Auxiliary analysis method and system for cervical cancer multi-category histopathological images

Abstract

The invention relates to the technical field of image processing, and discloses an auxiliary analysis method and system for cervical cancer multi-class tissue pathological images, and the method comprises the steps: obtaining a pathological image data set; a cervical cancer multi-class tissue pathology image segmentation network is constructed, the network comprises a multi-class attention unit, and the multi-class attention unit comprises a multi-channel gray level attention module, a multi-class expansion extrusion attention module and a multi-class space attention module; a pathological image data set is used to train a cervical cancer multi-class tissue pathological image segmentation network, after training, a canceration region, a solid region and a background part of cervical cancer in a pathological image are segmented, and an automatic grading method is used to carry out image FIGO grading on a segmentation result. When the method is used for segmenting different categories of the cervical cancer multi-category tissue pathology images, the accuracy is high, and the operation speed is high.

Description

Technical Field

[0001] The present invention relates to the field of image processing technology, and in particular to an auxiliary analysis method and system for multi-category histopathological images of cervical cancer. Background Art

[0002] In recent years, China's socioeconomic development, lifestyle changes, increased prevalence of metabolic syndrome, and an aging population have all led to an increase in the incidence of cervical cancer, especially in young women. This trend has significantly affected women's fertility and reproductive health. Accurate staging is very important for prognosis and treatment options. Over time, the pathological diagnosis stage of cervical cancer has gradually progressed from stage I to stage IV. According to the 2023 prognosis report, within 5 years, the proportion of stage I and stage IV expanded from 68.2% to 76.9%, an increase of 8.7%. Accurate staging is very important for prognosis and treatment options. Therefore, strengthening cervical cancer screening, improving the speed and accuracy of diagnosis, and early detection and treatment are very important to reduce the mortality rate of endometrium.

[0003] The paper (Ren L, Zhao D, Zhao X, Chen W, Li L, Wu T, Liang G, Cai Z, Xu S:Multi-level thresholding segmentation for pathological images: Optimal performance design of a new modified differential evolution, Computers in Biology and Medicine (2022).) proposed a multi-stage threshold pathological image segmentation method with significant convergence accuracy and the ability to escape local optimality (LO). The paper (Automatic Brain Tumor Tissue Detection based on Marker Controlled Watershed Segmentation in T1-weighted MR Images, International Journal of Recent Trends in Engineering & Research (2017).) proposed to use the intensity and edge amplitude information in the image histogram to calculate multiple thresholds for automatic segmentation of tumor lesions from magnetic resonance images. The paper (Wang J, Liu X: Medical image recognition and segmentation of pathological slices of gastric cancer based on Deeplab v3+ neural network, Computer Methods and Programs in Biomedicine (2021).) proposed a multi-scale input Deeplab v3+ network and compared it with SegNet and ICNet in terms of sensitivity, specificity, accuracy, Dice coefficient, etc.

[0004] With the development of technology, computer-aided diagnosis technology for cancer has gradually matured. However, when the existing technology is used to segment multiple categories of tumor tissues, the characteristics of different categories of tumor tissues are often similar and difficult to distinguish, resulting in poor segmentation effect and prone to over-segmentation and mis-segmentation problems. Based on these challenges and problems, the present invention proposes corresponding methods to solve the problem of low accuracy in segmenting different categories of tumor tissues. Summary of the invention

[0005] The purpose of the present invention is to solve the problems in the prior art.

[0006] The technical solution adopted by the present invention to solve the technical problem is: to provide an auxiliary analysis method for multi-category histopathological images of cervical cancer, comprising the following steps:

[0007] Collect and annotate multi-category histopathological images of cervical cancer to obtain a pathological image dataset;

[0008] A multi-category histopathological image segmentation network for cervical cancer is constructed using DeepLabV3+ as the backbone network, wherein a multi-category attention unit is added to the encoder, and the multi-category attention unit includes a multi-channel grayscale attention module, a multi-category expansion and compression attention module, and a multi-category spatial attention module;

[0009] Use the pathology image dataset to train the cervical cancer multi-category tissue pathology image segmentation network;

[0010] The trained cervical cancer multi-category tissue pathology image segmentation network is used to segment the cancerous area, solid area and background part of the cervical cancer in the pathology image to obtain the segmentation result;

[0011] The FIGO grade of the pathological image is obtained by automatic grading method based on the segmentation results.

[0012] Preferably, the cervical cancer multi-category histopathological image segmentation network includes an encoder and a decoder;

[0013] In the encoder, the input pathological image is extracted and compressed four times using multiple parallel convolutions to obtain preliminary effective features. After inputting into the multi-feature Transformer unit, convolutions of different rates are used to connect and merge, and then 1x1 convolution is performed to compress the features, and finally multi-category fusion features are output;

[0014] In the decoder, 1×1 convolution is used to adjust the number of channels of the initial effective feature layer that is compressed twice, and the original image features are obtained using the initial effective feature layer after adjusting the number of channels; the multi-scale fusion features output by the encoding part are upsampled four times using hollow convolution, stacked with the original image features, and then processed in sequence using two depthwise separable convolutions and one 1x1 convolution to obtain the final effective features, and finally resize is used for upsampling to obtain and output a segmentation result of the same size as the input pathological image.

[0015] Preferably, the internal execution of the multi-class attention unit comprises the following steps:

[0016] First, the input image is sent to the multi-channel grayscale feature extraction module, the input image is divided into feature images of different channels according to the grayscale value, and the feature information from different grayscale channels is fused through maximum pooling and average pooling to obtain a multi-channel grayscale attention map, which is expressed as:

[0017] ;

[0018] Among them, F' represents the multi-channel grayscale attention map obtained by the multi-channel grayscale feature extraction module of the feature map F, represents the characteristic element multiplication operation, Represents a multi-channel grayscale feature extraction operation;

[0019] Secondly, in order to more accurately extract the global feature information including the edge information and morphological features of the cancerous area and the solid area in the grayscale feature map, the obtained multi-channel grayscale attention map is input into the multi-category expansion and squeeze attention module to obtain the multi-category expansion and squeeze attention map, which is expressed as:

[0020] ;

[0021] Among them, F'' represents the multi-category expansion and squeeze attention map obtained by the multi-channel grayscale attention map through the multi-category expansion and squeeze attention module, Represents a multi-category expansion-squeezing attention operation;

[0022] Again, the multi-category expansion and squeeze attention map is input into the multi-category spatial attention module to obtain the multi-category spatial attention map, which is expressed as:

[0023] ;

[0024] Among them, F''' represents the multi-category spatial attention map obtained by inputting the multi-category expansion and squeeze attention map into the multi-category spatial attention module, and y q Represents a multi-category spatial attention operation;

[0025] Finally, the final attention feature map of multi-category histopathological images of cervical cancer is obtained through residual operation, which is expressed as:

[0026] ;

[0027] Among them, Fe represents the multi-category histopathological image of cervical cancer, Represents the residual operation.

[0028] Preferably, the multi-channel grayscale feature extraction operation is expressed as:

[0029] ;

[0030] in, Represents the sigmoid function, S0 and S1 represent the weights of the maximum pooling and average pooling fusion images, GS represents the image grayscale feature extraction operation, MaxPool and AvgPool represent the maximum pooling and average pooling operations respectively; and They represent the maximum pooling and average pooling functions respectively, c represents the grayscale feature map extracted when the grayscale value of the input image is c, 0≤c≤255.

[0031] Preferably, the multi-category expansion and compression attention operation is expressed as:

[0032] ;

[0033] Among them, F sc represents the expansion operation, T n represents the feature weights of different categories of the image, T1, T2 and T3 represent the weights of the cancerous region features, solid region features and background respectively, and S2 represents the feature that suppresses the useless features in the feature map except the solid region and cancerous region after the squeezing operation. The specific formula is:

[0034] ;

[0035] Among them, W1 and W2 represent the weights of the grayscale feature map after passing through the fully connected layer. Represents the Relu activation function, z represents the result obtained after the squeezing operation, and W represents the weight of the image when it passes through the fully connected layer.

[0036] Preferably, the multi-category spatial attention operation is expressed as:

[0037] ;

[0038] Among them, σ represents the sigmoid function, Represents a convolution operation with a filter size of 7×7; and They respectively represent the feature information graphs obtained after average pooling and maximum pooling when the channel is S.

[0039] Preferably, the pathological image dataset is used to train the cervical cancer multi-category tissue pathological image segmentation network, and the loss function used is expressed as:

[0040] ;

[0041] ;

[0042] ;

[0043] Among them, Loss represents the loss function of the cervical cancer pathology image segmentation network; L represents the cross entropy loss, x i Represents the i-th element of the true label, y i represents the probability that the model predicts that x belongs to the i-th category; DL represents the Dice loss, represents the number of elements that intersect X and Y, |X| and |Y| represent the number of elements in X and Y respectively; n1 and n2 represent the loss weights respectively.

[0044] Preferably, the step of obtaining the FIGO classification of the image by an automatic classification method for the segmentation result comprises the following steps:

[0045] The number of pixels with grayscale values ​​of 67 and 128 in the image of the segmentation result is counted, and the ratio of the number of pixels in the cancerous area to the number of pixels in the solid area is calculated, which is expressed as:

[0046] ;

[0047] Among them, A represents the ratio of the number of pixels in the solid area to the number of pixels in the cancerous area, S represents the number of pixels in the solid area, and T represents the number of pixels in the cancerous area;

[0048] The corresponding FIGO grade is obtained by the ratio of the number of pixels in the solid area to the cancerous area, expressed as:

[0049] .

[0050] The present invention also provides an auxiliary analysis system for multi-category histopathological images of cervical cancer, comprising:

[0051] The acquisition module collects and annotates multi-category pathological images of cervical cancer to obtain a pathological image dataset;

[0052] A construction module is used to construct a cervical cancer multi-category histopathological image segmentation network using DeepLabV3+ as the backbone network, wherein a multi-category attention unit is added to the encoder, and the multi-category attention unit includes a multi-channel grayscale attention module, a multi-category expansion and compression attention module, and a multi-category spatial attention module;

[0053] The training module uses the pathological image dataset to train the cervical cancer multi-category tissue pathological image segmentation network;

[0054] The segmentation module uses the trained cervical cancer multi-category tissue pathology image segmentation network to segment the cancerous area, solid area and background part of the cervical cancer in the pathology image to obtain the segmentation result;

[0055] The grading module obtains the FIGO grade of the pathological image through the automatic grading method for the segmentation results.

[0056] The present invention has the following beneficial effects:

[0057] (1) The present invention proposes a segmentation network for multi-category histopathological images of cervical cancer, and uses a multi-category attention module for cervical cancer and an improved loss function to improve the accuracy of segmentation;

[0058] (2) The present invention designs a cervical cancer category attention module in the sampling process of the segmentation network to improve the network's ability to extract pathological tissue entity areas and edge features, while improving the accuracy of fusing multi-classification features, thereby improving the accuracy of the network's segmentation of multi-classification features and edge features;

[0059] (3) The present invention obtains the FIGO grade of the image by calculating the ratio of the number of pixels in the cancerous area to the number of pixels in the solid area.

[0060] The present invention is further described in detail below in conjunction with the accompanying drawings and embodiments, but the present invention is not limited to the embodiments. BRIEF DESCRIPTION OF THE DRAWINGS

[0061] Figure 1 A method step diagram of an embodiment of the present invention;

[0062] Figure 2 This is a schematic diagram of a network structure for segmenting multi-category histopathological images of cervical cancer according to an embodiment of the present invention;

[0063] Figure 3 is a schematic diagram of a pathological image according to an embodiment of the present invention;

[0064] Figure 4 A schematic diagram of a multi-category attention unit structure according to an embodiment of the present invention;

[0065] Figure 5 A schematic diagram of a segmentation result according to an embodiment of the present invention;

[0066] Figure 6 4 is a system structure diagram of an embodiment of the present invention. DETAILED DESCRIPTION

[0067] See also Figure 1 FIG. 1 is a diagram showing steps of a method according to an embodiment of the present invention, comprising the following steps:

[0068] S101, collecting and annotating multi-category histopathological images of cervical cancer to obtain a pathological image dataset;

[0069] S102, using DeepLabV3+ as the backbone network to build a cervical cancer multi-category histopathological image segmentation network, wherein a multi-category attention unit is added to the encoder, and the multi-category attention unit includes a multi-channel grayscale attention module, a multi-category expansion and compression attention module, and a multi-category spatial attention module;

[0070] S103, using the pathological image dataset to train a cervical cancer multi-category tissue pathological image segmentation network;

[0071] S104, using the trained cervical cancer multi-category tissue pathology image segmentation network to segment the cancerous region, solid region and background portion of the cervical cancer in the pathology image to obtain a segmentation result;

[0072] S105, obtaining FIGO classification of the image by using an automatic classification method for the segmentation result.

[0073] For details, see Figure 2 As shown, it is a schematic diagram of the network structure of multi-category histopathological image segmentation of cervical cancer in an embodiment of the present invention; the DeepLab V3+ network composed of an encoder and a decoder is used as the backbone network for segmentation. In the encoder, parallel convolution is used to extract the features of the initial effective feature layer compressed four times, and convolutions of different rates are used to connect and merge them respectively, and then 1x1 convolution is performed to compress the features. In the decoder, 1×1 convolution is used to adjust the number of channels of the initial effective feature layer compressed twice, so that the number of channels becomes 3, which are solid feature channels, cancer feature channels and background channels respectively, and then hollow convolution is used to stack the sampling results of the output of the encoding part to obtain the fused feature information, and two depth-separable convolutions are used for processing, and the feature information is further compressed to obtain only solid features and cancer features, and the final effective feature layer containing the details of the solid area and the cancer area is obtained, and the channel is adjusted using 1x1 convolution to obtain 1 channel after the fusion feature information. Finally, resize is used for upsampling so that the size of the final output layer is the same as the input image.

[0074] For details, see Figure 3 As shown, it is a schematic diagram of a pathological image of an embodiment of the present invention. Since panoramic digital pathological images have problems of tissue adhesion and blurred boundaries, there are small solid areas, the lesion areas have different morphologies, and it is difficult to distinguish between solid areas and lesion areas, so an endometrial multi-category attention module is introduced to improve the segmentation accuracy of multi-category tissue pathological images of cervical cancer.

[0075] For details, see Figure 4 As shown, it is a schematic diagram of the structure of a multi-category attention unit in an embodiment of the present invention, and its internal execution process is as follows:

[0076] First, the input image is sent to the multi-channel grayscale feature extraction module, and the input image is divided into feature images of different channels according to the grayscale value. Since the grayscale level of the grayscale image is 8, the number of channels is also 8; and the feature information from different grayscale channels is fused through maximum pooling and average pooling to obtain a multi-channel grayscale attention map containing solid features and cancerous features.

[0077] ;

[0078] Among them, F' represents the multi-channel grayscale attention map obtained by the multi-channel grayscale feature extraction module of the feature map F, Represents the characteristic element multiplication operation, y o Represents the multi-channel grayscale feature extraction operation, the specific formula is:

[0079] ;

[0080] in, represents a sigmoid function, which is used to map the input of a neuron to the output end; S0 and S1 represent weights when the maximum pooling and average pooling are used to fuse images, and the values ​​of the embodiment of the present invention are 0.23 and 0.77 respectively; GS represents an image grayscale feature extraction operation, and MaxPool and AvgPool represent maximum pooling and average pooling operations respectively. and They represent the maximum pooling and average pooling functions respectively, c represents the grayscale feature map extracted when the grayscale value of the input image is c, 0≤c≤255.

[0081] Secondly, in order to more accurately extract the global feature information including the edge information and morphological features of the cancerous area and solid area in the grayscale feature map, the obtained multi-channel grayscale attention map is input into the multi-category expansion and squeeze attention module:

[0082] ;

[0083] Among them, F'' represents the multi-category expansion and squeeze attention map obtained by the multi-channel grayscale attention map through the multi-category expansion and squeeze attention module, y p It represents the multi-category squeezed incentive attention image obtained by the multi-category squeezed incentive attention module. The specific formula is:

[0084] ;

[0085] Among them, F sc represents the expansion operation, T n Indicates the feature weights of different categories of images. In the embodiment of the present invention, they are divided into three categories, namely background feature weight T1, solid feature weight T2 and cancerous feature weight T3, and the values ​​are 0.2, 0.35 and 0.45 respectively; S2 indicates that after the squeezing operation, the useless features in the feature map except the solid area and the cancerous area are suppressed. The specific formula is:

[0086] ;

[0087] Wherein, W1 and W2 represent the weights of the grayscale feature map after passing through the fully connected layer, and in the embodiment of the present invention, the values ​​are 0.3 and 0.7 respectively; Relu represents the Relu activation function, which is used to pass to the convolution solid features and cancerous features; z represents the result obtained after the squeezing operation, and W represents the weight when the image passes through the fully connected layer. The value of the embodiment of the present invention is 0.35. However, the multi-channel grayscale attention module and the multi-category expansion and squeezing attention module mainly extract the global feature information in the image, and the effect of extracting the local feature information of the image is poor. The solid area is included in the cancerous area, and it is necessary to distinguish the local detail differences to achieve high-accuracy multi-category segmentation.

[0088] Therefore, the multi-category expansion and squeeze attention map is input into the multi-category spatial attention module, expressed as:

[0089] ;

[0090] Among them, F''' represents the multi-category spatial attention map obtained by inputting the multi-category expansion and squeeze attention map into the multi-category spatial attention module, and y q Represents a multi-category spatial attention operation, which applies average and maximum pooling along the channel direction of each feature point, and fuses the attention features to generate a multi-category feature map that aggregates features of different channel categories. The specific formula is as follows:

[0091] ;

[0092] Among them, σ represents the sigmoid function, Represents a convolution operation with a filter size of 7×7. and They respectively represent the feature information graphs obtained after average pooling and maximum pooling when the channel is S.

[0093] Finally, through residual operation, the final attention feature map of multi-category histopathological images of cervical cancer is obtained:

[0094] ;

[0095] Among them, Fe represents the multi-category histopathological image of cervical cancer, Represents the residual operation.

[0096] Specifically, the loss function used in the training process in S103 is expressed as:

[0097] ;

[0098] ;

[0099] ;

[0100] Wherein, Loss represents the loss function used by the cervical cancer pathology image segmentation network, n1 and n2 represent loss weights, which are 0.3 and 0.7 respectively in the embodiment of the present invention; L represents the cross entropy loss, x i Represents the i-th element of the true label, y i represents the probability that the model predicts that x belongs to the i-th category; DL represents the Dice loss, represents the number of elements in the intersection of X and Y, and |X| and |Y| represent the number of elements in X and Y respectively.

[0101] Specifically, the S104 uses the improved DeepLabV3+ segmentation model to segment the cancerous part, solid part and background part of the tumor in the cervical cancer pathology image. In this embodiment, the software environment of the segmentation method is constructed using Python3.10.3 programming language and CUDA 10.2 as well as pytorch 1.3.1 and torchvision. The cervical cancer multi-category tissue pathology image segmentation network of the embodiment of the present invention is trained on an input image of size 512x512, with a batch size of 64 and a total number of generations of 200; the initial learning rate is set to 0.05, and the Adam optimizer is used for training. The data set consists of 1,782 images in total, which are divided into training set, test set and validation set in a ratio of 8:1:1, and data enhancement techniques including rotation and flip transformations are used. The final segmentation result of the embodiment of the present invention is shown in the figure. Figure 5 As shown in the figure, the average values ​​of segmentation accuracy, MIoU precision and Dice are 0.934, 0.922 and 0.945 respectively, which have high accuracy and precision.

[0102] Specifically, S105 calculates the ratio of the number of pixels in the cancerous area to the number of pixels in the solid area by counting the number of pixels in the segmented image with grayscale values ​​of 67 and 128 according to the segmentation result obtained in S104, which is expressed as:

[0103] ;

[0104] Among them, A represents the ratio of the number of pixels in the solid area to the cancerous area, S represents the number of pixels in the solid area, and T represents the number of pixels in the cancerous area. The corresponding FIGO grade is obtained by A, which is expressed as:

[0105] .

[0106] For details, see Figure 6 FIG. 1 is a system structure diagram of an embodiment of the present invention, including:

[0107] An acquisition module 601 acquires and annotates multi-category histopathological images of cervical cancer to obtain a pathological image dataset;

[0108] Building module 602, using DeepLabV3+ as the backbone network to build a cervical cancer multi-category tissue pathology image segmentation network, wherein a multi-category attention unit is added to the encoder, and the multi-category attention unit includes a multi-channel grayscale attention module, a multi-category expansion and compression attention module, and a multi-category spatial attention module;

[0109] A training module 603 is used to train a cervical cancer multi-category tissue pathology image segmentation network using a pathology image dataset;

[0110] The segmentation module 604 uses the trained cervical cancer multi-category tissue pathology image segmentation network to segment the cancerous area, solid area and background part of the cervical cancer in the pathology image to obtain a segmentation result;

[0111] The grading module 605 obtains the FIGO grade of the image by using an automatic grading method for the segmentation result.

[0112] The present invention has a higher accuracy and a faster running speed when segmenting different categories of multi-category tissue pathology images of cervical cancer.

[0113] The above are only preferred embodiments of the present invention and are not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principles of the present invention should be included in the protection scope of the present invention.

Claims

1. An auxiliary analysis method for multi-category histopathological images of cervical cancer, characterized in that: The steps include: Collect and annotate multi-category histopathological images of cervical cancer to obtain a pathological image dataset; A multi-category histopathological image segmentation network for cervical cancer is constructed using DeepLabV3+ as the backbone network, wherein a multi-category attention unit is added to the encoder, and the multi-category attention unit includes a multi-channel grayscale attention module, a multi-category expansion and compression attention module, and a multi-category spatial attention module; Use the pathology image dataset to train the cervical cancer multi-category tissue pathology image segmentation network; The trained cervical cancer multi-category tissue pathology image segmentation network is used to segment the cancerous area, solid area and background part of the cervical cancer in the pathology image to obtain the segmentation result; The FIGO grade of the pathological image is obtained by automatic grading method based on the segmentation results.

2. The auxiliary analysis method of multi-category histopathological images of cervical cancer according to claim 1, characterized in that: The cervical cancer multi-category histopathological image segmentation network includes an encoder and a decoder; In the encoder, the input pathological image is extracted and compressed four times using multiple parallel convolutions to obtain preliminary effective features. After inputting into the multi-feature Transformer unit, convolutions of different rates are used to connect and merge, and then 1x1 convolution is performed to compress the features, and finally multi-category fusion features are output; In the decoder, a 1×1 convolution is used to adjust the number of channels of the initial effective feature layer compressed twice, and the original image features are obtained using the initial effective feature layer after the number of channels is adjusted; After using hollow convolution to upsample the multi-scale fusion features of the encoding part output by four times, it is stacked with the original image features, and then processed using two depth-wise separable convolutions and a 1x1 convolution in sequence to obtain the final effective features. Finally, resize is used for upsampling to obtain and output a segmentation result of the same size as the input pathological image.

3. The auxiliary analysis method of multi-category histopathological images of cervical cancer according to claim 1, characterized in that: The internal execution of the multi-class attention unit includes the following steps: First, the input image is sent to the multi-channel grayscale feature extraction module, the input image is divided into feature images of different channels according to the grayscale value, and the feature information from different grayscale channels is fused through maximum pooling and average pooling to obtain a multi-channel grayscale attention map, which is expressed as: ； Among them, F' represents the multi-channel grayscale attention map obtained by the multi-channel grayscale feature extraction module of the feature map F, represents the characteristic element multiplication operation, Represents a multi-channel grayscale feature extraction operation; Secondly, in order to more accurately extract the global feature information including the edge information and morphological features of the cancerous area and the solid area in the grayscale feature map, the obtained multi-channel grayscale attention map is input into the multi-category expansion and squeeze attention module to obtain the multi-category expansion and squeeze attention map, which is expressed as: ； Among them, F'' represents the multi-category expansion and squeeze attention map obtained by the multi-channel grayscale attention map through the multi-category expansion and squeeze attention module, Represents a multi-category expansion-squeezing attention operation; Again, the multi-category expansion and squeeze attention map is input into the multi-category spatial attention module to obtain the multi-category spatial attention map, which is expressed as: ； Among them, F''' represents the multi-category spatial attention map obtained by inputting the multi-category expansion and squeeze attention map into the multi-category spatial attention module, and y q Represents a multi-category spatial attention operation; Finally, the final attention feature map of multi-category histopathological images of cervical cancer is obtained through residual operation, which is expressed as: ； Among them, Fe represents the multi-category histopathological image of cervical cancer, Represents the residual operation.

4. The auxiliary analysis method of multi-category histopathological images of cervical cancer according to claim 3, characterized in that: The multi-channel grayscale feature extraction operation is expressed as: ； in, Represents the sigmoid function, S0 and S1 represent the weights of the maximum pooling and average pooling fusion images, GS represents the image grayscale feature extraction operation, MaxPool and AvgPool represent the maximum pooling and average pooling operations respectively; and They represent the maximum pooling and average pooling functions respectively, c represents the grayscale feature map extracted when the grayscale value of the input image is c, 0≤c≤255.

5. The auxiliary analysis method of multi-category histopathological images of cervical cancer according to claim 3, characterized in that: The multi-category expansion and squeeze attention operation is expressed as: ； Among them, F sc represents the expansion operation, T n represents the feature weights of different categories of the image, T1, T2 and T3 represent the weights of the cancerous region features, solid region features and background respectively, and S2 represents the feature that suppresses the useless features in the feature map except the solid region and cancerous region after the squeezing operation. The specific formula is: ； Among them, W1 and W2 represent the weights of the grayscale feature map after passing through the fully connected layer. Represents the Relu activation function, z represents the result obtained after the squeezing operation, and W represents the weight of the image when it passes through the fully connected layer.

6. The auxiliary analysis method of multi-category histopathological images of cervical cancer according to claim 3, characterized in that: The multi-category spatial attention operation is expressed as: ； Among them, σ represents the sigmoid function, Represents a convolution operation with a filter size of 7×7; and They respectively represent the feature information graphs obtained after average pooling and maximum pooling when the channel is S.

7. The auxiliary analysis method of multi-category histopathological images of cervical cancer according to claim 1, characterized in that: The pathological image dataset is used to train the cervical cancer multi-category tissue pathological image segmentation network, and the loss function used is expressed as: ； ； ； Among them, Loss represents the loss function of the cervical cancer pathology image segmentation network; L represents the cross entropy loss, x i Represents the i-th element of the true label, y i represents the probability that the model predicts that x belongs to the i-th category; DL represents the Dice loss, represents the number of elements that intersect X and Y, |X| and |Y| represent the number of elements in X and Y respectively; n1 and n2 represent the loss weights respectively.

8. The auxiliary analysis method of multi-category histopathological images of cervical cancer according to claim 3, characterized in that: The step of obtaining the FIGO classification of the image by an automatic classification method for the segmentation result comprises the following steps: The number of pixels with grayscale values ​​of 67 and 128 in the image of the segmentation result is counted, and the ratio of the number of pixels in the cancerous area to the number of pixels in the solid area is calculated, which is expressed as: ； Among them, A represents the ratio of the number of pixels in the solid area to the number of pixels in the cancerous area, S represents the number of pixels in the solid area, and T represents the number of pixels in the cancerous area; The corresponding FIGO grade is obtained by the ratio of the number of pixels in the solid area to the cancerous area, expressed as: 。 9. An auxiliary analysis system for multi-category histopathological images of cervical cancer, characterized in that: include: The acquisition module collects and annotates multi-category pathological images of cervical cancer to obtain a pathological image dataset; A construction module is used to construct a cervical cancer multi-category histopathological image segmentation network using DeepLabV3+ as the backbone network, wherein a multi-category attention unit is added to the encoder, and the multi-category attention unit includes a multi-channel grayscale attention module, a multi-category expansion and compression attention module, and a multi-category spatial attention module; The training module uses the pathological image dataset to train the cervical cancer multi-category tissue pathological image segmentation network; The segmentation module uses the trained cervical cancer multi-category tissue pathology image segmentation network to segment the cancerous area, solid area and background part of the cervical cancer in the pathology image to obtain the segmentation result; The grading module obtains the FIGO grade of the pathological image through the automatic grading method for the segmentation results.

Images