNet model;

[0113] A model training module 530 is used to input the training 3D medical image into the medical image segmentation model to perform image segmentation prediction training to obtain a 3D segmented image;

[0114] The loss calculation module 540 is used to construct a loss function according to the preset 3D medical image characteristics; based on the loss function, the model segmentation loss is calculated according to the preset annotated 3D medical image and the 3D segmentation image; the loss function includes a Dice loss function, a cross entropy loss function and a BoundaryDoU loss function;

[0115] A model optimization module 550 is used to optimize the parameters of the medical image segmentation model according to the model segmentation loss to obtain an optimized image segmentation model;

[0116] The image segmentation module 560 is used to obtain the 3D medical image to be segmented; and perform image segmentation by optimizing the image segmentation model according to the 3D medical image to be segmented.

[0117] Optionally, the image processing module 510 is further configured to:

[0118] Resampling the initial 3D medical image to obtain a sampled 3D medical image with a reduced image size;

[0119] Performing Z-Score normalization processing on the sampled 3D medical image to obtain a processed 3D medical image;

[0120] The processed 3D medical images are subjected to data augmentation operations to obtain training 3D medical images; the data augmentation operations include random pixel perturbation, random image flipping, and random image rotation.

[0121] Among them, the medical image segmentation model includes an encoder module, a hierarchical shape prior module and a decoder module;

[0122] The encoder module includes a first downsampling module, a second downsampling module, a third downsampling module and a fourth downsampling module; the first downsampling module includes a compressed convolution layer, a comprehensive convolution layer and an extended convolution layer; the module structure of the second downsampling module is the same as the module structure of the first downsampling module; the module structure of the third downsampling module is the same as the module structure of the first downsampling module; the module structure of the fourth downsampling module is the same as the module structure of the first downsampling module;

[0123] The hierarchical shape prior module includes a first shape prior module, a second shape prior module and a third shape prior module; the model structure of the first shape prior module includes a first self-update module and a first cross-update module; the second shape prior module includes a second self-update module and a second cross-update module; the third shape prior module includes a third self-update module and a third cross-update module;

[0124] The decoder module includes a first upsampling module, a second upsampling module, a third upsampling module and a fourth upsampling module; the first upsampling module includes a compressed convolution layer, a comprehensive convolution layer and an expanded convolution layer; the module structure of the second upsampling module is the same as the module structure of the first upsampling module; the module structure of the third upsampling module is the same as the module structure of the first upsampling module; the module structure of the fourth upsampling module is the same as the module structure of the first upsampling module.

[0125] Optionally, the model training module 530 is further used to:

[0126] Inputting the training 3D medical image into the encoder module for feature extraction to obtain a first 3D image feature, a second 3D image feature, a third 3D image feature, and a fourth 3D image feature;

[0127] Inputting the first 3D image feature, the second 3D image feature and the third 3D image feature into a hierarchical shape prior module to capture global information, thereby obtaining a first updated shape prior, a second updated shape prior and a third updated shape prior;

[0128] The first updated shape prior, the second updated shape prior, the third updated shape prior and the fourth 3D image feature are input into a decoder module to obtain a 3D segmented image.

[0129] Optionally, the model training module 530 is further used to:

[0130] Inputting the first 3D image feature into a first self-update module for prior capture to obtain a first shape prior; inputting the first 3D image feature and the first shape prior into a first cross-update module for prior update to obtain a first updated shape prior;

[0131] Inputting the second 3D image feature and the first updated shape prior into the second self-update module for prior capture to obtain a second shape prior; inputting the second 3D image feature and the second shape prior into the second cross-update module for prior update to obtain a second updated shape prior;

[0132] The third 3D image feature and the second updated shape prior are input into the third self-update module for prior capture to obtain a third shape prior; the third 3D image feature and the third shape prior are input into the third cross-update module for prior update to obtain a third updated shape prior.

[0133] Optionally, the loss calculation module 540 is further configured to:

[0134] Based on the Dice loss function, the Dice loss is calculated according to the preset annotated 3D medical images and 3D segmented images;

[0135] Based on the cross entropy loss function, the cross entropy loss is calculated according to the preset annotated 3D medical image and the 3D segmented image;

[0136] Based on the BoundaryDoU loss function, the BoundaryDoU loss is calculated based on the preset annotated 3D medical images and 3D segmented images;

[0137] The model segmentation loss is obtained by weighted calculation based on Dice loss, cross entropy loss and BoundaryDoU loss.

[0138] The present invention proposes a medical image segmentation method based on a hierarchical shape prior enhanced ResUNet model; a hierarchical shape prior module composed of three layers of shape prior modules realizes the layer-by-layer transfer and update of shape prior, and enhances the model's comprehensive understanding of medical images; a comprehensive loss function composed of DiceLoss, Cross-EntropyLoss and is introduced to achieve balanced learning of the detail features and overall structure of medical images, and ensure that the model has high training efficiency when facing complex medical image segmentation scenarios such as high detail, low contrast between segmentation targets and surrounding tissues, and variable morphological positions of diseased organs. The present invention has good portability. The present invention enhances the jump connection part of the ResUNet model without modifying the encoder and decoder of the ResUNet model, so the present invention can be applied to variant models of other UNet models, and has high flexibility and practicality. The present invention is an accurate and efficient 3D medical image segmentation method based on a hierarchical shape prior enhanced ResUNet model.

[0139] Figure 6 is a schematic diagram of the structure of a medical image segmentation device provided by an embodiment of the present invention, such as Figure 6 As shown, the medical image segmentation device may include the above Figure 5 The medical image segmentation apparatus based on the hierarchical shape prior enhanced ResUNet model is shown. Optionally, the medical image segmentation device 610 may include a first processor 2001 .

[0140] Optionally, the medical image segmentation device 610 may further include a memory 2002 and a transceiver 2003 .

[0141] The first processor 2001, the memory 2002 and the transceiver 2003 may be connected via a communication bus.

[0142] Combine the following Figure 6 The components of the medical image segmentation device 610 are specifically introduced as follows:

[0143] The first processor 2001 is the control center of the medical image segmentation device 610, and may be a processor or a general term for multiple processing elements. For example, the first processor 2001 is one or more central processing units (CPUs), or may be application specific integrated circuits (ASICs), or may be one or more integrated circuits configured to implement the embodiments of the present invention, such as one or more microprocessors (digital signal processors, DSPs), or one or more field programmable gate arrays (field programmable gate arrays, FPGAs).

[0144] Optionally, the first processor 2001 may perform various functions of the medical image segmentation device 610 by running or executing a software program stored in the memory 2002 , and calling data stored in the memory 2002 .

[0145] In a specific implementation, as an embodiment, the first processor 2001 may include one or more CPUs, such as Figure 6 CPU0 and CPU1 are shown in FIG.

[0146] In a specific implementation, as an embodiment, the medical image segmentation device 610 may also include multiple processors, such as Figure 6 The first processor 2001 and the second processor 2004 are shown in FIG. Each of these processors may be a single-core processor (single-CPU) or a multi-core processor (multi-CPU). The processor here may refer to one or more devices, circuits, and / or processing cores for processing data (e.g., computer program instructions).

[0147] The memory 2002 is used to store the software program for executing the solution of the present invention, and is controlled to be executed by the first processor 2001. The specific implementation method can refer to the above method embodiment, which will not be repeated here.

[0148] Optionally, the memory 2002 may be a read-only memory (ROM) or other types of static storage devices that can store static information and instructions, a random access memory (RAM) or other types of dynamic storage devices that can store information and instructions, or an electrically erasable programmable read-only memory (EEPROM), a compact disc read-only memory (CD-ROM) or other optical disc storage, optical disc storage (including compressed optical disc, laser disc, optical disc, digital versatile disc, Blu-ray disc, etc.), a magnetic disk storage medium or other magnetic storage device, or any other medium that can be used to carry or store the desired program code in the form of instructions or data structures and can be accessed by a computer, but is not limited thereto. The memory 2002 may be integrated with the first processor 2001, or may exist independently and access the first processor 2001 through the interface circuit ( Figure 6 (not shown) is coupled to the first processor 2001, which is not specifically limited in this embodiment of the present invention.

[0149] The transceiver 2003 is used to communicate with a network device or a terminal device.

[0150] Optionally, the transceiver 2003 may include a receiver and a transmitter ( Figure 6 The receiver is used to implement a receiving function, and the transmitter is used to implement a sending function.

[0151] Optionally, the transceiver 2003 may be integrated with the first processor 2001, or may exist independently and communicate with the first processor 2001 through the interface circuit ( Figure 6 (not shown) is coupled to the first processor 2001, which is not specifically limited in this embodiment of the present invention.

[0152] It should be noted that Figure 6 The structure of the medical image segmentation device 610 shown in the figure does not constitute a limitation on the router, and the actual knowledge structure recognition device may include more or fewer components than those shown in the figure, or combine certain components, or arrange the components differently.

[0153] In addition, the technical effects of the medical image segmentation device 610 can refer to the technical effects of the medical image segmentation method based on the hierarchical shape prior enhanced ResUNet model described in the above method embodiment, which will not be repeated here.

[0154] It should be understood that the first processor 2001 in the embodiment of the present invention may be a central processing unit (CPU), and the processor may also be other general-purpose processors, digital signal processors (DSP), application specific integrated circuits (ASIC), field programmable gate arrays (FPGA) or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. A general-purpose processor may be a microprocessor or the processor may also be any conventional processor, etc.

[0155] It should also be understood that the memory in the embodiments of the present invention may be a volatile memory or a non-volatile memory, or may include both volatile and non-volatile memories. Among them, the non-volatile memory may be a read-only memory (ROM), a programmable read-only memory (PROM), an erasable programmable read-only memory (EPROM), an electrically erasable programmable read-only memory (EEPROM), or a flash memory. The volatile memory may be a random access memory (RAM), which is used as an external cache. By way of example and not limitation, many forms of random access memory (RAM) are available, such as static RAM (SRAM), dynamic random access memory (DRAM), synchronous DRAM (SDRAM), double data rate synchronous dynamic random access memory (DDR SDRAM), enhanced synchronous dynamic random access memory (ESDRAM), synchronous link DRAM (SLDRAM), and direct rambus RAM (DR RAM).

[0156] The above embodiments can be implemented in whole or in part by software, hardware (such as circuits), firmware or any other combination. When implemented by software, the above embodiments can be implemented in whole or in part in the form of a computer program product. The computer program product includes one or more computer instructions or computer programs. When the computer instructions or computer programs are loaded or executed on a computer, the process or function described in the embodiment of the present invention is generated in whole or in part. The computer can be a general-purpose computer, a special-purpose computer, a computer network, or other programmable device. The computer instructions can be stored in a computer-readable storage medium, or transmitted from one computer-readable storage medium to another computer-readable storage medium. For example, the computer instructions can be transmitted from one website, computer, server or data center to another website, computer, server or data center by wired (such as infrared, wireless, microwave, etc.). The computer-readable storage medium can be any available medium that can be accessed by a computer or a data storage device such as a server or data center that contains one or more available media sets. The available medium can be a magnetic medium (for example, a floppy disk, a hard disk, a tape), an optical medium (for example, a DVD), or a semiconductor medium. The semiconductor medium can be a solid-state hard disk.

[0157] It should be understood that the term "and / or" in this article is only a description of the association relationship of associated objects, indicating that there can be three relationships. For example, A and / or B can represent: A exists alone, A and B exist at the same time, and B exists alone. A and B can be singular or plural. In addition, the character " / " in this article generally indicates that the associated objects before and after are in an "or" relationship, but it may also indicate an "and / or" relationship. Please refer to the context for specific understanding.

[0158] In the present invention, "at least one" means one or more, and "plurality" means two or more. "At least one of the following" or similar expressions refers to any combination of these items, including any combination of single items or plural items. For example, at least one of a, b, or c can mean: a, b, c, ab, ac, bc, or abc, where a, b, and c can be single or multiple.

[0159] It should be understood that in various embodiments of the present invention, the size of the serial numbers of the above-mentioned processes does not mean the order of execution. The execution order of each process should be determined by its function and internal logic, and should not constitute any limitation on the implementation process of the embodiments of the present invention.

[0160] Those skilled in the art will appreciate that the units and algorithm steps of each example described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are performed in hardware or software depends on the specific application and design constraints of the technical solution. Professional and technical personnel can use different methods to implement the described functions for each specific application, but such implementation should not be considered to be beyond the scope of the present invention.

[0161] Those skilled in the art can clearly understand that, for the convenience and brevity of description, the specific working processes of the above-described equipment, devices and units can refer to the corresponding processes in the aforementioned method embodiments and will not be repeated here.

[0162] In the several embodiments provided by the present invention, it should be understood that the disclosed devices, apparatuses and methods can be implemented in other ways. For example, the device embodiments described above are only schematic. For example, the division of the units is only a logical function division. There may be other division methods in actual implementation, such as multiple units or components can be combined or integrated into another device, or some features can be ignored or not executed. Another point is that the mutual coupling or direct coupling or communication connection shown or discussed can be through some interfaces, indirect coupling or communication connection of devices or units, which can be electrical, mechanical or other forms.

[0163] The units described as separate components may or may not be physically separated, and the components shown as units may or may not be physical units, that is, they may be located in one place or distributed on multiple network units. Some or all of the units may be selected according to actual needs to achieve the purpose of the solution of this embodiment.

[0164] In addition, each functional unit in each embodiment of the present invention may be integrated into one processing unit, or each unit may exist physically separately, or two or more units may be integrated into one unit.

[0165] If the functions are implemented in the form of software functional units and sold or used as independent products, they can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention, or the part that contributes to the prior art, or the part of the technical solution, can be embodied in the form of a software product. The computer software product is stored in a storage medium, including several instructions for enabling a computer device (which can be a personal computer, server, or network device, etc.) to perform all or part of the steps of the methods described in various embodiments of the present invention. The aforementioned storage medium includes: U disk, mobile hard disk, read-only memory (ROM), random access memory (RAM), disk or optical disk, and other media that can store program codes.

[0166] The above is only a specific embodiment of the present invention, but the protection scope of the present invention is not limited thereto. Any person skilled in the art can easily think of changes or substitutions within the technical scope disclosed by the present invention, which should be included in the protection scope of the present invention. Therefore, the protection scope of the present invention should be based on the protection scope of the claims.

Claims

1. A medical image segmentation method based on a hierarchical shape prior enhanced ResUNet model, characterized in that: The method comprises: Acquire an initial 3D medical image; perform data processing on the initial 3D medical image based on preset processing parameters to obtain a training 3D medical image; Constructing a medical image segmentation model based on the ResUnet model structure; the medical image segmentation model is a hierarchical shape-based prior-enhanced ResUNet model; Inputting the training 3D medical image into the medical image segmentation model to perform image segmentation prediction training to obtain a 3D segmented image; Constructing a loss function according to the preset 3D medical image characteristics; based on the loss function, calculating according to the preset annotated 3D medical image and the 3D segmented image to obtain the model segmentation loss; the loss function includes a Dice loss function, a cross entropy loss function and a BoundaryDoU loss function; According to the model segmentation loss, optimizing the parameters of the medical image segmentation model to obtain an optimized image segmentation model; Acquire a 3D medical image to be segmented; and perform image segmentation using the optimized image segmentation model according to the 3D medical image to be segmented.

2. The medical image segmentation method based on the hierarchical shape prior enhanced ResUNet model according to claim 1, characterized in that: The step of performing data processing on the initial 3D medical image based on preset processing parameters to obtain a training 3D medical image includes: Resampling the initial 3D medical image to obtain a sampled 3D medical image with a reduced image size; Performing Z-Score normalization processing on the sampled 3D medical image to obtain a processed 3D medical image; Performing a data enhancement operation on the processed 3D medical image to obtain a training 3D medical image; the data enhancement operation includes random pixel perturbation, random image flipping and random image rotation.

3. The medical image segmentation method based on the hierarchical shape prior enhanced ResUNet model according to claim 1, characterized in that: The medical image segmentation model includes an encoder module, a hierarchical shape prior module and a decoder module; The encoder module includes a first down-sampling module, a second down-sampling module, a third down-sampling module and a fourth down-sampling module; the first down-sampling module includes a compressed convolution layer, a comprehensive convolution layer and an extended convolution layer; the module structure of the second down-sampling module is the same as the module structure of the first down-sampling module; the module structure of the third down-sampling module is the same as the module structure of the first down-sampling module; the module structure of the fourth down-sampling module is the same as the module structure of the first down-sampling module; The hierarchical shape prior module includes a first shape prior module, a second shape prior module and a third shape prior module; the model structure of the first shape prior module includes a first self-update module and a first cross-update module; the second shape prior module includes a second self-update module and a second cross-update module; the third shape prior module includes a third self-update module and a third cross-update module; The decoder module includes a first upsampling module, a second upsampling module, a third upsampling module and a fourth upsampling module; the first upsampling module includes a compressed convolution layer, a comprehensive convolution layer and an extended convolution layer; the module structure of the second upsampling module is the same as the module structure of the first upsampling module; the module structure of the third upsampling module is the same as the module structure of the first upsampling module; the module structure of the fourth upsampling module is the same as the module structure of the first upsampling module.

4. The medical image segmentation method based on the hierarchical shape prior enhanced ResUNet model according to claim 3, characterized in that: The step of inputting the training 3D medical image into the medical image segmentation model to perform image segmentation prediction training to obtain a 3D segmented image comprises: Inputting the training 3D medical image into the encoder module for feature extraction to obtain a first 3D image feature, a second 3D image feature, a third 3D image feature, and a fourth 3D image feature; Inputting the first 3D image feature, the second 3D image feature and the third 3D image feature into the hierarchical shape prior module to capture global information, thereby obtaining a first updated shape prior, a second updated shape prior and a third updated shape prior; The first updated shape prior, the second updated shape prior, the third updated shape prior and the fourth 3D image feature are input into the decoder module to obtain a 3D segmented image.

5. The medical image segmentation method based on the hierarchical shape prior enhanced ResUNet model according to claim 4, characterized in that: The step of inputting the first 3D image feature, the second 3D image feature and the third 3D image feature into the hierarchical shape prior module to capture global information and obtain a first updated shape prior, a second updated shape prior and a third updated shape prior comprises: Inputting the first 3D image feature into the first self-update module for prior capture to obtain a first shape prior; inputting the first 3D image feature and the first shape prior into the first cross-update module for prior update to obtain a first updated shape prior; Inputting the second 3D image feature and the first updated shape prior into the second self-update module for prior capture to obtain a second shape prior; inputting the second 3D image feature and the second shape prior into the second cross-update module for prior update to obtain a second updated shape prior; The third 3D image feature and the second updated shape prior are input into the third self-update module for prior capture to obtain a third shape prior; the third 3D image feature and the third shape prior are input into the third cross-update module for prior update to obtain a third updated shape prior.

6. The medical image segmentation method based on the hierarchical shape prior enhanced ResUNet model according to claim 1, characterized in that: The step of calculating based on the loss function according to the preset annotated 3D medical image and the 3D segmented image to obtain the model segmentation loss includes: Based on the Dice loss function, a Dice loss is obtained by performing calculation according to the preset annotated 3D medical image and the 3D segmented image; Based on the cross entropy loss function, a cross entropy loss is obtained by calculating according to the preset annotated 3D medical image and the 3D segmented image; Based on the BoundaryDoU loss function, a BoundaryDoU loss is obtained by performing calculation according to the preset annotated 3D medical image and the 3D segmented image; A weighted calculation is performed according to the Dice loss, the cross entropy loss, and the BoundaryDoU loss to obtain a model segmentation loss.

7. A medical image segmentation device based on a hierarchical shape prior enhanced ResUNet model, the medical image segmentation device based on a hierarchical shape prior enhanced ResUNet model is used to implement the medical image segmentation method based on a hierarchical shape prior enhanced ResUNet model as claimed in any one of claims 1 to 6, characterized in that: The device comprises: An image processing module is used to obtain an initial 3D medical image; based on preset processing parameters, the initial 3D medical image is processed to obtain a training 3D medical image; A model building module, used to build a medical image segmentation model based on the ResUNet model structure; the medical image segmentation model is a hierarchical shape-based prior-enhanced ResUNet model; A model training module, used for inputting the training 3D medical image into the medical image segmentation model to perform image segmentation prediction training to obtain a 3D segmented image; A loss calculation module, used to construct a loss function according to the preset 3D medical image characteristics; based on the loss function, calculate according to the preset annotated 3D medical image and the 3D segmented image to obtain the model segmentation loss; the loss function includes a Dice loss function, a cross entropy loss function and a BoundaryDoU loss function; A model optimization module, used to optimize the parameters of the medical image segmentation model according to the model segmentation loss to obtain an optimized image segmentation model; The image segmentation module is used to obtain the 3D medical image to be segmented; and perform image segmentation based on the 3D medical image to be segmented using the optimized image segmentation model.

8. The medical image segmentation device based on the hierarchical shape prior enhanced ResUNet model according to claim 7, characterized in that: The image processing module is further used for: Resampling the initial 3D medical image to obtain a sampled 3D medical image with a reduced image size; Performing Z-Score normalization processing on the sampled 3D medical image to obtain a processed 3D medical image; Performing a data enhancement operation on the processed 3D medical image to obtain a training 3D medical image; the data enhancement operation includes random pixel perturbation, random image flipping and random image rotation.

9. A medical image segmentation device, characterized in that: The medical image segmentation device comprises: processor; A memory having computer-readable instructions stored thereon, wherein when the computer-readable instructions are executed by the processor, the method according to any one of claims 1 to 6 is implemented.

10. A computer-readable storage medium, characterized in that: The computer-readable storage medium stores program codes, which can be called by a processor to execute the method according to any one of claims 1 to 6.

Images