Prediction method and system for prognosis of bladder cancer

Abstract

The invention provides a bladder cancer prognosis prediction method and system, and relates to the technical field of intelligent detection, and the method specifically comprises the steps: obtaining and preprocessing a bladder cancer hematoxylin-eosin (HE) staining pathological image to obtain an effective color block, and marking to obtain a tumor and necrosis training set; respectively training the two initial models based on the tumor training set and the necrosis training set to correspondingly obtain a tumor model and a necrosis model; acquiring a to-be-predicted bladder cancer pathological image, and performing tumor and necrosis region identification based on the tumor model and the necrosis model to correspondingly obtain a tumor region and a necrosis region; based on the space coordinate information of the tumor area and the necrotic area, carrying out numerical calculation to obtain a space distribution index of the necrotic area relative to the tumor area; and carrying out survival analysis based on the spatial distribution index to obtain survival prognosis information of the bladder cancer patient. The deep learning technology is applied to the prognosis direction of the bladder cancer, a clinician is assisted in evaluating survival prognosis of a patient, and a reference basis is provided for diagnosis and treatment decisions.

Description

Technical Field

[0001] The present invention relates to the field of intelligent detection technology, and in particular to a method and system for predicting the prognosis of bladder cancer. Background Art

[0002] Bladder cancer is the most common malignant tumor of the urothelial system. According to pathological diagnosis, bladder cancer can be divided into non-muscle-invasive bladder cancer (NMIBC) and muscle-invasive bladder cancer (MIBC). Most bladder cancers have already invaded or metastasized to the muscles at the time of initial diagnosis and have a worse prognosis.

[0003] The tumor node metastasis (TNM) staging system plays a key role in the risk stratification and management of MIBC, but due to the heterogeneity of tumors, the prognosis of patients in the same TNM stage may vary significantly. Studies have shown that tumor necrosis is associated with poor prognosis in a variety of solid tumors, including renal cell carcinoma, hepatocellular carcinoma, pancreatic cancer, breast cancer, colorectal cancer, and osteosarcoma. Therefore, tumor necrosis is expected to become a supplementary prognostic biomarker for MIBC.

[0004] Necrosis is the death of local tissue cells in the living body characterized by enzymatic changes. Necrosis can be caused directly by strong pathogenic factors, but most of them develop from reversible damage. Its basic manifestations are cell swelling, organelle disintegration and protein denaturation. Necrosis often affects multiple cells at the same time.

[0005] It is generally believed that after tumor treatment, tumor cells undergo necrosis or the solid components of the tumor are significantly reduced. During systemic anti-tumor treatment, the tumor undergoes ischemia and necrosis, which are closely related to a better prognosis. If necrosis is found inside the tumor before tumor treatment, it is believed that during the continuous growth of the tumor, due to the lack of blood supply inside the tumor, local ischemia and hypoxia in the center of the tumor will cause tumor cell necrosis. Considering that the tumor is relatively large and the tumor load may be relatively heavy during treatment, the difficulty of treatment will increase.

[0006] Finding out the existence, size, spatial location and other indicators of necrotic areas in tumors before and after cancer treatment can provide a basis for medical staff to make judgments on patient treatment and prognosis, and thus provide patients with more reasonable and accurate treatment plans. It also provides clinicians with more new ideas and data support for anti-cancer treatment.

[0007] At present, the existing technology mainly uses HE staining and microscopic identification to identify and observe necrotic areas within tumors. However, these indicators require professional pathologists to perform histological measurements and analysis on the necrotic areas to obtain them. This process is repetitive, time-consuming, highly professional, and difficult to quantify.

[0008] Therefore, it is urgent to develop a solution to solve the above problems. Summary of the invention

[0009] The purpose of the present invention is to provide a method and system for predicting the prognosis of bladder cancer. By utilizing the advantages of convolutional neural networks, which are efficient, simple, accurate and low-cost, a model for identifying necrotic areas of bladder cancer and calculating spatial distribution indicators is trained to provide a solid scientific foundation and technical support for the development of precision medicine.

[0010] The present invention provides a method and system for predicting the prognosis of bladder cancer, which adopts the following technical solutions: In a first aspect, a method for predicting the prognosis of bladder cancer specifically comprises: Acquire and preprocess bladder cancer stained pathology images to obtain effective color blocks, and annotate and divide the effective color blocks to obtain a tumor training set and a necrosis training set; Construct two initial models, and train the two initial models based on the tumor training set and the necrosis training set, respectively, to obtain a tumor model for identifying the tumor parenchyma area and a necrosis model for identifying the necrosis area; Acquire a bladder cancer pathology image to be predicted, identify the bladder cancer tumor substance and necrotic area of ​​the bladder cancer pathology image based on the tumor model and the necrosis model, respectively, obtain the tumor area and the necrotic area accordingly, and numerically calculate the spatial distribution index of the necrotic area relative to the tumor area based on the spatial coordinate information of the tumor area and the necrotic area; Survival analysis is performed based on the spatial distribution index to obtain the survival prognosis information of bladder cancer patients.

[0011] The beneficial effect of a method for predicting the prognosis of bladder cancer provided by the present invention is that the present invention can automatically identify the tumor substance and necrotic area in bladder cancer WSI (Whole Slide Imaging), and quantify the area ratio of the necrotic area in different areas of the tumor, thereby assisting clinicians in evaluating the patient's survival prognosis and providing a reference basis for digital and individualized diagnosis and treatment decisions.

[0012] Optionally, the effective color blocks are labeled and divided to obtain a tumor training set and a necrosis training set, including: The valid color blocks are marked and divided into tumor color blocks, non-tumor color blocks, necrotic color blocks, and non-necrotic color blocks; Based on the tumor color block and non-tumor color block, the training set and test set of the tumor area and non-tumor area are obtained; Based on the necrotic color blocks and non-necrotic color blocks, the training set and test set of the necrotic area and the non-necrotic area are obtained.

[0013] Optionally, two initial models are constructed, which are constructed based on a residual network architecture and consist of multiple stacked residual blocks. Each residual block contains a convolutional layer, an activation function and a cross-layer connection. The convolutional layer is used to extract image features, the activation function is used to introduce nonlinearity, and the cross-layer connection is used to directly connect the input to the output of the residual block.

[0014] Optionally, two initial models are trained based on the tumor training set and the necrosis training set, respectively, to obtain a tumor model for identifying a tumor parenchymal region and a necrosis model for identifying a necrosis region, respectively, including: Initialize the network weights of the two initial models, input the tumor training set and the necrosis training set respectively, and forward propagate through multiple residual blocks; The loss between the output of the initial model and the true label is calculated based on the loss function, the cross entropy loss function is minimized based on the gradient descent method, and the network weights of the initial model are iterated based on back propagation until the loss tends to be stable or the maximum number of iterations is reached, and the training is terminated; After evaluation and verification, the tumor model and necrosis model were obtained.

[0015] Optionally, a tumor model and a necrosis model are obtained after evaluation and verification, and the evaluation and verification specifically includes: Visualize the weighted regions of tumor and necrosis models based on gradient-weighted class activation mapping; Known bladder cancer pathology image slices are obtained as a transfer learning verification set, and the feasibility of the tumor model and the necrosis model in predicting tissue types is evaluated based on the transfer learning verification set to obtain the tumor model and the necrosis model.

[0016] Optionally, based on the tumor model and the necrosis model, the bladder cancer tumor substance and the necrosis area are identified in the bladder cancer pathology image respectively, and the tumor area and the necrosis area are obtained correspondingly, including: Based on the tumor model, the bladder cancer pathological image is subjected to bladder cancer tumor substance identification to obtain a tumor area; Based on the necrosis model, the bladder cancer pathological image is subjected to bladder cancer tumor necrosis identification to obtain a necrotic area; The edge of the tumor area is analyzed and evaluated based on the spatial distribution of the tumor area, and the tumor tissue is divided into the tumor interior, boundary area and external area.

[0017] Optionally, based on the spatial coordinate information of the tumor region and the necrotic region, a spatial distribution index of the necrotic region relative to the tumor region is numerically calculated, including: Calculate the distribution ratios of tumor parenchyma tissue, necrotic tissue, and other tissues in the tumor area, and calculate the area ratios of tumor tissue and necrotic tissue in the tumor interior, border area, and external area respectively; Based on the above calculation results, a spatial distribution index of the necrotic area relative to the tumor area is constructed; Among them, the spatial distribution indicators include: the total proportion of tumor substance, the proportion of tumor inside the tumor, the proportion of tumor in the boundary area, the proportion of tumor in the external area, the total proportion of necrotic tissue, the proportion of necrotic tissue inside the tumor, the proportion of necrotic tissue in the boundary area and the proportion of necrotic tissue in the external area.

[0018] Optionally, performing survival analysis based on the spatial distribution index to obtain survival prognosis information of bladder cancer patients includes: Based on the Cox proportional hazard regression model, a correlation analysis was performed on the spatial distribution index combined with the survival rate of bladder cancer patients, and a forest plot was constructed with the survival rate as the y-axis and the spatial distribution index as the x-axis to obtain the survival prognosis information of bladder cancer patients.

[0019] In a second aspect, a bladder cancer prognosis prediction system specifically comprises: A data collection module, used for acquiring and preprocessing bladder cancer stained pathological images to obtain effective color blocks, and annotating and dividing the effective color blocks to obtain a tumor training set and a necrosis training set; A model training module is used to construct two initial models. The two initial models are trained based on a tumor training set and a necrosis training set, respectively, to obtain a tumor model for identifying a tumor parenchymal area and a necrosis model for identifying a necrosis area; An index calculation module is used to obtain a bladder cancer pathology image to be predicted, identify the bladder cancer tumor substance and necrotic area of ​​the bladder cancer pathology image based on a tumor model and a necrosis model, obtain a tumor area and a necrotic area respectively, and numerically calculate the spatial distribution index of the necrotic area relative to the tumor area based on the spatial coordinate information of the tumor area and the necrotic area; The survival analysis module performs survival analysis based on the spatial distribution index to obtain the survival prognosis information of bladder cancer patients.

[0020] Optionally, in the model training module, the network structures of the two initial models constructed include: an input layer, a convolutional layer, an activation layer, a pooling layer and a fully connected layer; The input layer is used to receive bladder cancer pathology image data and transmit it to the convolution layer; The convolution layer consists of multiple convolution kernels. Each convolution kernel is responsible for extracting local features and extracting the features of bladder cancer pathology image data. The convolution kernel is slid on the input data, and the dot product between the convolution kernel and the local area of ​​the input data is calculated to generate a feature map. The activation layer is located immediately after the convolutional layer and is used to introduce nonlinear calculations to enhance the expressiveness of the initial model. It uses the ReLU activation function. The pooling layer is used to reduce the spatial dimension of the feature map, reduce the number of parameters, and improve the generalization ability of the initial model. It takes the maximum value in the local area, retains the most important features, or calculates the average value in the local area to make the features smoother. Fully connected layer: the neurons in the fully connected layer are connected to all neurons in the previous layer, linearly combined through weights and biases, and multi-category classification is performed using the softmax function to output the classification results; Among them, the convolutional layers are connected by jump connections, and the input is directly added to the output to form a residual block.

[0021] The beneficial effects of the second aspect can refer to the description of the first aspect. BRIEF DESCRIPTION OF THE DRAWINGS

[0022] Figure 1 A flowchart of the overall steps of a method for predicting the prognosis of bladder cancer provided by the present invention; Figure 2 A schematic diagram of the marked area of ​​a method for predicting the prognosis of bladder cancer provided by the present invention; Figure 3 A schematic diagram of four types of color blocks GradCAM of a method for predicting the prognosis of bladder cancer provided by the present invention; Figure 4 A schematic diagram of the spatial distribution of tumor and necrotic areas in a method for predicting the prognosis of bladder cancer provided by the present invention. DETAILED DESCRIPTION

[0023] In order to make the purpose, technical solutions and advantages of the embodiments of the present invention clearer, the technical solutions in the embodiments of the present invention will be clearly and completely described below. Obviously, the described embodiments are part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention. Unless otherwise defined, the technical terms or scientific terms used herein should be understood by people with general skills in the field to which the present invention belongs. "Including" and similar words used in this article mean that the elements or objects appearing before the word include the elements or objects listed after the word and their equivalents, without excluding other elements or objects.

[0024] The present invention provides a method for predicting the prognosis of bladder cancer. Figure 1 , specifically including the following steps: S1. Acquire and preprocess bladder cancer stained pathological images to obtain effective color blocks, and annotate and divide the effective color blocks to obtain a tumor training set and a necrosis training set; S2. Construct two initial models, and train the two initial models based on the tumor training set and the necrosis training set, respectively, to obtain a tumor model for identifying the tumor parenchyma area and a necrosis model for identifying the necrosis area; S3, obtaining a bladder cancer pathology image to be predicted, identifying the bladder cancer tumor substance and the necrotic area of ​​the bladder cancer pathology image based on the tumor model and the necrosis model, respectively, obtaining the tumor area and the necrotic area accordingly, and numerically calculating the spatial distribution index of the necrotic area relative to the tumor area based on the spatial coordinate information of the tumor area and the necrotic area; S4. Performing survival analysis based on the spatial distribution index to obtain survival prognosis information of bladder cancer patients.

[0025] In some embodiments, when executing step S1, it includes: S1-1, obtaining and preprocessing bladder cancer staining pathology images to obtain effective color blocks; S1-2. Annotate and divide the effective color blocks to obtain a tumor training set and a necrosis training set.

[0026] Specifically, when executing step S1-1, the criteria for obtaining the bladder cancer staining pathology image include: Inclusion criteria: pathologically confirmed MIBC (muscle-invasive bladder cancer), complete clinical data, and pathology experts confirming that the pathological images have clear and uniformly stained tissue structures; Exclusion criteria: pathological images with a magnification of less than 40×, damage under naked eye observation, low definition, or large-area contamination.

[0027] Furthermore, preprocessing includes image segmentation and quality control. Image segmentation is to segment the manually annotated digital pathology image into color blocks of 512×512 pixels. Quality control includes: eliminating color blocks with inconsistent size, low clarity, blank background, and zero memory.

[0028] Specifically, when executing step S1-2, the effective color blocks are labeled and divided to obtain a tumor training set and a necrosis training set, including: The valid color blocks are marked and divided into tumor color blocks, non-tumor color blocks, necrotic color blocks, and non-necrotic color blocks; Based on the tumor color block and the non-tumor color block, the training set and the test set of the tumor area and the non-tumor area are obtained. Based on the necrotic color block and the non-necrotic color block, the training set and the test set of the necrotic area and the non-necrotic area are obtained.

[0029] In fact, the effective color blocks obtained from the bladder cancer stained pathology image are handed over to the pathologist for annotation, and the obtained blocks are tumor color blocks, non-tumor color blocks, necrotic color blocks, and non-necrotic color blocks.

[0030] Furthermore, the color blocks after quality control are divided into training sets and test sets for tumor areas and non-tumor areas, and training sets and test sets for necrotic areas and non-necrotic areas.

[0031] In fact, see Figure 2 , where A is the bladder cancer stained pathological image and the pathological image after the tumor area and necrotic area were annotated by the pathologist; where B is the ROC curve of the ResNet model for predicting tumor color blocks (AUC = 0.9439, 95% CI: 0.942-0.945); where C is the ROC curve of the ResNet model for predicting necrotic color blocks (AUC = 0.9357, 95% CI: 0.855-0.860).

[0032] In some embodiments, when executing step S2, it specifically includes: S2-1, construct two initial models; S2-2, training two initial models based on the tumor training set and the necrosis training set, respectively, to obtain a tumor model for identifying the tumor parenchyma area and a necrosis model for identifying the necrosis area; Specifically, when executing step S2-1, two initial models are constructed. The initial models are constructed based on the residual network architecture and are composed of multiple residual blocks stacked together. Each residual block contains a convolutional layer, an activation function and a cross-layer connection. The convolutional layer is used to extract image features, the activation function is used to introduce nonlinearity, and the cross-layer connection is used to directly connect the input to the output of the residual block.

[0033] In fact, the data transmission process in the initial model includes: the size of the input image is 512 pixels × 512 pixels to the first convolution layer (Conv2D, which contains 64 7x7 filters with a stride of 2), connected to a batch normalization layer for normalizing the output of the convolution layer, and then using the ReLU activation function, the output image shape is 512 pixels * 512 pixels * 32 channels, and then connected to a 3x3 maximum pooling layer (MaxPooling2D) with a stride of 2, and the output shape is 256 pixels * 256 pixels * 32 channels, and then transmitted to the second convolution layer (implementing a convolution operation with a residual connection, which contains two 3x3 convolution kernels and one 1x1 convolution kernel), the down-sampled output shape is 256 pixels * 256 pixels * 32 channels, followed by another maximum pooling layer with a stride of 2, and the output shape is 128 pixels * 128 pixels * 64 channels.

[0034] Furthermore, the identity block (identity_block) implements the identity mapping in the residual block by adding the input directly to the convolutional and normalized output, which helps to solve the gradient vanishing problem in deep networks. Each convolutional layer in the model is followed by several identity blocks, and the number of filters in the identity blocks gradually increases, namely 64, 128, 256 and 512, respectively. Convolutional layers with a step size of 2 are used to reduce the spatial dimension of the feature map. After all convolutional and residual blocks, a global average pooling layer (GlobalAveragePooling2D) is used to convert the feature map into a one-dimensional feature vector. The last fully connected layer (Dense) uses the ReLU activation function and adds a Dropout layer to reduce the risk of overfitting. 50% of the input neurons are randomly discarded, the feature vector is mapped to the number of categories, and the softmax activation function is used for multi-category classification.

[0035] Specifically, when executing step S2-2, two initial models are trained based on the tumor training set and the necrosis training set, respectively, to obtain a tumor model for identifying a tumor parenchymal area and a necrosis model for identifying a necrosis area, including: Initialize the network weights of the two initial models, input the tumor training set and the necrosis training set respectively, and forward propagate through multiple residual blocks; The loss between the output of the initial model and the true label is calculated based on the loss function, the cross entropy loss function is minimized based on the gradient descent method, and the network weights of the initial model are iterated based on back propagation until the loss tends to be stable or the maximum number of iterations is reached, and the training is terminated; After evaluation and verification, the tumor model and necrosis model were obtained.

[0036] In fact, the tumor model and the necrosis model are obtained after evaluation and verification, and the evaluation and verification specifically include: Visualize the weighted regions of tumor and necrosis models based on gradient-weighted class activation mapping; Known bladder cancer pathology image slices are obtained as a transfer learning verification set, and the feasibility of the tumor model and the necrosis model in predicting tissue types is evaluated based on the transfer learning verification set to obtain the tumor model and the necrosis model.

[0037] Furthermore, the test data was input into the tumor model and the necrosis model, and Grad-CAM was used to obtain the model focus point heat map, where the remaining bladder cancer pathology images were used as the transfer learning verification set. The test data came from the test data obtained by randomly dividing the tumor training set and the necrosis training set in a ratio of 2:8.

[0038] For further information, see Figure 3 The figure shows the Grad-CAM visualization of four types of color blocks, where A is a tumor color block, highlighting the characteristic area determined by the ResNet model as the tumor cell nucleus; B is a necrotic color block, and the highlighted area in the heat map represents the necrotic cells; C is a color block where tumor cells and necrotic cells coexist, proving the accuracy of the ResNet model's recognition; D is a color block without tumor cells or necrotic cells, defined as other color blocks.

[0039] In some embodiments, when executing step S3, it includes: S3-1, obtaining a predicted bladder cancer pathology image, and identifying the bladder cancer tumor substance and necrosis area of ​​the bladder cancer pathology image based on a tumor model and a necrosis model, respectively, to obtain a tumor area and a necrosis area accordingly; S3-2. Based on the spatial coordinate information of the tumor region and the necrotic region, a spatial distribution index of the necrotic region relative to the tumor region is numerically calculated.

[0040] Specifically, when executing step S3-1, the bladder cancer tumor substance and necrotic area are identified on the bladder cancer pathological image based on the tumor model and the necrosis model, and the tumor area and the necrotic area are obtained correspondingly, including: Based on the tumor model, the bladder cancer pathological image is subjected to bladder cancer tumor substance identification to obtain a tumor area; Based on the necrosis model, the bladder cancer pathological image is subjected to bladder cancer tumor necrosis identification to obtain a necrotic area; The edge of the tumor area is analyzed and evaluated based on the spatial distribution of the tumor area, and the tumor tissue is divided into the tumor interior, boundary area and external area.

[0041] In fact, the SPIAT software package was used to analyze the spatial distribution of the tumor region and to evaluate the margins of the tumor region to divide each region.

[0042] Specifically, when executing step S3-2, based on the spatial coordinate information of the tumor region and the necrotic region, a spatial distribution index of the necrotic region relative to the tumor region is numerically calculated, including: Calculate the distribution ratios of tumor parenchyma tissue, necrotic tissue, and other tissues in the tumor area, and calculate the area ratios of tumor tissue and necrotic tissue in the tumor interior, border area, and external area respectively; Based on the above calculation results, the spatial distribution index of the necrotic area relative to the tumor area is constructed.

[0043] Furthermore, the spatial distribution indicators include: the total proportion of tumor substance, the proportion of tumor inside the tumor, the proportion of tumor in the boundary area, the proportion of tumor in the external area, the total proportion of necrotic tissue, the proportion of necrotic tissue inside the tumor, the proportion of necrotic tissue in the boundary area and the proportion of necrotic tissue in the external area (tumor%, tumor_inside%, tumor_border%, tumor_outside%, necrosis%, necrosis_inside%, necrosis_border% and necrosis_outside%).

[0044] For further information, see Figure 4 , are the spatial distributions of tumor and necrotic areas in the two ROIs, where C means that in H&E-1, the proportion of necrotic tissue inside the tumor parenchyma is higher than that in the external area, and in H&E-2, the proportion of necrotic tissue in the external area is higher than that in the tumor; A and B are the proportions of tumor blocks, necrotic blocks and other blocks in different areas of the tumor parenchyma in H&E-1 and H&E-2.

[0045] In some embodiments, when executing step S4, performing survival analysis based on the spatial distribution index to obtain survival prognosis information of bladder cancer patients includes: Based on the Cox proportional hazard regression model, a correlation analysis was performed on the spatial distribution index combined with the survival rate of bladder cancer patients, and a forest plot was constructed with the survival rate as the y-axis and the spatial distribution index as the x-axis to obtain the survival prognosis information of bladder cancer patients.

[0046] The embodiment of the present invention further provides a prediction system for bladder cancer prognosis, which specifically includes: A data collection module, used for acquiring and preprocessing bladder cancer stained pathological images to obtain effective color blocks, and annotating and dividing the effective color blocks to obtain a tumor training set and a necrosis training set; A model training module is used to construct two initial models. The two initial models are trained based on a tumor training set and a necrosis training set, respectively, to obtain a tumor model for identifying a tumor parenchymal area and a necrosis model for identifying a necrosis area; An index calculation module is used to obtain a bladder cancer pathology image to be predicted, identify the bladder cancer tumor substance and necrotic area of ​​the bladder cancer pathology image based on a tumor model and a necrosis model, obtain a tumor area and a necrotic area respectively, and numerically calculate the spatial distribution index of the necrotic area relative to the tumor area based on the spatial coordinate information of the tumor area and the necrotic area; The survival analysis module performs survival analysis based on the spatial distribution index to obtain the survival prognosis information of bladder cancer patients.

[0047] In some embodiments, in the model training module, the network structures of the two initial models constructed include: an input layer, a convolutional layer, an activation layer, a pooling layer, and a fully connected layer; The input layer is used to receive bladder cancer pathology image data and transmit it to the convolution layer; The convolution layer consists of multiple convolution kernels. Each convolution kernel is responsible for extracting local features and extracting the features of bladder cancer pathology image data. The convolution kernel is slid on the input data, and the dot product between the convolution kernel and the local area of ​​the input data is calculated to generate a feature map. The activation layer is located immediately after the convolutional layer and is used to introduce nonlinear calculations to enhance the expressiveness of the initial model. It uses the ReLU activation function. The pooling layer is used to reduce the spatial dimension of the feature map, reduce the number of parameters, and improve the generalization ability of the initial model. It takes the maximum value in the local area, retains the most important features, or calculates the average value in the local area to make the features smoother. Fully connected layer: The neurons in the fully connected layer are connected to all the neurons in the previous layer, linearly combined through weights and biases, and multi-category classification is performed using the softmax function to output the classification results.

[0048] Furthermore, the convolutional layers are connected through skip connections, and the input is directly added to the output to form a residual block.

[0049] Although the embodiments of the present invention are described in detail above, it is obvious to those skilled in the art that various modifications and variations can be made to these embodiments. However, it should be understood that such modifications and variations are within the scope and spirit of the present invention as described in the claims. Moreover, the present invention described herein may have other embodiments and may be implemented or realized in a variety of ways.

Claims

1. A method for predicting the prognosis of bladder cancer, It is characterized in that include: Acquire and preprocess bladder cancer stained pathology images to obtain effective color blocks, and annotate and divide the effective color blocks to obtain a tumor training set and a necrosis training set; Construct two initial models, and train the two initial models based on the tumor training set and the necrosis training set, respectively, to obtain a tumor model for identifying the tumor parenchyma area and a necrosis model for identifying the necrosis area; Acquire a bladder cancer pathology image to be predicted, identify the bladder cancer tumor substance and necrotic area of ​​the bladder cancer pathology image based on the tumor model and the necrosis model, respectively, obtain the tumor area and the necrotic area accordingly, and numerically calculate the spatial distribution index of the necrotic area relative to the tumor area based on the spatial coordinate information of the tumor area and the necrotic area; Survival analysis is performed based on the spatial distribution index to obtain the survival prognosis information of bladder cancer patients.

2. A method for predicting bladder cancer prognosis according to claim 1, It is characterized in that Based on the effective color blocks, the effective color blocks are labeled and divided to obtain a tumor training set and a necrosis training set, including: The valid color blocks are marked and divided into tumor color blocks, non-tumor color blocks, necrotic color blocks, and non-necrotic color blocks; Based on the tumor color block and non-tumor color block, the training set and test set of the tumor area and non-tumor area are obtained; Based on the necrotic color blocks and non-necrotic color blocks, the training set and test set of the necrotic area and the non-necrotic area are obtained.

3. The method for predicting the prognosis of bladder cancer according to claim 1, It is characterized in that Two initial models are constructed. The initial models are constructed based on the residual network architecture and consist of multiple stacked residual blocks. Each residual block contains a convolutional layer, an activation function and a cross-layer connection. The convolutional layer is used to extract image features, the activation function is used to introduce nonlinearity, and the cross-layer connection is used to directly connect the input to the output of the residual block.

4. A method for predicting the prognosis of bladder cancer according to claim 3, It is characterized in that Two initial models are trained based on the tumor training set and the necrosis training set, respectively, to obtain a tumor model for identifying the tumor parenchyma area and a necrosis model for identifying the necrosis area, including: Initialize the network weights of the two initial models, input the tumor training set and the necrosis training set respectively, and forward propagate through multiple residual blocks; The loss between the output of the initial model and the true label is calculated based on the loss function, the cross entropy loss function is minimized based on the gradient descent method, and the network weights of the initial model are iterated based on back propagation until the loss tends to be stable or the maximum number of iterations is reached, and the training is terminated; After evaluation and verification, the tumor model and necrosis model were obtained.

5. A method for predicting the prognosis of bladder cancer according to claim 4, It is characterized in that After evaluation and verification, the tumor model and necrosis model are obtained, and the evaluation and verification specifically include: Visualize the weighted regions of tumor and necrosis models based on gradient-weighted class activation mapping; Known bladder cancer pathology image slices are obtained as a transfer learning verification set, and the feasibility of the tumor model and the necrosis model in predicting tissue types is evaluated based on the transfer learning verification set to obtain the tumor model and the necrosis model.

6. A method for predicting the prognosis of bladder cancer according to claim 1, It is characterized in that Based on the tumor model and the necrosis model, the bladder cancer tumor substance and the necrosis area are identified in the bladder cancer pathology image respectively, and the tumor area and the necrosis area are obtained correspondingly, including: Based on the tumor model, the bladder cancer pathological image is subjected to bladder cancer tumor substance identification to obtain a tumor area; Based on the necrosis model, the bladder cancer pathological image is subjected to bladder cancer tumor necrosis identification to obtain a necrotic area; The edge of the tumor area is analyzed and evaluated based on the spatial distribution of the tumor area, and the tumor tissue is divided into the tumor interior, boundary area and external area.

7. A method for predicting the prognosis of bladder cancer according to claim 6, It is characterized in that Based on the spatial coordinate information of the tumor region and the necrotic region, the spatial distribution index of the necrotic region relative to the tumor region is obtained by numerical calculation, including: Calculate the distribution ratios of tumor parenchyma tissue, necrotic tissue, and other tissues in the tumor area, and calculate the area ratios of tumor tissue and necrotic tissue in the tumor interior, border area, and external area respectively; Based on the above calculation results, a spatial distribution index of the necrotic area relative to the tumor area is constructed; Among them, the spatial distribution indicators include: the total proportion of tumor substance, the proportion of tumor inside the tumor, the proportion of tumor in the boundary area, the proportion of tumor in the external area, the total proportion of necrotic tissue, the proportion of necrotic tissue inside the tumor, the proportion of necrotic tissue in the boundary area and the proportion of necrotic tissue in the external area.

8. A method for predicting the prognosis of bladder cancer according to claim 7, It is characterized in that Survival analysis based on the spatial distribution index is performed to obtain survival prognosis information of bladder cancer patients, including: Based on the Cox proportional hazard regression model, a correlation analysis was performed on the spatial distribution index combined with the survival rate of bladder cancer patients, and a forest plot was constructed with the survival rate as the y-axis and the spatial distribution index as the x-axis to obtain the survival prognosis information of bladder cancer patients.

9. A prediction system for bladder cancer prognosis. It is characterized in that Specifically include: A data collection module, used for acquiring and preprocessing bladder cancer stained pathological images to obtain effective color blocks, and annotating and dividing the effective color blocks to obtain a tumor training set and a necrosis training set; A model training module is used to construct two initial models. The two initial models are trained based on a tumor training set and a necrosis training set, respectively, to obtain a tumor model for identifying a tumor parenchymal area and a necrosis model for identifying a necrosis area; An index calculation module is used to obtain a bladder cancer pathology image to be predicted, identify the bladder cancer tumor substance and necrotic area of ​​the bladder cancer pathology image based on a tumor model and a necrosis model, obtain a tumor area and a necrotic area respectively, and numerically calculate the spatial distribution index of the necrotic area relative to the tumor area based on the spatial coordinate information of the tumor area and the necrotic area; The survival analysis module performs survival analysis based on the spatial distribution index to obtain the survival prognosis information of bladder cancer patients.

10. A bladder cancer prognosis prediction system according to claim 9, It is characterized in that In the model training module, the network structures of the two initial models constructed include: input layer, convolution layer, activation layer, pooling layer and fully connected layer; The input layer is used to receive bladder cancer pathology image data and transmit it to the convolution layer; The convolution layer consists of multiple convolution kernels. Each convolution kernel is responsible for extracting local features and extracting the features of bladder cancer pathology image data. The convolution kernel is slid on the input data, and the dot product between the convolution kernel and the local area of ​​the input data is calculated to generate a feature map. The activation layer is located immediately after the convolutional layer and is used to introduce nonlinear calculations to enhance the expressiveness of the initial model. It uses the ReLU activation function. The pooling layer is used to reduce the spatial dimension of the feature map, reduce the number of parameters, and improve the generalization ability of the initial model. It takes the maximum value in the local area, retains the most important features, or calculates the average value in the local area to make the features smoother. Fully connected layer: the neurons in the fully connected layer are connected to all neurons in the previous layer, linearly combined through weights and biases, and multi-category classification is performed using the softmax function to output the classification results; Among them, the convolutional layers are connected by jump connections, and the input is directly added to the output to form a residual block.

Images