Pathological image colon gland segmentation method, device, equipment, medium and program product

Abstract

The invention discloses a pathological image colon gland segmentation method, device and equipment, a medium and a program product, and the method comprises the steps: carrying out the feature extraction of an image sample obtained after preprocessing based on two processing branches, and obtaining a first candidate feature map and a second candidate feature map, performing feature extraction on the second candidate feature map along multiple directions to obtain global feature information, obtaining local feature information of the second candidate feature map, aggregating the local feature information and the global feature information to obtain a second target feature map, and performing feature fusion on the first candidate feature map and the second target feature map to obtain a second target feature map; the target fusion feature is decoded, the decoding result is projected to the same dimension as the original pathological image, and the colon gland segmentation result of the pathological image is obtained, so that the global modeling and local detail sensing ability of the image is enhanced, the processing ability of the overall structure and details of the image is remarkably improved, and the processing efficiency of the colon gland segmentation is improved. Therefore, the accuracy of colon gland segmentation of the pathological image is improved.

Description

Technical Field

[0001] The present invention relates to the field of pathological image recognition, and particularly to a method, device, equipment, medium and program product for segmenting colon glands in pathological images. Background Art

[0002] Currently, the segmentation of glands in pathological images mainly relies on experienced pathologists for manual annotation. However, this process is not only time-consuming and laborious, but also easily affected by individual subjective factors, resulting in large variability and limited repeatability of the segmentation results. With the rapid development of computer technology and artificial intelligence technology, deep learning technology has been increasingly widely used in medical image processing. Deep learning technology can achieve automated processing and intelligent analysis of medical images. Using deep learning technology to perform intelligent analysis on colon glands in pathological images can provide important diagnostic basis for doctors, reduce misjudgments caused by subjective analysis, improve the accuracy and efficiency of diagnosis, and reduce waste of medical resources. Since malignant glands are usually irregular in shape, unclear in edge due to severe deformation and canceration, and are accompanied by infiltration and destruction of surrounding tissues, it is extremely difficult to segment the gland region from surrounding irrelevant tissues. In addition, benign glands have various shapes, uneven distributions, and severe adhesions between glands, making it very difficult to accurately segment benign glands. Moreover, due to the continuity between the gland region and the background tissue and its blurred edge, existing segmentation methods often have difficulty in effectively distinguishing the two. Summary of the Invention

[0003] The main objective of the present invention is to provide a method, device, equipment, medium and program product for segmenting colon glands in pathological images, aiming to solve the technical problem that the prior art is limited by the continuity between the gland region and the background tissue and its blurred edge, and cannot effectively identify and distinguish the gland region from the background tissue, resulting in the inability to accurately segment colon glands from pathological images.

[0004] To achieve the above objective, the present invention provides a method for segmenting colon glands in pathological images, the method comprising the following steps: Preprocess the original pathological image to obtain a candidate image sample; Respectively perform feature extraction on the candidate image sample based on a first processing branch and a second processing branch to obtain a first candidate feature map and a second candidate feature map; Extract features of the second candidate feature map in multiple directions, and aggregate the feature extraction results in each direction to obtain global feature information; Perform local feature extraction on the second candidate feature map through a convolution operation to obtain local feature information, and aggregate the local feature information with the global feature information to obtain a second target feature map; Fuse the first candidate feature map and the second target feature map to obtain a target fused feature; Decode the target fused feature and project the decoding result to the same dimension as the original pathological image to obtain the colon gland segmentation result of the pathological image.

[0005] Optionally, the preprocessing of the original pathological image to obtain candidate image samples includes: Denoise the original pathological image; Perform stain normalization on the denoised original pathological image to obtain a stain-normalized pathological image; Perform data augmentation on the stain-normalized pathological image and generate an initial image dataset based on the data-augmented stain-normalized pathological image, where the initial image dataset includes multiple initial pathological image samples; Perform image feature serialization processing on the initial image dataset to obtain candidate image samples.

[0006] Optionally, the feature extraction of the candidate image samples based on the first processing branch and the second processing branch respectively to obtain the first candidate feature map and the second candidate feature map includes: Extract features from the candidate image samples based on the first processing branch through a linear layer and an activation function to obtain the first candidate feature map : where, represents the first candidate feature map, represents the candidate image samples, represents the activation function, represents the weight matrix, represents the bias vector; Extract features from the candidate image samples based on the second processing branch through a linear layer, a depthwise separable convolutional layer and an activation function to obtain the second candidate feature map: where, represents the second candidate feature map, represents the feature map without depthwise separable convolutional operation, represents the weight matrix, represents the bias vector, represents the depthwise separable convolutional operation.

[0007] Optionally, the feature extraction of the second candidate feature map in multiple directions and the aggregation of the feature extraction results in each direction to obtain global feature information includes: Unfold the second candidate feature map along multiple directions to obtain one-dimensional features in multiple directions; Input the one-dimensional features into multiple linear layers, and update the hidden state based on the learnable matrix parameters output by the multiple linear layers: where, 、 and respectively represent the learnable matrix parameters output by the linear layers, represents the learnable matrix parameter output by the linear layer, respectively represent the learnable matrix parameters after discretization, represents the derivative of the hidden state at the current moment, represents the hidden state at the previous moment, represents the two-dimensional feature of the current input, represents the identity matrix, represents the initial matrix; Obtain global feature information based on the updated hidden state, the one-dimensional features, and the output learnable matrix parameters of each linear layer: where, represents the learnable matrix parameter output by the linear layer, represents the extracted global feature information; Aggregate the global features after global modeling of the one-dimensional features in multiple directions to obtain two-dimensional features.

[0008] Optionally, the decoding of the target fusion feature and the projection of the decoding result to the same dimension as the original pathological image to obtain the colon gland segmentation result of the pathological image includes: Decode the target fusion feature and project the decoding result to the same dimension as the original pathological image to obtain an initial segmentation result; Label each connected region in the initial segmentation result to obtain multiple labeled regions; Based on the area of each labeled region, screen out the target region from the labeled regions, and remove the labeled regions other than the target region from the initial segmentation result to obtain a candidate segmentation result; Perform filtering processing on the candidate segmentation result, and perform hole filling on the filtered candidate segmentation result to obtain the colon gland segmentation result of the pathological image.

[0009] Optionally, after preprocessing the original pathological image to obtain candidate image samples, the method further includes: Inputting the candidate image samples into an image segmentation model for colon gland segmentation of pathological images to obtain the colon gland segmentation results of pathological images; Wherein, the image segmentation model includes an image patch embedding module, a plurality of encoders, a plurality of decoders corresponding to the plurality of encoders, and a hybrid dilated convolution module, and the hybrid dilated convolution module is connected between the encoder and the decoder; The image patch embedding module is used for image feature serialization and layer normalization processing to obtain candidate image samples; The encoder is used to respectively perform feature extraction on the candidate image samples based on a first processing branch and a second processing branch to obtain a first candidate feature map and a second candidate feature map; The encoder is further used to perform feature extraction on the second candidate feature map in multiple directions, and aggregate the feature extraction results in each direction to obtain global feature information; The encoder is further used to perform local feature extraction on the second candidate feature map through convolution operations, obtain local feature information, and aggregate the local feature information with the global feature information to obtain a second target feature map; The encoder is further used to perform feature fusion on the first candidate feature map and the second target feature map to obtain a target fusion feature; The hybrid dilated convolution module is used to perform multi-scale feature perception on the target fusion feature output by the encoder to obtain multi-scale perception features, and input the multi-scale perception features into the decoder for decoding; The decoder is used to decode the target fusion feature output by the encoder and / or the multi-scale perception features output by the hybrid dilated convolution module, and project the decoding result to the same dimension as the original pathological image to obtain the colon gland segmentation results of pathological images.

[0010] In addition, to achieve the above object, the present invention further provides a colon gland segmentation device for pathological images, and the colon gland segmentation device for pathological images includes: An image processing module for preprocessing the original pathological image to obtain candidate image samples; A multi-branch feature extraction module for respectively performing feature extraction on the candidate image samples based on a first processing branch and a second processing branch to obtain a first candidate feature map and a second candidate feature map; A global feature extraction module for performing feature extraction on the second candidate feature map in multiple directions, and aggregating the feature extraction results in each direction to obtain global feature information; A local feature extraction module, configured to perform local feature extraction on the second candidate feature map through convolutional operations, obtain local feature information, and aggregate the local feature information with the global feature information to obtain a second target feature map; A feature fusion module, configured to fuse the first candidate feature map with the second target feature map to obtain a target fusion feature; A feature decoding module, configured to decode the target fusion feature and project the decoding result onto the same dimension as the original pathological image to obtain a colon gland segmentation result of the pathological image.

[0011] In addition, to achieve the above object, the present application also provides a colon gland segmentation device for pathological images, the device includes: a memory, a processor, and a computer program stored on the memory and executable on the processor, the computer program is configured to implement the steps of the colon gland segmentation method for pathological images as described above.

[0012] In addition, to achieve the above object, the present application also provides a computer-readable storage medium, on which a computer program is stored, and when the computer program is executed by a processor, the steps of the colon gland segmentation method for pathological images as described above are implemented.

[0013] In addition, to achieve the above object, the present application also provides a computer program product, the computer program product includes a computer program, and when the computer program is executed by a processor, the steps of the colon gland segmentation method for pathological images as described above are implemented.

[0014] The present invention preprocesses the original pathological image to obtain candidate image samples; respectively performs feature extraction on the candidate image samples based on a first processing branch and a second processing branch to obtain a first candidate feature map and a second candidate feature map; performs feature extraction on the second candidate feature map in multiple directions, and aggregates the feature extraction results in each direction to obtain global feature information; performs local feature extraction on the second candidate feature map through a convolution operation to obtain local feature information, and aggregates the local feature information and the global feature information to obtain a second target feature map; performs feature fusion on the first candidate feature map and the second target feature map to obtain a target fusion feature; decodes the target fusion feature and projects the decoding result to the same dimension as the original pathological image to obtain a pathological image colon gland segmentation result; since the present invention performs feature extraction on candidate images based on two processing branches, and respectively extracts the global feature information and local feature information of the second candidate feature map based on the second processing branch, realizes collaborative optimization of global and local feature representations, thereby enhancing the multi-scale perception ability of gland structures, improving the segmentation accuracy, performs feature fusion on the first candidate feature map and the second target feature map, thereby making up for the problem of the loss of small-scale feature correlation relationships in the single-branch linear scanning process, enhancing the perception ability of global dependence and local details of the image, significantly improving the processing ability of image details, and thus improving the accuracy of pathological image colon gland segmentation. Description of the Drawings

[0015] In order to more clearly illustrate the technical solutions in the embodiments of the present application or the prior art, the following will briefly introduce the drawings required for use in the description of the embodiments or the prior art. Obviously, for those of ordinary skill in the art, without creative efforts, other drawings can also be obtained based on these drawings.

[0016] Figure 1 It is a schematic structural diagram of a pathological image colon gland segmentation device in the hardware operating environment related to the embodiment solution of the present invention; Figure 2 It is a schematic flowchart of the first embodiment of the pathological image colon gland segmentation method of the present invention; Figure 3 It is a schematic structural diagram of the CSS2D module in the first embodiment of the pathological image colon gland segmentation method of the present invention; Figure 4 It is a schematic diagram of the post-processing flow in an embodiment of the pathological image colon gland segmentation method of the present invention; Figure 5 It is a schematic diagram of the image segmentation model network architecture in the second embodiment of the pathological image colon gland segmentation method of the present invention; Figure 6Schematic diagram of the processing flow of the Ls-VSS module in the second embodiment of the method for segmenting colon glands in pathological images of the present invention; Figure 7 Schematic diagram of the structure of the hybrid dilated convolution module in the second embodiment of the method for segmenting colon glands in pathological images of the present invention; Figure 8 Block diagram of the structure of an embodiment of the device for segmenting colon glands in pathological images of the present invention.

[0017] The realization, functional features and advantages of the object of the present invention will be further described with reference to the embodiments and the accompanying drawings. Detailed implementation manners

[0018] It should be understood that the specific embodiments described herein are only used to explain the present invention and are not used to limit the present invention.

[0019] Refer to Figure 1 , Figure 1 Schematic diagram of the structure of the device for segmenting colon glands in pathological images, which is the hardware operating environment involved in the embodiment solution of the present invention.

[0020] As Figure 1 shown, the device for segmenting colon glands in pathological images may include: a processor 1001, such as a central processing unit (CPU), a communication bus 1002, a user interface 1003, a network interface 1004, and a memory 1005. Among them, the communication bus 1002 is used to realize the connection and communication between these components. The user interface 1003 may include a display screen (Display) and an input unit such as a keyboard (Keyboard). Optionally, the user interface 1003 may further include a standard wired interface and a wireless interface. The network interface 1004 may optionally include a standard wired interface and a wireless interface (such as a wireless-fidelity (WI-FI) interface). The memory 1005 may be a high-speed random access memory (Random Access Memory, RAM), or a stable non-volatile memory (Non-Volatile Memory, NVM), such as a disk memory. Optionally, the memory 1005 may also be a storage device independent of the aforementioned processor 1001.

[0021] Those skilled in the art can understand that Figure 1 the structure shown in

[0022] does not constitute a limitation on the device for segmenting colon glands in pathological images, and may include more or fewer components than shown in the figure, or combine some components, or have different component arrangements. Figure 1As shown in the figure, the memory 1005, which is a computer-readable storage medium, may include an operating system, a network communication module, a user interface module, and a pathological image colon gland segmentation program.

[0023] In Figure 1 In the pathological image colon gland segmentation device shown in the figure, the network interface 1004 is mainly used for data communication with a network server; the user interface 1003 is mainly used for data interaction with a user; the processor 1001 and the memory 1005 in the pathological image colon gland segmentation device of the present invention may be arranged in the pathological image colon gland segmentation device. The pathological image colon gland segmentation device calls the pathological image colon gland segmentation program stored in the memory 1005 through the processor 1001 and executes the pathological image colon gland segmentation method provided by the embodiments of the present invention.

[0024] The embodiments of the present invention provide a pathological image colon gland segmentation method. Referring to Figure 2 , Figure 2 is a schematic flowchart of the first embodiment of the pathological image colon gland segmentation method of the present invention.

[0025] In the first embodiment, the pathological image colon gland segmentation method includes the following steps: Step S10: Preprocess the original pathological image to obtain a candidate image sample.

[0026] It should be understood that the execution subject of this embodiment may be a computing service device with data processing, network communication, and program running functions, such as a tablet computer, a personal computer, a mobile phone, etc., or a terminal electronic device capable of implementing the above functions. Hereinafter, the pathological image colon gland segmentation device (referred to as the segmentation device for short) is taken as an example to illustrate this embodiment and the following embodiments.

[0027] It should be noted that the preprocessing may include noise reduction processing and data enhancement processing on the original pathological image, so as to improve the data quality of the image sample, improve the segmentation efficiency of the pathological image, and the gland segmentation accuracy.

[0028] Further, in order to improve the image segmentation efficiency and reduce the segmentation, and reduce the influence of different dyes on the image segmentation, the above step S10 may include: Step S101: Perform noise reduction processing on the original pathological image; Step S102: Perform staining normalization on the original pathological image after noise reduction processing to obtain a stained-normalized pathological image; Step S103: Perform data enhancement processing on the stained-normalized pathological image, and generate an initial image dataset based on the data-enhanced stained-normalized pathological image; Step S104: Perform image feature serialization processing on the initial image dataset to obtain candidate image samples.

[0029] It should be noted that the initial image dataset includes multiple initial pathological image samples, and the data augmentation processing includes image rotation, Gaussian noise addition, random occlusion, and random contrast enhancement.

[0030] In some embodiments, due to the concentration of the H&E reagent used and the differences in the image generation devices, the original pathological images often have problems in effectively extracting image features, seriously affecting the segmentation accuracy. For the original pathological images of colon glands, the segmentation device can use the Staintools standard library and adopt the vahadane staining method to perform staining normalization on all colon gland pathological images.

[0031] In addition, data augmentation processing of the data is also required. To solve the problem of the small number of samples, the segmentation device can perform data augmentation operations to expand the data training samples. The Imgaug data augmentation library is used to implement operations such as rotation, Gaussian noise addition, random occlusion, and random contrast enhancement to construct a high-quality dataset with rich samples and consistent staining conditions.

[0032] It can be understood that this embodiment can be applied to the training of deep learning models. If the training data volume is too small, it will make the network training difficult and prone to overfitting. Especially in medical images, the high annotation cost results in very little labeled data available for training. Increasing the number of datasets can improve the generalization and segmentation performance of the deep neural network. In this embodiment, the original pathological images are denoised, the denoised original pathological images are stained and normalized to obtain stained and normalized pathological images, the stained and normalized pathological images are subjected to data augmentation processing, and an initial image dataset is generated based on the data-augmented stained and normalized pathological images. By normalizing and expanding the colon pathological image data, a high-quality dataset with rich samples and consistent staining conditions is constructed to make it compatible with the deep learning model.

[0033] In specific implementation, the segmentation device can input the initial image dataset obtained after preprocessing into the image patch embedding layer, and divide the image into 4x4 non-overlapping image patches. This process will convert the image size to , and then use layer normalization to normalize the image patches , and then input them into the encoder for feature extraction. The input feature map X undergoes layer normalization, and the layer normalization process refers to the following formula: Among them, The features after layer normalization The image patches after image size conversion and represent the mean and standard deviation respectively and represent the trainable scaling parameter and translation parameter respectively

[0034] Step S20: Feature extraction is performed on the candidate image samples based on the first processing branch and the second processing branch respectively to obtain a first candidate feature map and a second candidate feature map

[0035] In a specific implementation, after the original pathological image undergoes preprocessing and layer normalization, the input feature map is divided into a first processing branch and a second processing branch for feature extraction

[0036] Furthermore, in order to balance the linear transfer and saturation characteristics and enhance the fitting ability for complex images, the above step S20 may include Step S201: Feature extraction is performed on the candidate image samples based on the first processing branch through a linear layer and an activation function to obtain a first candidate feature map .

[0037] It should be noted that in the first processing branch, the input candidate image samples pass through a linear layer and then through an activation function .

[0038] The activation function has the following formula The normalized input feature map first passes through a linear layer and an activation function for processing, and the formula is as follows where represents the first candidate feature map represents the candidate image sample represents the activation function represents the weight matrix represents the bias vector

[0039] Step S202: Feature extraction is performed on the candidate image samples based on the second processing branch through a linear layer, a depthwise separable convolutional layer and an activation function to obtain a second candidate feature map

[0040] It should be noted that in the second branch, the input candidate image samples are processed through a linear layer, a depthwise separable convolution and an activation function, with reference to the following formula Among them, represents the second candidate feature map, represents the feature map that has not undergone depthwise separable convolution operation, represents the weight matrix, represents the bias vector, represents the depthwise separable convolution operation.

[0041] Step S30: Extract features of the second candidate feature map in multiple directions, and aggregate the feature extraction results in each direction to obtain global feature information.

[0042] In some embodiments, the segmentation device can extract features of the second candidate feature map in multiple directions such as horizontal, vertical, diagonal, etc., input the one-dimensional sequence features in each direction into the S6 module for global modeling, and then aggregate the feature extraction results in each direction to obtain global feature information.

[0043] Further, in order to accurately capture the global feature information in the second candidate feature map, the above step S30 may include: Step S301: Expand the second candidate feature map in multiple directions to obtain one-dimensional features in multiple directions; Step S302: Input the one-dimensional features into multiple linear layers, and update the hidden state based on the learnable matrix parameters output by the multiple linear layers; Step S303: Obtain global feature information based on the updated hidden state, the one-dimensional features, and the output learnable matrix parameters of each linear layer; Step S304: Aggregate the global features after global modeling of the one-dimensional features in multiple directions to obtain two-dimensional features.

[0044] It should be noted that in the second branch, the input is processed through linear layers, depthwise separable convolution, and activation functions, and then further feature screening is performed through a pre-constructed CSS2D module. The CSS2D module consists of a traditional SS2D branch and a local spatial block. This structure can synergistically optimize global and local feature representations, thereby enhancing the model's multi-scale perception ability of glandular structures and improving the segmentation accuracy.

[0045] It can be understood that the CSS2D module focuses on extracting and enhancing the expression of specific categories or features through information in different channels. This module consists of three parts: a scanning expansion operation, an S6 module, and a scanning merge operation. Refer to Figure 3 , Figure 3It is a schematic diagram of the structure of the CSS2D module. The scan expansion unfolds the second candidate feature map in four different directions (from top left to bottom right, from bottom right to top left, from top right to bottom left, and from bottom left to top right), forming multiple feature sequences to achieve dimensionality reduction of the feature map. These sequences are used by the S6 module to extract features, ensuring that information from all directions is thoroughly scanned, thereby capturing different features. The S6 module extracts information through a selective scanning mechanism, and the specific formula is as follows: Among them, 、 and respectively represent the learnable matrix parameters output by the linear layer, represents the learnable matrix parameter output by the linear layer, respectively represent the learnable matrix parameters after discretization, represents the derivative of the hidden state at the current moment, represents the hidden state at the previous moment, represents the two-dimensional feature of the current input, represents the identity matrix, represents the initial matrix.

[0046] The hidden state is recursively updated with reference to the following formula: Among them, represents the learnable matrix parameter output by the linear layer, represents the extracted global feature information, and the final output is the sequence . Subsequently, the global features after global modeling of the one-dimensional features in multiple directions are aggregated to obtain two-dimensional features.

[0047] It should be understood that the CSS2D module also includes a local space module. Referring to Figure 3 , the local space module is used to perform local feature extraction on the second candidate feature map, obtain local feature information, and aggregate the local feature information with the global feature information to obtain a second target feature map.

[0048] It should be noted that in this embodiment, the CSS2D module performs two-way scanning in the vertical and horizontal directions, mainly focusing on long-distance spatial relationships to capture the dependencies between distant regions in the feature map. However, there are often problems with weak connections in local regions of the feature map. Especially in small-scale regions, the feature map may not fully express the complex associations between regions, resulting in insufficient expression of local information. To address this problem of insufficient local feature information, this embodiment introduces a spatial attention mechanism (SAM), which improves the model's performance at the detail level by enhancing the connections between local regions. Specifically, the spatial attention mechanism first extracts the global information of the feature map through average and maximum statistical operations among channels. The channel maximum statistical operation captures the most significant responses of each spatial position in different channels, while the channel average statistical operation provides global background information. Then, the features obtained from these two operations are fused and processed through convolution to generate a spatial attention map, reflecting the importance of each spatial position.

[0049] Different from traditional spatial attention mechanisms, the spatial attention mechanism in this embodiment uses the features obtained from max pooling and average pooling as weights to weight the original feature map pixel by pixel. The original feature map is multiplied element-wise with the calculated spatial attention map to highlight the responses in important regions of the feature map while suppressing the influence of irrelevant regions. This process enhances the connections between local regions, especially in regions with weak local features, effectively compensating for their insufficient local information.

[0050] Finally, the local feature map enhanced by spatial attention is added to the feature map of the main branch, thus completing the operation of the entire module. Through this weighted fusion, the features of the main branch are finely adjusted and enhanced, making the local information more abundant and the feature expression more accurate. This design can effectively improve the model's performance in tasks such as gland segmentation. Especially when dealing with complex pathological images, it can highlight details more, improving the segmentation accuracy and robustness. Refer to Figure 3 , Figure 3 shows the CSS2D structure proposed in this embodiment. This figure details the two-way scanning main branch and the local feature branch, as well as how the feature fusion method finally achieves optimized feature expression.

[0051] Step S40: Perform local feature extraction on the second candidate feature map through convolution operations to obtain local feature information, and aggregate the local feature information with the global feature information to obtain a second target feature map.

[0052] In some embodiments, the segmentation device may perform local spatial feature extraction in a small range through a local spatial module, that is, capture local fine features in the spatial dimension through convolution to enhance the understanding of different spatial positions and structures. For example, through the convolution operation of a small-size convolution kernel, local feature extraction is performed on the second candidate feature map to obtain local feature information, and then the local feature information is aggregated with the global feature information to obtain a second target feature map, enhancing the connection between local regions and improving the performance at the detail level in the image segmentation process.

[0053] Step S50: Perform feature fusion on the first candidate feature map and the second target feature map to obtain a target fusion feature.

[0054] In some embodiments, the segmentation device may fuse the first candidate feature map with the second target feature map to generate a target fusion feature wherein, represents element-wise multiplication.

[0055] Step S60: Decode the target fusion feature and project the decoding result to the same dimension as the original pathological image to obtain the segmentation result of the colon glands in the pathological image.

[0056] In a specific implementation, the segmentation device may decode the target fusion feature. After decoding, by adding the encoder features in the skip connection and using the final projection layer to restore the size of the features to match the segmentation target, the height and width of the features are restored by upsampling the image patch by 4 times, and then the number of channels is restored by the projection layer, thereby outputting an image, that is, obtaining the segmentation result of the colon glands in the pathological image.

[0057] Further, in order to effectively remove the noise and irregular regions in the segmentation result and further improve the coherence and accuracy of the segmentation effect, the segmentation device may perform post-processing on the decoding result. Referring to Figure 4 Figure 4 which is a schematic diagram of the post-processing flow in an embodiment, the above step S60 may include: Step S601: Decode the target fusion feature and project the decoding result to the same dimension as the original pathological image to obtain an initial segmentation result; Step S602: Label each connected region in the initial segmentation result to obtain a plurality of labeled regions; ​Step S603: Screen out target regions from the labeled regions based on the areas of the labeled regions, and remove the labeled regions other than the target regions from the initial segmentation result to obtain a candidate segmentation result; Step S604: Perform filtering processing on the candidate segmentation result, and perform hole filling on the filtered candidate segmentation result to obtain a segmentation result of colon glands in the pathological image.

[0058] In specific implementation, the segmentation device first marks the initial segmentation result, independently labels each connected region, and then removes regions smaller than a set threshold by calculating the area of each region, effectively removing noise or false positive regions of mis-segmentation. Then, median filtering is used to smooth the segmentation result to eliminate rough parts in the segmentation boundary and reduce the influence of small-range noise on the segmentation result. Finally, through hole filling technology, small holes that may exist inside the gland region are filled to ensure the integrity and coherence of the segmentation result. The implementation of this series of post-processing operations not only optimizes the accuracy of the segmentation result, but also improves the ability to capture image details. Especially when processing complex pathological images, it can effectively make up for problems such as uneven boundaries and incomplete regions that occur in the model segmentation process.

[0059] In this embodiment, candidate image samples are obtained by preprocessing the original pathological image; first candidate feature maps and second candidate feature maps are obtained by respectively performing feature extraction on the candidate image samples based on a first processing branch and a second processing branch; the second candidate feature map is subjected to feature extraction in multiple directions, and the feature extraction results in each direction are aggregated to obtain global feature information; local feature information is obtained by performing local feature extraction on the second candidate feature map through a convolution operation, and the local feature information is aggregated with the global feature information to obtain a second target feature map; the first candidate feature map is fused with the second target feature map to obtain a target fusion feature; the target fusion feature is decoded, and the decoding result is projected onto the same dimension as the original pathological image to obtain a segmentation result of colon glands in the pathological image; since this embodiment performs feature extraction on the candidate image based on two processing branches, and respectively extracts the global feature information and local feature information of the second candidate feature map based on the second processing branch, realizing collaborative optimization of global and local feature representations, thereby enhancing the multi-scale perception ability of the gland structure, improving the segmentation accuracy, and fusing the first candidate feature map with the second target feature map, thus making up for the problem of loss of small-range feature correlation relationships in the single-branch linear scanning process, enhancing the global dependence and local detail perception ability of the image, significantly improving the ability to process image details, and thus improving the accuracy of colon gland segmentation in pathological images.

[0060] ReferenceFigure 5 , Figure 5 It is a schematic diagram of the network architecture of the image segmentation model in the second embodiment of the method for segmenting colon glands in pathological images of the present invention.

[0061] Based on the above embodiment, in the second embodiment, the step S10 further includes: Inputting the candidate image samples into an image segmentation model for segmenting colon glands in pathological images to obtain the segmentation results of colon glands in pathological images.

[0062] It should be noted that in this embodiment, the pre-processed pathological images can be segmented for colon glands through a pre-constructed image segmentation model. The image segmentation model in this embodiment can be an SVM-UNet neural network model.

[0063] It should be noted that the image segmentation model includes an image patch embedding module, multiple encoders, multiple decoders corresponding to the multiple encoders, and a hybrid dilated convolution module. The hybrid dilated convolution module is connected between the encoder and the decoder; the image patch embedding module is used to perform image feature serialization and layer normalization processing to obtain candidate image samples.

[0064] It should be noted that the Ls-VSS module is the core module of SVM-UNet. Referring to Figure 6 , Figure 6 As the schematic diagram of the processing flow of the Ls-VSS module, the encoder respectively performs feature extraction on the candidate image samples based on the first processing branch and the second processing branch to obtain the first candidate feature map and the second candidate feature map. The second candidate feature map is subjected to feature extraction in multiple directions, and the feature extraction results in each direction are aggregated to obtain global feature information. Local feature information is obtained by performing local feature extraction on the second candidate feature map through convolution operations, and the local feature information is aggregated with the global feature information to obtain the second target feature map. The first candidate feature map is fused with the second target feature map to obtain the target fusion feature. The target fusion feature output by the encoder and / or the multi-scale perception feature output by the hybrid dilated convolution module are decoded, and the decoding result is projected to the same dimension as the original pathological image to obtain the segmentation results of colon glands in pathological images.

[0065] It can be understood that, as shown in Figure 5 , the encoder consists of four stages. An image patch merging operation is applied at the end of the first three stages to reduce the height and width of the input features while increasing the number of channels. The Ls-VSS module of [2, 2, 2, 2] is adopted in the four stages, and the number of channels in each stage is [C, 2C, 4C, 8C].

[0066] After the output of the fourth layer of the encoder, it will be processed layer by layer through three-layer hybrid dilated convolution. Different dilation rates (set to 1, 2, 4) are used for detail processing. While keeping the size of the feature map basically unchanged, the receptive field is expanded to capture a larger range of context information, enhancing the perception ability of different-scale features.

[0067] After completing the feature extraction of the hybrid dilated convolution module, this feature is input into the decoder to start decoding. Combining the feature map processed by dilated convolution and the upsampling information of the decoder can capture image details more comprehensively, making the restored resolution contain richer context information.

[0068] The decoder is also divided into four stages. Starting from the penultimate stage, the image patch expansion operation is used to reduce the number of feature channels and increase the height and width. In these four stages, the Ls-VSS module with a scale of [2, 2, 2, 1] is used, and the number of channels in each stage is [8C, 4C, 2C, C].

[0069] After the decoder, by adding the features of the encoder in the skip connection, the final projection layer is used to restore the size of the features to match the segmentation target. The height and width of the features are restored by performing 4-fold upsampling through image patch expansion, and then the number of channels is restored through the projection layer, thereby outputting the image.

[0070] It should be noted that the hybrid dilated convolution module is used to aggregate the target fusion features output by the encoder and the multi-scale convolution output features to obtain multi-scale perception features, and input the multi-scale perception features into the decoder for decoding. The hybrid dilated convolution module realizes multi-scale feature extraction through three consecutive convolutional layers, in which dilated convolutions with dilation rates of 1, 2, and 4 are respectively used to expand the receptive field. In addition, after the multi-scale features are concatenated in the channel dimension, they are compressed through 1×1 convolution to avoid feature redundancy and improve the efficiency of feature expression at the same time. Refer to Figure 7 , Figure 7 is the structural schematic diagram of the hybrid dilated convolution module. A residual connection is also designed inside the hybrid dilated convolution module to add the input features and the output features after multi-scale convolution. This structure not only effectively guarantees the transmission of the original feature information, but also alleviates the problem of gradient disappearance that may occur in deep networks, thus significantly improving the stability of network training.

[0071] It should be noted that the transition layer plays a crucial role in connecting the upper and lower parts in the deep network structure. Introducing the hybrid convolution module in SVM-UNet not only increases the network depth, but also further enhances the model's ability to express detail features. The core idea of this module is to enhance the network's capture of detail information through multi-scale dilated convolution combined with the residual structure, while maintaining the integrity and diversity of feature expression.

[0072] The hybrid convolution module achieves multi-scale feature extraction through three consecutive convolutional layers, where dilated convolutions with dilation rates of 1, 2, and 4 are respectively used to expand the receptive field. The design of the dilation rate avoids the information loss or insufficiency that may be caused by a single dilation rate, making the model perform more stably and accurately when dealing with the gland segmentation task with complex and diverse deformations. In addition, this progressive dilation rate setting also reduces the problem of incomplete features caused by excessive sparsity. The introduction of dilated convolution enables the module to capture a wider range of context information, thus effectively handling complex problems such as deformation and blurred boundaries in the gland segmentation task. In addition, after the multi-scale features are concatenated in the channel dimension, they are compressed through 1×1 convolution to avoid feature redundancy and improve the efficiency of feature expression at the same time.

[0073] A residual connection is also designed inside the module to add the input features to the output features after multi-scale convolution. This structure not only effectively ensures the transmission of the original feature information, but also alleviates the problem of vanishing gradients that may occur in deep networks, thus significantly improving the stability of network training. In addition, by combining the use of normalization operations and the SiLU activation function, the module can dynamically adjust the feature distribution and further enhance the feature expression ability. In the characterization of complex patterns, the module demonstrates excellent adaptability and robustness, laying a solid foundation for the network to capture richer multi-scale semantic information.

[0074] In some embodiments, the segmentation device can train the image segmentation model using a combined loss function, which is composed of a Dice loss function and a cross-entropy loss function. The formula is as follows: The Dice loss function is the complement of the Dice coefficient. By minimizing the Dice Loss, the model directly optimizes the overlap degree of the segmentation results. Its specific formula is: The cross-entropy loss function pays more attention to the pixel-level classification accuracy, that is, it tries to achieve correct classification at each pixel point of the image. This refined characteristic makes it have an advantage when dealing with segmentation tasks with rich details and complex structures.

[0075] The specific formula of the cross-entropy loss function is: In some embodiments, during the training of an image segmentation model, if the amount of training data is too small, it will make network training difficult and prone to overfitting. Especially in medical images, the high annotation cost results in very little labeled data available for training. Increasing the number of data sets can improve the generalization and segmentation performance of deep neural networks. The segmentation device can adopt a series of data processing methods: denoising, stain normalization, data augmentation, etc. to normalize and expand the colon pathological image data, and construct a high-quality data set with rich samples and consistent staining conditions to make it compatible with the deep learning model.

[0076] In some embodiments, taking the construction of an image segmentation model with SVM-UNet as an example, through qualitative analysis, the segmentation results of SVM-UNet demonstrate its advantages in dealing with complex glandular morphology and boundary details. Especially in terms of complex boundaries and deformation degree, it outperforms the comparison model. However, qualitative analysis mainly relies on visual comparison. Although it can intuitively reflect the performance of the model, it lacks quantitative support. Therefore, in order to more comprehensively and objectively evaluate the segmentation performance of SVM-UNet, this embodiment will use multiple evaluation metrics for quantitative analysis, and further verify the effectiveness and superiority of the model in the pathological image segmentation task through metric comparison. Tables 1 and 2 respectively show the quantitative evaluation results of the image segmentation model on the GlaS data set and the CRAG data set.

[0077] Table 1. Quantitative evaluation result table of GlaS data set Referring to Table 1 above, on the GlaS data set, SVM-UNet shows significant advantages. Compared with the basic network VM-UNet, the F1 coefficient in TestA increased by 2.5% (from 0.921 to 0.944), the Object-Dice coefficient increased by 1.4% (from 0.927 to 0.940), and the Hausdorff distance decreased by 6.1% (from 35.172 to 33.032). Compared with DE-MambaUNet, the F1 coefficient increased by 0.9%, the Object-Dice coefficient increased by 0.6%, and the Hausdorff distance decreased significantly by 51.0%. In TestB, the F1 coefficient of SVM-UNet increased by 1.6% compared with VM-UNet, and the Object-Dice coefficient increased by 0.2%. Although the F1 coefficient is slightly lower than that of DE-MambaUNet, the Object-Dice coefficient remains the same, and the Hausdorff distance decreased by 25.3%, verifying the model's ability to capture overall features.

[0078] Table 2. Quantitative evaluation result table of CRAG data set Referring to Table 2 above, on the CRAG dataset, SVM-UNet also performs excellently. The Object-Dice coefficient is on par with DE-MambaUNet (0.933), slightly higher than VM-UNet (0.929), with a 0.4% improvement; the F1 coefficient is 0.933, representing a 1.3% improvement compared to VM-UNet. In addition, its Hausdorff distance is only 26.249, which is a 34.1% and 57.8% reduction compared to VM-UNet and DE-MambaUNet respectively, fully demonstrating its precise ability to capture gland boundaries.

[0079] Among them, the F1 coefficient is mainly used to evaluate the detection accuracy of the algorithm for individual glands. The closer it is to 1, the better the segmentation algorithm, and it takes into account both precision and recall.

[0080] The Object-Dice coefficient represents the Object-Dice coefficient, which is calculated for each individual object and focuses on the segmentation quality of each individual object, making it more suitable for the gland segmentation task. The value range of the Object-Dice coefficient is between 0 and 1, and the higher the value, the more accurate the segmentation result.

[0081] The Hausdorff distance represents the Object-Hausdorff distance, which is used to evaluate the shape similarity between the segmentation result and the ground truth annotation. Shape similarity is a very important consideration in gland segmentation.

[0082] Through the qualitative and quantitative experimental results, the overall effectiveness of SVM-UNet has been verified. To further quantify the specific contributions of each key module in the model to performance improvement, in some embodiments, ablation experiments can be designed to deeply analyze the innovative structure of the model by gradually adding modules. The experiment starts from the VM-UNet model as a benchmark for comparison. First, the hybrid dilated convolution module is added to enhance the ability to extract global features and evaluate its effect on improving segmentation accuracy. On this basis, the Ls-VSS module is used to replace the original VSS module to further enhance the ability to capture local spatial information and examine its role in improving boundary capture accuracy and segmentation result coherence. Finally, post-processing operations are added to the model containing the above modules to analyze its contribution to result smoothness optimization and small noise suppression.

[0083] To more intuitively show the impact of each module on the model performance, Table 3 summarizes the quantitative results of the ablation experiments, especially using the TestA part of the GlaS dataset as a reference. By comparing the performance metrics under different configurations, the table shows the specific contributions of each module introduced to segmentation accuracy and boundary capture ability.

[0084] Table 3, Comparative Evaluation Results Table of Ablation Experiments Referring to Table 3 above, it can be seen from the ablation experiment that as the number of modules gradually increases, the model performance is gradually improved. After introducing the Hybrid Atrous Convolution module (HAC), the F1 coefficient is significantly improved, the Object-Dice coefficient is slightly improved, and the Hausdorff distance decreases, indicating that the boundary capture ability is enhanced and the feature extraction ability of the network is greatly improved. After further adding the Ls-VSS module, with the increase of local spatial information, although the F1 coefficient decreases slightly, the Object-Dice coefficient and the Hausdorff distance are further optimized, improving the segmentation accuracy and the accuracy of the boundary. SVM-UNet has achieved excellent performance in all indicators, thanks to the effective fusion and synergy of each module.

[0085] The ablation experiment effectively verifies the overall performance of the SVM-UNet structure. At the same time, in some embodiments, different parameters in each module can also be experimented to confirm their optimal configuration. Table 4 shows the quantitative experimental results of the normalization method and convolution kernel size used in the Hybrid Atrous Convolution module (HAC) on the model performance. The data in the table are the convolution kernel sizes used. Table 5 shows the impact of different configurations on the model performance in the Ls-VSS module.

[0086] Table 4, Comparative Table of Different Parameter Results in HAC Table 5, Comparative Table of Different Parameter Results in Ls-VSS Module Referring to Table 4 and Table 5 above, it can be seen from the ablation experiment that the parameters of the HAC and Ls-VSS modules adopted by SVM-UNet are optimal and can obtain the best performance.

[0087] In this embodiment, by introducing the Hybrid Atrous Convolution module into the image segmentation model, the feature map is processed by convolution with different dilation rates to extract multi-scale detailed information. This module is stacked between the encoder and the decoder to retain the subtle features in the feature map, greatly improving the accuracy of gland segmentation. Finally, by introducing a local feature acquisition branch, this branch makes up for the small-scale feature association relationship lost in the linear scanning process and enhances the model's perception ability of local details. This improvement significantly enhances the model's ability to process details, thereby improving the accuracy of pathological image analysis.

[0088] In addition, an embodiment of the present invention also provides a computer-readable storage medium, on which a pathological image colon gland segmentation program is stored. When the pathological image colon gland segmentation program is executed by a processor, the steps of the pathological image colon gland segmentation method described above are implemented.

[0089] The computer-readable storage medium provided by this application can be, for example, a USB flash drive, but is not limited to electrical, magnetic, optical, electromagnetic, infrared, or semiconductor systems, devices, or any combination of the above. More specific examples of computer-readable storage media may include, but are not limited to: electrical connections with one or more wires, portable computer disks, hard disks, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM) or flash memory, optical fibers, portable compact disk read-only memory (CD-ROM), optical storage devices, magnetic storage devices, or any suitable combination of the above. In this embodiment, the computer-readable storage medium can be any tangible medium that contains or stores a program, and this program can be used by or in combination with an instruction execution system, device, or device. The program code contained on the computer-readable storage medium can be transmitted by any suitable medium, including but not limited to: wires, optical cables, RF (Radio Frequency), etc., or any suitable combination of the above.

[0090] The above computer-readable storage medium can be included in the pathological image colon gland segmentation device; it can also exist separately and not be assembled into the pathological image colon gland segmentation device.

[0091] In addition, an embodiment of the present invention also provides a computer program product, including a pathological image colon gland segmentation program. When the pathological image colon gland segmentation program is executed by a processor, the steps of the pathological image colon gland segmentation method described above are implemented.

[0092] The specific implementation manner of the computer program product of the present invention is basically the same as that of each embodiment of the above pathological image colon gland segmentation method, and will not be elaborated here.

[0093] Refer to Figure 8 , Figure 8 which is a structural block diagram of an embodiment of the pathological image colon gland segmentation device of the present invention.

[0094] As Figure 8 shown, the pathological image colon gland segmentation device proposed by the embodiment of the present invention includes: The image processing module 10 is used to preprocess the original pathological image to obtain candidate image samples; The multi-branch feature extraction module 20 is used to extract features from the candidate image samples respectively based on the first processing branch and the second processing branch to obtain a first candidate feature map and a second candidate feature map; The global feature extraction module 30 is used to extract features from the second candidate feature map in multiple directions and aggregate the feature extraction results in each direction to obtain global feature information; The local feature extraction module 40 is used to perform local feature extraction on the second candidate feature map through convolution operations, obtain local feature information, and aggregate the local feature information with the global feature information to obtain a second target feature map; The feature fusion module 50 is used to fuse the features of the first candidate feature map and the second target feature map to obtain a target fusion feature; The feature decoding module 60 is used to decode the target fusion feature and project the decoding result to the same dimension as the original pathological image to obtain the segmentation result of the colon glands in the pathological image.

[0095] In this embodiment, candidate image samples are obtained by preprocessing the original pathological image; features are extracted from the candidate image samples respectively based on the first processing branch and the second processing branch to obtain a first candidate feature map and a second candidate feature map; features are extracted from the second candidate feature map in multiple directions and the feature extraction results in each direction are aggregated to obtain global feature information; local feature extraction is performed on the second candidate feature map through convolution operations to obtain local feature information, and the local feature information is aggregated with the global feature information to obtain a second target feature map; the features of the first candidate feature map and the second target feature map are fused to obtain a target fusion feature; the target fusion feature is decoded and the decoding result is projected to the same dimension as the original pathological image to obtain the segmentation result of the colon glands in the pathological image; since this embodiment extracts features from the candidate image based on two processing branches and extracts the global feature information and local feature information of the second candidate feature map respectively based on the second processing branch, it realizes the collaborative optimization of global and local feature representations, thereby enhancing the multi-scale perception ability of the gland structure, improving the segmentation accuracy, fusing the first candidate feature map and the second target feature map, thereby making up for the problem of the loss of small-scale feature correlation relationships in the single-branch linear scanning process, enhancing the global dependence and local detail perception ability of the image, significantly improving the processing ability of image details, and thus improving the accuracy of colon gland segmentation in pathological images.

[0096] The colon gland segmentation device for pathological images provided by this application adopts the colon gland segmentation method for pathological images in the above-mentioned embodiment, and can solve the technical problem of colon gland segmentation in pathological images. Compared with the prior art, the beneficial effects of the colon gland segmentation device for pathological images provided by this application are the same as those of the colon gland segmentation method for pathological images provided by the above-mentioned embodiment, and other technical features in the colon gland segmentation device for pathological images are the same as the features disclosed in the method of the above-mentioned embodiment, and will not be elaborated here.

[0097] It should be understood that the above is only for illustration and does not constitute any limitation to the technical solution of the present invention. In specific applications, those skilled in the art can set according to needs, and the present invention does not limit this.

[0098] It should be noted that the above-described work process is only illustrative and does not limit the protection scope of the present invention. In actual applications, those skilled in the art can select some or all of them according to actual needs to achieve the purpose of the solution of this embodiment, and there is no limitation here.

[0099] In addition, for the technical details not described in detail in this embodiment, reference can be made to the colon gland segmentation method for pathological images provided in any embodiment of the present invention, and details will not be repeated here.

[0100] It should be noted that in this article, the terms "include", "comprise" or any other variant thereof are intended to cover non-exclusive inclusion, so that a process, method, article or system including a series of elements not only includes those elements, but also includes other elements not explicitly listed, or also includes elements inherent to such process, method, article or system. Without further limitation, an element defined by the statement "including one..." does not exclude the existence of another identical element in the process, method, article or system including that element.

[0101] The serial numbers of the above-mentioned embodiments of the present invention are only for description and do not represent the advantages or disadvantages of the embodiments.

[0102] Through the description of the above embodiments, those skilled in the art can clearly understand that the above-mentioned embodiment methods can be implemented by means of software plus a necessary general hardware platform. Of course, they can also be implemented by hardware, but in many cases the former is a better implementation method. Based on such an understanding, the technical solution of the present invention, in essence, or the part that contributes to the prior art can be embodied in the form of a software product. This computer software product is stored in a storage medium (such as read-only memory / random access memory, magnetic disk, optical disk), and includes several instructions to enable a terminal device (which can be a mobile phone, computer, server, air conditioner, or network device, etc.) to execute the methods described in various embodiments of the present invention.

[0103] The above are only the preferred embodiments of the present invention, and do not limit the patent scope of the present invention accordingly. Any equivalent structure or equivalent process transformation made by using the content of the specification and drawings of the present invention, or directly or indirectly applied in other related technical fields, shall be similarly included in the patent protection scope of the present invention.

Claims

1. A method for segmenting colon glands in pathological images, characterized in that, The method for segmenting colon glands in pathological images includes: Preprocessing the original pathological image to obtain candidate image samples; Performing feature extraction on the candidate image samples respectively based on a first processing branch and a second processing branch to obtain a first candidate feature map and a second candidate feature map; Performing feature extraction on the second candidate feature map in multiple directions, and aggregating the feature extraction results in each direction to obtain global feature information; Performing local feature extraction on the second candidate feature map through a convolution operation to obtain local feature information, and aggregating the local feature information with the global feature information to obtain a second target feature map; Performing feature fusion on the first candidate feature map and the second target feature map to obtain a target fusion feature; Decoding the target fusion feature and projecting the decoding result to the same dimension as the original pathological image to obtain the segmentation result of colon glands in the pathological image.

2. The method for segmenting colon glands in a pathological image according to claim 1, wherein The preprocessing of the original pathological image to obtain candidate image samples includes: Performing denoising processing on the original pathological image; Performing stain normalization on the denoised original pathological image to obtain a stain-normalized pathological image; Performing data augmentation processing on the stain-normalized pathological image, and generating an initial image dataset based on the stain-normalized pathological image after data augmentation, where the initial image dataset includes multiple initial pathological image samples; Performing image feature serialization processing on the initial image dataset to obtain candidate image samples.

3. The pathological image colon gland segmentation method according to claim 2, wherein, The performing feature extraction on the candidate image samples respectively based on a first processing branch and a second processing branch to obtain a first candidate feature map and a second candidate feature map includes: Based on the first processing branch, feature extraction is performed on the candidate image sample through a linear layer and an activation function to obtain a first candidate feature map : Among them, represents the first candidate feature map, represents the candidate image sample, represents the activation function, represents the weight matrix, represents the bias vector; Based on the second processing branch, feature extraction is performed on the candidate image sample through a linear layer, a depthwise separable convolutional layer, and an activation function to obtain a second candidate feature map : Among them, represents the second candidate feature map, represents the feature map that has not undergone the depthwise separable convolution operation, represents the weight matrix, represents the bias vector, represents the depthwise separable convolution operation.

4. The pathological image colon gland segmentation method according to claim 3, wherein, The performing feature extraction on the second candidate feature map in multiple directions, and aggregating the feature extraction results in each direction to obtain global feature information includes: Unfolding the second candidate feature map in multiple directions to obtain one-dimensional features in multiple directions; Inputting the one-dimensional features into multiple linear layers, and updating the hidden state based on the learnable matrix parameters output by the multiple linear layers: Among them, , and respectively represent the learnable matrix parameters output by the linear layer, represents the learnable matrix parameters output by the linear layer, respectively represent the learnable matrix parameters after discretization, represents the derivative of the hidden state at the current moment, represents the hidden state at the previous moment, represents the two-dimensional feature of the current input, represents the identity matrix, represents the initial matrix; Obtaining global feature information based on the updated hidden state, the one-dimensional features, and the output learnable matrix parameters of each linear layer: Among them, represents the learnable matrix parameter of the linear layer output, represents the extracted global feature information; Aggregating the global features after global modeling of the one-dimensional features in multiple directions to obtain two-dimensional features.

5. The method for segmenting colon glands in a pathological image according to any one of claims 1 to 4, characterized in that, The decoding the target fusion feature and projecting the decoding result to the same dimension as the original pathological image to obtain the segmentation result of colon glands in the pathological image includes: Decoding the target fusion feature and projecting the decoding result to the same dimension as the original pathological image to obtain an initial segmentation result; Labeling each connected region in the initial segmentation result to obtain multiple labeled regions; Screening out target regions from the labeled regions based on the areas of the labeled regions, and removing the labeled regions other than the target regions from the initial segmentation result to obtain a candidate segmentation result; Performing filtering processing on the candidate segmentation result, and performing hole filling on the candidate segmentation result after filtering processing to obtain the segmentation result of colon glands in the pathological image.

6. The pathological image colon gland segmentation method according to any one of claims 1 to 4, characterized in that, After preprocessing the original pathological image to obtain candidate image samples, the following steps are further included: Inputting the candidate image samples into an image segmentation model for colon gland segmentation of pathological images to obtain the colon gland segmentation results of pathological images; Among them, the image segmentation model includes an image patch embedding module, multiple encoders, multiple decoders corresponding to the multiple encoders, and a hybrid dilated convolution module, and the hybrid dilated convolution module is connected between the encoder and the decoder; The image patch embedding module is used to perform image feature serialization and layer normalization processing to obtain candidate image samples; The encoder is used to perform feature extraction on the candidate image samples respectively based on a first processing branch and a second processing branch to obtain a first candidate feature map and a second candidate feature map; The encoder is further used to perform feature extraction on the second candidate feature map in multiple directions and aggregate the feature extraction results in each direction to obtain global feature information; The encoder is further used to perform local feature extraction on the second candidate feature map through a convolution operation, obtain local feature information, and aggregate the local feature information with the global feature information to obtain a second target feature map; The encoder is further used to perform feature fusion on the first candidate feature map and the second target feature map to obtain a target fusion feature; The hybrid dilated convolution module is used to perform multi-scale feature perception on the target fusion feature output by the encoder to obtain multi-scale perception features, and input the multi-scale perception features into the decoder for decoding; The decoder is used to decode the target fusion feature output by the encoder and / or the multi-scale perception feature output by the hybrid dilated convolution module, and project the decoding result to the same dimension as the original pathological image to obtain the colon gland segmentation results of pathological images.

7. A pathological image colon gland segmentation device, characterized in that, The colon gland segmentation device for pathological images includes: An image processing module for preprocessing the original pathological image to obtain candidate image samples; A multi-branch feature extraction module for performing feature extraction on the candidate image samples respectively based on a first processing branch and a second processing branch to obtain a first candidate feature map and a second candidate feature map; A global feature extraction module for performing feature extraction on the second candidate feature map in multiple directions and aggregating the feature extraction results in each direction to obtain global feature information; A local feature extraction module for performing local feature extraction on the second candidate feature map through a convolution operation, obtaining local feature information, and aggregating the local feature information with the global feature information to obtain a second target feature map; A feature fusion module for performing feature fusion on the first candidate feature map and the second target feature map to obtain a target fusion feature; A feature decoding module for decoding the target fusion feature and projecting the decoding result to the same dimension as the original pathological image to obtain the colon gland segmentation results of pathological images.

8. A pathological image colon gland segmentation device, characterized in that, The pathological image colon gland segmentation device described above includes: a memory, a processor, and a pathological image colon gland segmentation program stored on the memory and executable on the processor. The pathological image colon gland segmentation program is configured to implement the pathological image colon gland segmentation method according to any one of claims 1 to 6.

9. A computer-readable storage medium, characterized in that, A pathological image colon gland segmentation program is stored on the computer-readable storage medium. When the pathological image colon gland segmentation program is executed by a processor, it implements the pathological image colon gland segmentation method according to any one of claims 1 to 6.

10. A computer program product, characterized in that, The computer program product includes a pathological image colon gland segmentation program. When the pathological image colon gland segmentation program is executed by a processor, it implements the steps of the pathological image colon gland segmentation method according to any one of claims 1 to 6.

Images