A method and system for bladder cancer immunophenotyping

Abstract

The application provides a bladder cancer immunophenotyping recognition method and system, comprising: preprocessing pathological image data to obtain preprocessed pathological image data; constructing a tumor and stroma region segmentation model based on a semantic segmentation network TransUNet, training the tumor and stroma region segmentation model using the preprocessed pathological image data and label data, and obtaining a trained tumor and stroma region segmentation model; constructing a TIL recognition neural network model classification model based on Resnet50 using a tumor and stroma segmentation result output by the trained tumor and stroma region segmentation model; and obtaining a bladder cancer immunophenotyping result based on a lymphocyte infiltration score output by the TIL recognition neural network model classification model. The application automatically recognizes a tumor region and immune cell distribution using digital pathological images, constructs a standardized immunoscore and typing method, and improves the individual decision-making ability of clinical immunotherapy.

Description

Technical Field

[0001] This invention relates to the field of medical image processing technology, and in particular to a method and system for identifying bladder cancer immunophenotyping, used for assisting diagnosis, predicting immunotherapy response, and supporting individualized treatment decisions. Background Technology

[0002] Bladder cancer is the ninth most common malignant tumor worldwide, accounting for approximately 4.2% of all new cancer cases globally, and is one of the most common malignant tumors of the urinary system. Based on clinicopathological classification, bladder cancer is mainly divided into non-muscle-invasive bladder cancer (NMIBC) and muscle-invasive bladder cancer (MIBC). The five-year survival rate for NMIBC patients can reach 96%, while the five-year survival rate for advanced MIBC is only about 8%, indicating a very poor prognosis. Evidence-based guidelines recommend transurethral resection of bladder tumor (TURBT) combined with intravesical instillation therapy for NMIBC, while radical cystectomy combined with pelvic lymph node dissection is the standard treatment for high-risk NMIBC or MIBC patients. However, even with radical treatment, approximately 50% of MIBC patients experience recurrence or distant metastasis, and 10%-15% of patients have metastases at initial diagnosis, resulting in a median survival of only about 14 months.

[0003] In recent years, with a deeper understanding of tumor biology, tumors have been recognized as complex ecosystems. Immune cells, matrix components, and their interactions within the tumor microenvironment (TME) play a crucial role in tumor development, progression, and treatment response. Immune checkpoint blockade (ICB), particularly drugs targeting the PD-1 / PD-L1 pathway, has shown some efficacy in advanced bladder cancer, with some patients achieving long-term survival. However, bladder cancer is highly immunogenic, exhibiting a high tumor mutational burden and neoantigen expression; approximately 70% of patients still exhibit primary or secondary resistance to ICB treatment.

[0004] Tumor-infiltrating lymphocytes (TILs), as a key immune component of the tumor microenvironment (TME), are closely related to immunotherapy response and patient prognosis in terms of their quantity and distribution. Existing literature indicates that increased TIL numbers are generally associated with good treatment response and longer survival, and their predictive ability in various cancers using immune scores is superior to traditional clinicopathological indicators such as TNM staging. In recent years, the spatial distribution characteristics of immune phenotypes have attracted widespread attention. The three types of immune phenotypes—"immune activation type," "immune rejection type," and "immune desert type"—provide a theoretical basis for precision immunotherapy.

[0005] However, current clinical assessment of tumor-associated lymphoid tissue (TILs) in bladder cancer largely relies on the subjective interpretation of slides by pathologists, which suffers from limitations such as poor standardization and low reproducibility, hindering high-throughput clinical application. With the rapid development of artificial intelligence and computational pathology, digital pathological image analysis technology has been used for quantitative research on tumor tissue endothelial growth factor (TME). Deep learning models can accurately identify and segment tumor regions, stromal regions, and TILs within tumor tissue, assisting in the construction of standardized immune scoring systems and overcoming the bottlenecks of manual assessment. Summary of the Invention

[0006] This invention provides a method and system for identifying immune typing in bladder cancer, addressing the deficiencies in existing technologies. By using deep learning technology, it automatically identifies and quantifies the spatial distribution characteristics of tumor-infiltrating lymphocytes (TILs) in tumor regions and stromal regions of tumor tissue, enabling precise assessment of the immune microenvironment in bladder cancer patients. This leads to the construction of a standardized immune typing system, which in turn assists in predicting immunotherapy responses and clinical prognoses, providing a scientific basis and intelligent decision support for developing individualized treatment plans for bladder cancer patients.

[0007] In a first aspect, the present invention provides a method for identifying bladder cancer through immunophenotyping, comprising: Collect pathological image data from patients; The pathological image data is preprocessed to obtain preprocessed pathological image data; A tumor and stroma region segmentation model was constructed based on the semantic segmentation network TransUNet. The preprocessed pathological image data and label data were used to train the tumor and stroma region segmentation model to obtain a trained tumor and stroma region segmentation model. Using the tumor and stroma segmentation results output by the trained tumor and stroma region segmentation model, a TIL recognition neural network model based on ResNet50 is constructed for classification. Based on the lymphocyte infiltration score output by the TIL recognition neural network model classification model, the bladder cancer immunophenotyping results are obtained.

[0008] According to the present invention, a method for identifying bladder cancer immunophenotyping involves collecting pathological image data of patients, including: Obtain surgically removed tissue from the patient; HE-stained digital pathological whole-section images were prepared from the surgically removed tissue; A digital slide scanner was used to scan the preprocessed HE-stained digital pathological whole slide images at a preset resolution to generate full-field images.

[0009] According to a bladder cancer immunophenotyping identification method provided by the present invention, the pathological image data is preprocessed to obtain preprocessed pathological image data, including: The full-view image is sequentially processed by color normalization, noise reduction, and patch segmentation to output the preprocessed pathological image data.

[0010] According to the present invention, a method for identifying bladder cancer immunophenotyping involves constructing a tumor-stromal region segmentation model based on the semantic segmentation network TransUNet, and training the tumor-stromal region segmentation model using the preprocessed pathological image data and label data to obtain a trained tumor-stromal region segmentation model, including: The tumor region and stroma region in each image were manually delineated using biological image analysis software. If any discrepancies in the annotation were found, they were then subject to final review and confirmation by experts. The tumor region and stromal region are segmented into patches of a preset size to generate the original pathological image and a mask with labeled regions; The original pathological images are divided into training set and training set, and the tumor and stroma region segmentation model is trained. The Dice loss function is used to optimize the model to obtain the trained tumor and stroma region segmentation model.

[0011] According to the present invention, a method for identifying bladder cancer immunophenotyping utilizes the tumor and stroma segmentation results output by the trained tumor and stroma region segmentation model to construct a ResNet50-based TIL identification neural network classification model, which outputs a lymphocyte infiltration score, including: Obtain the TIL database, which includes a dataset consisting of several pixel patch images of a preset size. Enlarge the dataset by a preset factor and label each patch as TIL positive or TIL negative. The dataset was divided into training and validation sets. A ResNet-50 feature extractor pre-trained based on ImageNet-1K was used, the output categories were set, and a patch classification model was trained based on the cross-entropy loss function to obtain the TIL recognition neural network classification model.

[0012] According to the present invention, a method for identifying bladder cancer immunophenotyping, based on the lymphocyte infiltration score output by the TIL identification neural network model classification model, obtains bladder cancer immunophenotyping results, including: The number of TIL-positive patches in the tumor epithelium and the number of patches in the tumor epithelial region are used to obtain the lymphocyte infiltration score of the tumor region; The peritumoral lymphocyte infiltration score is obtained by the number of TIL-positive patches within a predetermined range of the tumor tissue edge and the number of patches within the predetermined range of the tumor tissue edge. The tumor region lymphocyte infiltration score is used to differentiate between immune-activated and immune-cold tumors; The immune rejection type and immune desert type of the immune cold tumor are distinguished based on the peritumoral lymphocyte infiltration score.

[0013] The bladder cancer immunophenotyping identification method provided by the present invention further includes: The accuracy of tumor region identification, the accuracy of TIL identification and classification, and the retrospective cohort study of patients were validated respectively.

[0014] Secondly, the present invention also provides a bladder cancer immune typing identification system, comprising: The acquisition module is used to acquire pathological image data of patients; The preprocessing module is used to preprocess the pathological image data to obtain preprocessed pathological image data. The module is used to construct a tumor and stroma region segmentation model based on the semantic segmentation network TransUNet. The preprocessed pathological image data and label data are used to train the tumor and stroma region segmentation model to obtain a trained tumor and stroma region segmentation model. The classification module is used to construct a TIL recognition neural network model based on ResNet50 using the tumor and stroma segmentation results output by the trained tumor and stroma region segmentation model. The identification module is used to obtain the bladder cancer immunophenotyping result based on the lymphocyte infiltration score output by the TIL identification neural network model classification model.

[0015] Thirdly, the present invention also provides an electronic device, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the program to implement the bladder cancer immune typing identification method as described above.

[0016] Fourthly, the present invention also provides a non-transitory computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the bladder cancer immune typing identification method as described above.

[0017] The bladder cancer immunophenotyping identification method and system provided by this invention automatically identifies and quantifies the types and spatial distribution characteristics of tumor regions, stromal regions, and tumor-infiltrating lymphocytes (TILs) in tumor tissue through deep learning technology. By calculating multiple TIL scoring indicators, it achieves a quantitative assessment of the degree of immune infiltration, significantly improving the accuracy and standardization of diagnosis and avoiding the problems of poor standardization and low repeatability caused by the subjective interpretation of slides by traditional pathologists. This system supports high-throughput clinical applications, which helps to promote the precise immunophenotyping of bladder cancer on a wider scale and improve the efficiency of clinical diagnosis and treatment. At the same time, it constructs an objective and standardized immunophenotyping system, providing quantitative evidence for the precise immunotherapy of bladder cancer, which helps to develop individualized treatment plans for patients, promotes the development of precision medicine for bladder cancer, and is expected to improve the treatment effect and quality of life of bladder cancer patients, reduce medical costs, and has significant social and economic benefits. Furthermore, this system can also reduce the workload of pathologists, improve the overall level of medical services, and adopts a modular design, which is easy to integrate into existing pathological diagnostic systems, facilitating subsequent functional expansion and upgrades to meet the ever-evolving clinical needs. Attached Figure Description

[0018] To more clearly illustrate the technical solutions in this invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of this invention. For those skilled in the art, other drawings can be obtained from these drawings without creative effort.

[0019] Figure 1 This is a flowchart illustrating the bladder cancer immunophenotyping identification method provided by the present invention; Figure 2 This is the overall technical roadmap provided by the present invention; Figure 3 This is a diagram of the TransUnet neural network architecture provided by this invention; Figure 4 These are images showing the results of tumor and stroma segmentation at the Patch level and whole slice level provided by this invention. Figure 5 This is the identification diagram of positive and negative patches (TIL-) of lymphocyte infiltration provided by the present invention; Figure 6 This is a Kaplan-Meier survival analysis diagram of three immunophenotyping types in a cohort study provided by the present invention; Figure 7 This is a schematic diagram of the Dice Score for tumor region segmentation provided by the present invention; Figure 8 This is a diagram illustrating the training process of the TIL recognition model provided by this invention; Figure 9 This is a schematic diagram of the bladder cancer immune typing and identification system provided by the present invention; Figure 10 This is a schematic diagram of the structure of the electronic device provided by the present invention. Detailed Implementation

[0020] To make the objectives, technical solutions, and advantages of this invention clearer, the technical solutions of this invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some, not all, of the embodiments of this invention. All other embodiments obtained by those skilled in the art based on the embodiments of this invention without creative effort are within the scope of protection of this invention.

[0021] Figure 1 This is a flowchart illustrating the bladder cancer immunophenotyping identification method provided in an embodiment of the present invention, as shown below. Figure 1 As shown, it includes: Step 100: Collect the patient's pathological image data; Step 200: Preprocess the pathological image data to obtain preprocessed pathological image data; Step 300: Construct a tumor and stroma region segmentation model based on the semantic segmentation network TransUNet, and train the tumor and stroma region segmentation model using the preprocessed pathological image data and label data to obtain a trained tumor and stroma region segmentation model. Step 400: Using the tumor and stroma segmentation results output by the trained tumor and stroma region segmentation model, construct a TIL recognition neural network model based on ResNet50 for classification. Step 500: Based on the lymphocyte infiltration score output by the TIL recognition neural network model classification model, obtain the bladder cancer immunophenotyping result.

[0022] Specifically, such as Figure 2The technical roadmap shown in this invention first acquires H&E-stained digital whole-slide images (WSI) of bladder cancer patients and performs image preprocessing, including color normalization, noise reduction, and patch segmentation, to improve image quality and model recognition efficiency. A deep learning-based semantic segmentation network (TransUNet) is used to automatically segment the tumor and stromal regions in the WSI, forming a structured spatial region mask to provide precise region localization for subsequent analysis. Based on the segmentation results, the tumor epithelial and peritumoral tissue regions are located. A ResNet50-based TIL recognition neural network model is constructed to achieve automatic TIL subtype identification and spatial localization. A tumor region lymphocyte infiltration score (iTIL) and a peritumoral lymphocyte infiltration score (sTIL) are constructed to achieve quantitative assessment of the degree of immune infiltration. Based on the number and distribution of TILs in the tumor and stromal regions, combined with preset classification rules, bladder cancer is divided into three immune subtypes: Inflamed, Excluded, and Desert.

[0023] This invention uses deep learning technology to automatically identify and quantify the spatial distribution characteristics of tumor-infiltrating lymphocytes (TILs) in tumor regions and stromal regions of tumor tissue, enabling precise assessment of the immune microenvironment of bladder cancer patients. It constructs a standardized immunophenotyping system, thereby assisting in predicting immunotherapy response and clinical prognosis, and providing scientific basis and intelligent decision support for developing individualized treatment plans for bladder cancer patients.

[0024] Based on the above embodiments, step 100 includes: Obtain surgically removed tissue from the patient; HE-stained digital pathological whole-section images were prepared from the surgically removed tissue; A digital slide scanner was used to scan the preprocessed HE-stained digital pathological whole slide images at a preset resolution to generate full-field images.

[0025] Specifically, in this embodiment of the invention, pathological image data is first collected and processed. Surgically removed tissue from bladder cancer patients is paraffin-embedded to prepare hematoxylin and eosin (H&E) stained pathological sections. Subsequently, a digital slide scanner is used to scan the entire section under a 20× objective lens at a resolution of 0.24 μm / pixel to generate high-quality full-field-of-view (WSI) images.

[0026] Based on the above embodiments, step 200 includes: The full-view image is sequentially processed by color normalization, noise reduction, and patch segmentation to output the preprocessed pathological image data.

[0027] Specifically, in the image acquisition process of this embodiment of the invention, in order to ensure data quality and remove slices with image defects, such as bubbles, uneven staining, tissue folds, severe fading, and image defocusing, the foreground area can be initially extracted using Otsu's color thresholding method.

[0028] Based on the above embodiments, step 300 includes: The tumor region and stroma region in each image were manually delineated using biological image analysis software. If any discrepancies in the annotation were found, they were then subject to final review and confirmation by experts. The tumor region and stromal region are segmented into patches of a preset size to generate the original pathological image and a mask with labeled regions; The original pathological images are divided into training set and training set, and the tumor and stroma region segmentation model is trained. The Dice loss function is used to optimize the model to obtain the trained tumor and stroma region segmentation model.

[0029] Specifically, the embodiments of the present invention mainly construct a tumor and stromal region segmentation model.

[0030] The first step is to label the tumor and stroma regions. For pathological WSI data, pathologists with extensive experience in the pathological diagnosis of urinary tract diseases manually delineate the tumor and stroma regions in each WSI using QuPath software (version 0.5). If there is a labeling dispute, it will be reviewed and confirmed by other senior pathology experts.

[0031] The second step is the segmentation of the whole slice. In order to reduce the computational complexity, based on the manually labeled tumor and stroma regions mentioned above, the WSI is segmented into smaller 1024*1024 pixel patches to generate the original pathological image and a mask with labeled regions for training the segmentation model below.

[0032] The third step is training the semantic segmentation model. Based on class oversampling and downsampling, resampling ensures consistent data distribution. Furthermore, we divide the dataset into training and validation sets for training the TransUNet semantic segmentation model. The network structure is as follows: Figure 3 As shown in the diagram, 80% of the patch-mask paired data was used for training, and 20% for validation. The Dice loss function was selected during training to optimize model performance. A schematic diagram illustrating the tumor and stroma recognition results is shown below. Figure 4 As shown.

[0033] Based on the above embodiments, step 400 includes: Obtain the TIL database, which includes a dataset consisting of several pixel patch images of a preset size. Enlarge the dataset by a preset factor and label each patch as TIL positive or TIL negative. The dataset was divided into training and validation sets. A ResNet-50 feature extractor pre-trained based on ImageNet-1K was used, the output categories were set, and a patch classification model was trained based on the cross-entropy loss function to obtain the TIL recognition neural network classification model.

[0034] Specifically, this embodiment of the invention aims to perform bladder cancer immunophenotyping, and plans to construct a recognition model based on the TIL database to quantitatively assess immune infiltration. This dataset contains 86,104 100×100 pixel patch images, magnified 20 times, with each patch labeled as TIL-positive or TIL-negative. The dataset is divided into training and validation sets in an 80:20 ratio. A ResNet-50 feature extractor pre-trained based on ImageNet-1K is used, with two output classes. A patch classification model is trained based on the cross-entropy loss function. The relevant recognition results are illustrated in the diagram below. Figure 5 As shown.

[0035] Based on the above embodiments, step 500 includes: The number of TIL-positive patches in the tumor epithelium and the number of patches in the tumor epithelial region are used to obtain the lymphocyte infiltration score of the tumor region; The peritumoral lymphocyte infiltration score is obtained by the number of TIL-positive patches within a predetermined range of the tumor tissue edge and the number of patches within the predetermined range of the tumor tissue edge. The tumor region lymphocyte infiltration score is used to differentiate between immune-activated and immune-cold tumors; The immune rejection type and immune desert type of the immune cold tumor are distinguished based on the peritumoral lymphocyte infiltration score.

[0036] Specifically, based on the aforementioned tumor-stromal segmentation and TIL classification model and reasoning results, this embodiment of the invention calculates the number of TIL-positive patches in the tumor epithelial and stromal regions to construct a lymphocyte infiltration score. The TIL density score calculates the proportion of TIL-positive areas per unit area in different regions. The tumor-associated lymphocyte infiltration score (iTIL) is defined as the proportion of TIL-positive patches in the tumor epithelium to the tumor region; the peritumoral lymphocyte infiltration score (sTIL) is defined as the proportion of TIL-positive patches within 1 mm of the tumor edge to the area within 1 mm of the edge.

[0037]

[0038]

[0039] Immune hot tumors (immunely activated) and immune cold tumors are distinguished based on the iTIL score: an iTIL score greater than 20% is defined as an immune hot tumor (Inflamed), and an iTIL score less than or equal to 20% is defined as an immune cold tumor. For immune cold tumors, sTIL is further evaluated: an sTIL score greater than 30% is considered immune exclusion (Excluded), and an sTIL score less than or equal to 30% is classified as immune desert tumors.

[0040]

[0041] Based on the above embodiments, it also includes: The accuracy of tumor region identification, the accuracy of TIL identification and classification, and the retrospective cohort study of patients were validated respectively.

[0042] Specifically, in this embodiment of the invention, after the construction of the above-mentioned artificial intelligence system for bladder cancer immunophenotyping based on digital pathological images was completed, the system underwent rigorous data verification to ensure its accuracy, reliability and clinical applicability.

[0043] The first step was tumor region identification and validation, which involved comparing the tumor region segmentation results generated by the system with the manually annotated results from pathology experts. The results showed that the system achieved a Dice accuracy of 92% in tumor region identification. Figure 7 As shown, the system can identify tumor regions with high accuracy, providing a solid foundation for subsequent TIL identification and classification.

[0044] The second step is TIL identification and classification verification, which further verifies the accuracy of TIL identification and classification. Experimental results show that the system achieves an accuracy of 93%, a sensitivity of 90%, and a specificity of 95% in TIL identification and classification. The training process is as follows: Figure 8 As shown.

[0045] The third step was a retrospective cohort study for validation. To further verify the system's effectiveness in clinical application, this embodiment conducted a retrospective cohort study based on 450 bladder cancer patients. We performed Kaplan-Meier survival analysis on the immunophenotyping results generated by the system—immune desert, immune rejection, and immune activation—and patient survival. The results showed that immunophenotyping had a significant impact on patient survival, and the survival differences among the three immunophenotyping types were statistically significant (log-rank test, p<0.001). Specifically, patients with tumor immune activation had the best prognosis, patients with non-invasive immune rejection had a moderate prognosis, and patients with immune desert had the worst prognosis. Figure 6 The survival curves clearly demonstrate the significant differences in prognosis among the three types of immune subtypes.

[0046] The bladder cancer immune typing identification system provided by the present invention is described below. The bladder cancer immune typing identification system described below can be referred to in correspondence with the bladder cancer immune typing identification method described above.

[0047] Figure 9 This is a schematic diagram of the bladder cancer immune typing identification system provided in an embodiment of the present invention, as shown below. Figure 9 As shown, it includes: a data acquisition module 91, a preprocessing module 92, a construction module 93, a classification module 94, and a recognition module 95, wherein: The acquisition module 91 is used to acquire the patient's pathological image data; the preprocessing module 92 is used to preprocess the pathological image data to obtain preprocessed pathological image data; the construction module 93 is used to construct a tumor and stroma region segmentation model based on the semantic segmentation network TransUNet, and to train the tumor and stroma region segmentation model using the preprocessed pathological image data and label data to obtain a trained tumor and stroma region segmentation model; the classification module 94 is used to construct a TIL recognition neural network model based on ResNet50 using the tumor and stroma segmentation results output by the trained tumor and stroma region segmentation model; the recognition module 95 is used to obtain bladder cancer immunophenotyping results based on the lymphocyte infiltration score output by the TIL recognition neural network model classification model.

[0048] Figure 10 An example is a schematic diagram of the physical structure of an electronic device, such as... Figure 10As shown, the electronic device may include: a processor 1010, a communication interface 1020, a memory 1030, and a communication bus 1040, wherein the processor 1010, the communication interface 1020, and the memory 1030 communicate with each other through the communication bus 1040. The processor 1010 can call logical instructions in the memory 1030 to execute a bladder cancer immunophenotyping identification method, which includes: acquiring pathological image data of the patient; preprocessing the pathological image data to obtain preprocessed pathological image data; constructing a tumor and stroma region segmentation model based on the semantic segmentation network TransUNet; training the tumor and stroma region segmentation model using the preprocessed pathological image data and label data to obtain a trained tumor and stroma region segmentation model; constructing a ResNet50-based TIL recognition neural network model classification model using the tumor and stroma region segmentation results output by the trained tumor and stroma region segmentation model; and obtaining the bladder cancer immunophenotyping result based on the lymphocyte infiltration score output by the TIL recognition neural network model classification model.

[0049] Furthermore, the logical instructions in the aforementioned memory 1030 can be implemented as software functional units and, when sold or used as independent products, can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention, or the part that contributes to the prior art, or a part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of the present invention. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0050] On the other hand, the present invention also provides a non-transitory computer-readable storage medium storing a computer program thereon, which, when executed by a processor, implements the bladder cancer immunophenotyping identification method provided by the above methods. This method includes: acquiring pathological image data of a patient; preprocessing the pathological image data to obtain preprocessed pathological image data; constructing a tumor and stroma region segmentation model based on the semantic segmentation network TransUNet; training the tumor and stroma region segmentation model using the preprocessed pathological image data and label data to obtain a trained tumor and stroma region segmentation model; constructing a ResNet50-based TIL recognition neural network model classification model using the tumor and stroma region segmentation results output by the trained model; and obtaining bladder cancer immunophenotyping results based on the lymphocyte infiltration score output by the TIL recognition neural network model classification model.

[0051] The device embodiments described above are merely illustrative. The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the modules can be selected to achieve the purpose of this embodiment according to actual needs. Those skilled in the art can understand and implement this without any creative effort.

[0052] Through the above description of the embodiments, those skilled in the art can clearly understand that each embodiment can be implemented by means of software plus necessary general-purpose hardware platforms, and of course, it can also be implemented by hardware. Based on this understanding, the above technical solutions, 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 can be stored in a computer-readable storage medium, such as ROM / RAM, disk, optical disk, etc., and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute the methods described in the various embodiments or some parts of the embodiments.

[0053] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, and not to limit them; although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features; and these modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A method for identifying bladder cancer through immunophenotyping, characterized in that, include: Collect pathological image data from patients; The pathological image data is preprocessed to obtain preprocessed pathological image data; A tumor and stroma region segmentation model was constructed based on the semantic segmentation network TransUNet. The preprocessed pathological image data and label data were used to train the tumor and stroma region segmentation model to obtain a trained tumor and stroma region segmentation model. Using the tumor and stroma segmentation results output by the trained tumor and stroma region segmentation model, a TIL recognition neural network model based on ResNet50 is constructed for classification. Based on the lymphocyte infiltration score output by the TIL recognition neural network model classification model, the bladder cancer immunophenotyping results are obtained.

2. The bladder cancer immunophenotyping identification method according to claim 1, characterized in that, Collect the patient's pathological image data, including: Obtain surgically removed tissue from the patient; HE-stained digital pathological whole-section images were prepared from the surgically removed tissue; A digital slide scanner was used to scan the preprocessed HE-stained digital pathological whole slide images at a preset resolution to generate full-field images.

3. The bladder cancer immunophenotyping identification method according to claim 2, characterized in that, The pathological image data is preprocessed to obtain preprocessed pathological image data, including: The full-view image is sequentially processed by color normalization, noise reduction, and patch segmentation to output the preprocessed pathological image data.

4. The bladder cancer immunophenotyping identification method according to claim 1, characterized in that, A tumor and stroma region segmentation model is constructed based on the semantic segmentation network TransUNet. The preprocessed pathological image data and labeled data are used to train the tumor and stroma region segmentation model, resulting in a trained model, including: The tumor region and stroma region in each image were manually delineated using biological image analysis software. If any discrepancies in the annotation were found, they were then subject to final review and confirmation by experts. The tumor region and stromal region are segmented into patches of a preset size to generate the original pathological image and a mask with labeled regions; The original pathological images are divided into training set and training set, and the tumor and stroma region segmentation model is trained. The Dice loss function is used to optimize the model to obtain the trained tumor and stroma region segmentation model.

5. The bladder cancer immunophenotyping identification method according to claim 1, characterized in that, Using the tumor and stroma segmentation results output by the trained tumor and stroma region segmentation model, a ResNet50-based TIL identification neural network classification model is constructed to output a lymphocyte infiltration score, including: Obtain the TIL database, which includes a dataset consisting of several pixel patch images of a preset size. Enlarge the dataset by a preset factor and label each patch as TIL positive or TIL negative. The dataset was divided into training and validation sets. A ResNet-50 feature extractor pre-trained based on ImageNet-1K was used, the output categories were set, and a patch classification model was trained based on the cross-entropy loss function to obtain the TIL recognition neural network classification model.

6. The bladder cancer immunophenotyping identification method according to claim 1, characterized in that, Based on the lymphocyte infiltration score output by the TIL recognition neural network model classification model, the bladder cancer immunophenotyping results are obtained, including: The number of TIL-positive patches in the tumor epithelium and the number of patches in the tumor epithelial region are used to obtain the lymphocyte infiltration score of the tumor region; The peritumoral lymphocyte infiltration score is obtained by the number of TIL-positive patches within a predetermined range of the tumor tissue edge and the number of patches within the predetermined range of the tumor tissue edge. The tumor region lymphocyte infiltration score is used to differentiate between immune-activated and immune-cold tumors; The immune rejection type and immune desert type of the immune cold tumor are distinguished based on the peritumoral lymphocyte infiltration score.

7. The bladder cancer immunophenotyping identification method according to claim 1, characterized in that, Also includes: The accuracy of tumor region identification, the accuracy of TIL identification and classification, and the retrospective cohort study of patients were validated respectively.

8. A bladder cancer immunophenotyping and identification system, characterized in that, include: The acquisition module is used to acquire pathological image data of patients; The preprocessing module is used to preprocess the pathological image data to obtain preprocessed pathological image data. The module is used to construct a tumor and stroma region segmentation model based on the semantic segmentation network TransUNet. The preprocessed pathological image data and label data are used to train the tumor and stroma region segmentation model to obtain a trained tumor and stroma region segmentation model. The classification module is used to construct a TIL recognition neural network model based on ResNet50 using the tumor and stroma segmentation results output by the trained tumor and stroma region segmentation model. The identification module is used to obtain the bladder cancer immunophenotyping result based on the lymphocyte infiltration score output by the TIL identification neural network model classification model.

9. An electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that, When the processor executes the program, it implements the bladder cancer immune typing identification method as described in any one of claims 1 to 7.

10. A non-transitory computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by the processor, it implements the bladder cancer immunophenotyping identification method as described in any one of claims 1 to 7.

Images