PD-L1 immunohistochemical section tumor recognition detection method and system

Abstract

The invention provides a PD-L1 immunohistochemical section tumor identification detection method and system, and the method comprises the steps: collecting various organ PD-L1 sections of a clinical pathological experiment, carrying out high-resolution scanning, pre-identification screening and cloud data storage, and obtaining a PD-L1 section data set of various organs; automatically labeling each tumor cell labeling pixel-level mask, automatically generating a labeling bounding box, and obtaining a tumor cell region labeling data set; according to the tumor cell region labeling data set, preprocessing all positions labeled as effective regions, creating a YOLO v8 deep learning model, training to gradually adjust network weight, and outputting a prediction result of the YOLO v8 deep learning model; and analyzing and calculating whether the prediction evaluation index information of the YOLO v8 deep learning model accords with a training target evaluation index, completing model training, and carrying out accurate positioning and pixel-level region identification detection on PD-L1 immunohistochemical pathological section tumor cells.

Description

Technical Field

[0001] The present invention relates to the field of automated screening mask model training target recognition technology, and more specifically, to a PD-L1 immunohistochemical slice tumor recognition detection method and system. Background Art

[0002] PD-L1 (Programmed Death-Ligand 1) is an immune checkpoint molecule that plays an important regulatory role in the immune system. It inhibits the activation of T cells by binding to the PD-1 receptor, thereby avoiding excessive immune response. However, some tumor cells can express high levels of PD-L1 to escape the attack of the immune system, leading to tumor immune escape and progression. Therefore, the quantitative detection and localization of PD-L1 are crucial for the effectiveness evaluation of tumor immunotherapy and the formulation of treatment strategies. Currently, immunohistochemistry (IHC) technology is a common method for detecting PD-L1 expression. However, traditional IHC methods have some limitations in tumor cell localization and counting. First, manual localization and counting of tumor cells is a time-consuming and subjective task, which can easily lead to Errors and inconsistencies; secondly, the positioning of tumor cells is usually limited to visual judgment and lacks precise quantitative information; these problems limit the accuracy and reliability of PD-L1 expression detection; in order to overcome these problems, how to further expand the application of computer-assisted methods in immunohistochemical pathological slice image analysis, how to use deep learning models to enhance image analysis and further improve efficiency and accuracy in various medical image recognition and segmentation tasks, how to create training and optimization target detection models in tumor cell recognition and enhance model performance in target positioning and recognition, etc. have yet to be solved; therefore, it is necessary to propose a PD-L1 immunohistochemical slice tumor recognition detection method and system to at least partially solve the problems existing in the prior art. Summary of the invention

[0003] A series of simplified concepts are introduced in the content of the invention, which will be further described in detail in the specific implementation method section; the content of the invention of the present invention does not mean to attempt to limit the key features and essential technical features of the technical solution claimed for protection, nor does it mean to attempt to determine the scope of protection of the technical solution claimed for protection.

[0004] In order to at least partially solve the above problems, the present invention provides a PD-L1 immunohistochemical section tumor recognition detection method, comprising:

[0005] S100 collects PD-L1 slices of various organs for clinical pathology experiments, performs high-resolution scanning, pre-identification screening and cloud data storage, and obtains PD-L1 slice data sets of various organs;

[0006] S200, based on the PD-L1 slice datasets of various organs, automatically annotates each tumor cell with a pixel-level mask, automatically generates annotated bounding boxes, and obtains tumor cell region annotated datasets;

[0007] S300, based on the tumor cell region annotation dataset, preprocesses all locations annotated as valid regions, creates a YOLO v8 deep learning model and performs training to gradually adjust the network weights, and outputs the prediction results of the YOLO v8 deep learning model;

[0008] S400, analyzes and calculates whether the prediction evaluation index information of the YOLO v8 deep learning model meets the training target evaluation index, completes the model training, and performs accurate positioning and pixel-level region recognition detection of tumor cells in PD-L1 immunohistochemistry pathology sections.

[0009] Preferably, S100 includes:

[0010] S101, collect PD-L1 slices of various organs for clinical pathology experiments, and obtain PD-L1 slice data of various organs through high-resolution digital scanning of PD-L1 slices;

[0011] S102, pre-identify and screen the PD-L1 slice data of various organs to obtain the PD-L1 slice data of various organs containing sufficient tumor cell samples and sufficient PD-L1 expression areas;

[0012] S103, storing the PD-L1 slice data of various organs in the cloud, and obtaining the PD-L1 slice data sets of various organs.

[0013] Preferably, S200 includes:

[0014] S201, based on the PD-L1 slice datasets of various organs, the PD-L1 slice annotation expert model is used to input the cell morphological characteristics of the PD-L1 slices and the PD-L1 staining results, and output the location of the tumor cells to be annotated and the distribution area of ​​the tumor cells to be annotated;

[0015] S202, according to the position of the tumor cells to be annotated and the distribution area of ​​the tumor cells to be annotated, setting the minimum number of annotated cells for each slice; the minimum number of annotated cells for each slice is 5000;

[0016] S203, automatically annotate each tumor cell annotated pixel-level mask through the pixel-level mask automatic annotation module; automatically generate an annotation boundary box through the boundary marking program module; the boundary marking program module first generates a minimum bounding box according to the center of any tumor cell annotated pixel-level mask of the pixel-level mask automatic annotation module; expand layer by layer from the inside to the outside with the minimum bounding box as the center until the bounding box covers any tumor cell annotated pixel-level mask, and generate an annotation boundary box; in the annotation process, select the boundary box to be annotated using the valid area label, and then annotate all tumor cells in it without omission; select the boundary box for the undetermined area using the ignore area label; obtain the tumor cell area annotation data set.

[0017] Preferably, S300 includes:

[0018] S301, based on the tumor cell region labeling dataset, pre-process all the locations marked as valid regions and divide the training image blocks; assign 1 pixel labels to all tumor regions in the training image blocks and give boundaries; assign 255 pixel labels to the pixels in the training image blocks that exceed the valid region or are marked as ignored; assign 0 pixel labels to all other pixels in the training image blocks; during the training process, remove the loss function contributed by all 255 pixels;

[0019] S302, using a single neural network to directly predict the entire image, and output the target prediction bounding box and target prediction category probability, and output the pixel-level segmentation result to create a YOLO v8 deep learning model;

[0020] S303, using the tumor cell region annotation dataset as a training sample, train the YOLO v8 deep learning model; optimize the model parameters through the back propagation algorithm; use the cross entropy loss and bounding box regression loss to measure the difference between the model prediction results and the labels, and gradually adjust the network weights to output the YOLO v8 deep learning model prediction results.

[0021] Preferably, S400 includes:

[0022] S401, using an untrained PD-L1 image independent test dataset, comparing the prediction results of the YOLO v8 deep learning model with the test dataset labels, calculating the model prediction accuracy, model prediction recall, and model prediction F1 score, and obtaining model prediction evaluation indicator information;

[0023] S402, determine whether the model prediction evaluation index information meets the training target evaluation index; when the model prediction evaluation index information does not meet the training target evaluation index, retrain the YOLO v8 deep learning model until the model prediction evaluation index information meets the training target evaluation index; when the model prediction evaluation index information meets the training target evaluation index, complete the training to obtain the PD-L1 image target evaluation YOLO v8 deep learning model; S403, perform actual PD-L1 immunohistochemical pathology section image recognition through the PD-L1 image target evaluation YOLO v8 deep learning model, count tumor cells, extract morphological features and analyze PD-L1 expression levels for the recognition results, and achieve accurate positioning of tumor cells in PD-L1 immunohistochemical pathology sections and pixel-level area recognition and detection.

[0024] The present invention provides a PD-L1 immunohistochemical section tumor recognition and detection system, comprising:

[0025] The PD-L1 slice collection and processing subsystem collects PD-L1 slices of various organs for clinical pathology experiments, performs high-resolution scanning, pre-identification screening and cloud data storage, and obtains PD-L1 slice data sets of various organs;

[0026] The tumor cell mask annotation subsystem automatically annotates the pixel-level mask of each tumor cell based on the PD-L1 slice datasets of various organs, automatically generates annotated bounding boxes, and obtains the tumor cell region annotation dataset;

[0027] The YOLO v8 model creates a training subsystem that pre-processes all locations marked as valid areas based on the tumor cell region annotation dataset, creates a YOLO v8 deep learning model, performs training, and gradually adjusts the network weights, outputting the prediction results of the YOLO v8 deep learning model.

[0028] The prediction results evaluate the tumor recognition and detection subsystem, analyze and calculate whether the prediction evaluation index information of the YOLO v8 deep learning model meets the training target evaluation index, and complete the model training to accurately locate the tumor cells in the PD-L1 immunohistochemistry pathology sections and identify the pixel-level area.

[0029] Preferably, the PD-L1 slice collection and processing subsystem includes:

[0030] The PD-L1 slice acquisition high-resolution scanning subsystem collects PD-L1 slices of various organs for clinical pathology experiments, and obtains PD-L1 slice data of various organs through high-resolution digital scanning of PD-L1 slices;

[0031] The slice data pre-identification and screening subsystem pre-identifies and screens the PD-L1 slice data of various organs, and obtains PD-L1 slice data of various organs that contain sufficient tumor cell samples and sufficient PD-L1 expression areas;

[0032] The PD-L1 slice data cloud storage subsystem stores the PD-L1 slice data of various organs in the cloud and obtains the PD-L1 slice data sets of various organs.

[0033] Preferably, the tumor cell mask annotation subsystem includes:

[0034] The PD-L1 slice annotation expert model subsystem uses the PD-L1 slice annotation expert model to input the cell morphological characteristics of the PD-L1 slices and the PD-L1 staining results based on the PD-L1 slice datasets of various organs, and outputs the location of the tumor cells to be annotated and the distribution area of ​​the tumor cells to be annotated;

[0035] The subsystem for setting the number of labeled cells sets the minimum number of labeled cells for each slice according to the location of the tumor cells to be labeled and the distribution area of ​​the tumor cells to be labeled; the minimum number of labeled cells for each slice is 5000;

[0036] The mask automatic annotation box selection subsystem automatically annotates each tumor cell annotation pixel-level mask through the pixel-level mask automatic annotation module; automatically generates an annotation bounding box through the boundary marking program module; the boundary marking program module first generates a minimum bounding box according to the center of any tumor cell annotation pixel-level mask of the pixel-level mask automatic annotation module; expands layer by layer from the inside to the outside with the minimum bounding box as the center until the bounding box covers any tumor cell annotation pixel-level mask to generate an annotation bounding box; in the annotation process, the bounding box selection is annotated with the valid area label for the area to be annotated, and then all tumor cells are annotated without omission; the bounding box selection is annotated with the ignore area label for the undetermined area; and the tumor cell area annotation data set is obtained.

[0037] Preferably, the YOLO v8 model creation training subsystem includes:

[0038] The labeled region preprocessing subsystem preprocesses all the locations marked as valid regions according to the tumor cell region labeling dataset and divides the training image blocks; assigns 1 pixel labels to all tumor regions in the training image blocks and gives boundaries; assigns 255 pixel labels to pixels in the training image blocks that exceed the valid region or are marked as ignored; assigns 0 pixel labels to all other pixels in the training image blocks; during the training process, removes the loss function contributed by all 255 pixels;

[0039] The single neural network YOLO v8 model subsystem uses a single neural network to make predictions directly on the entire image and output the target prediction bounding box and target prediction class probability, while outputting pixel-level segmentation results to create a YOLO v8 deep learning model;

[0040] The tumor annotation data model training subsystem uses the tumor cell region annotation dataset as training samples to train the YOLO v8 deep learning model; optimizes the model parameters through the back propagation algorithm; uses the cross entropy loss and bounding box regression loss to measure the difference between the model prediction results and the labels, and gradually adjusts the network weights to output the YOLO v8 deep learning model prediction results.

[0041] Preferably, the prediction result evaluation tumor recognition detection subsystem includes:

[0042] The model prediction comparison calculation subsystem compares the prediction results of the YOLO v8 deep learning model with the test dataset labels through an independent test dataset of untrained PD-L1 images, calculates the model prediction accuracy, model prediction recall, and model prediction F1 score, and obtains model prediction evaluation indicator information;

[0043] The training target evaluation determination subsystem determines whether the model prediction evaluation index information meets the training target evaluation index; when the model prediction evaluation index information does not meet the training target evaluation index, the YOLO v8 deep learning model is retrained until the model prediction evaluation index information meets the training target evaluation index; when the model prediction evaluation index information meets the training target evaluation index, the training is completed to obtain the PD-L1 image target evaluation YOLO v8 deep learning model;

[0044] The PD-L1 immunohistochemistry analysis image recognition subsystem uses the PD-L1 image target evaluation YOLO v8 deep learning model to perform actual PD-L1 immunohistochemistry pathology slice image recognition, and performs tumor cell counting, morphological feature extraction, and PD-L1 expression level analysis on the recognition results to achieve accurate positioning of tumor cells in PD-L1 immunohistochemistry pathology slices and pixel-level region recognition and detection.

[0045] Compared with the prior art, the present invention has at least the following beneficial effects:

[0046] The present invention provides a PD-L1 immunohistochemical section tumor recognition and detection method and system, including: collecting PD-L1 sections of various organs in clinical pathology experiments, performing high-resolution scanning, pre-recognition screening and cloud data storage, and obtaining PD-L1 section data sets of various organs; according to the PD-L1 section data sets of various organs, automatically labeling each tumor cell labeling pixel-level mask, automatically generating a labeling boundary box, and obtaining a tumor cell area labeling data set; according to the tumor cell area labeling data set, pre-processing all positions labeled as valid areas, creating a YOLO v8 deep learning model and training it to gradually adjust the network weights, and outputting the YOLO v8 deep learning model prediction results; analyzing and calculating whether the YOLO v8 deep learning model prediction evaluation index information meets the training target evaluation index, completing the model training to accurately locate the PD-L1 immunohistochemical pathology section tumor cells and pixel-level area recognition and detection; the present invention scheme uses YOLO The YOLO v8 deep learning model can accurately locate tumor cells and identify pixel-level regions by annotating tumor cells in PD-L1 slices. Compared with traditional methods, firstly, the automated and standardized image analysis process can reduce manual operations and subjective errors, and improve the consistency and reliability of tumor cell positioning and counting. Secondly, the application of the YOLO v8 model can achieve fast and accurate tumor cell recognition, providing an effective solution for the processing of large-scale image data. Finally, by obtaining pixel-level regional information, the morphological characteristics of tumor cells and the PD-L1 expression level can be further analyzed to provide support for the evaluation of tumor immunotherapy and the formulation of individualized treatment plans. The present invention is applicable to a wide range of scenarios in the field of tumor cell recognition in PD-L1 immunohistochemical pathological slices. It can be widely used in clinical pathology for quantitative detection and positioning of tumor PD-L1 expression, providing an important reference for the efficacy evaluation of tumor immunotherapy and the formulation of individualized treatment plans. It can also promote an in-depth understanding of the mechanism of tumor immune escape in the field of scientific research, and provide strong support for the development and optimization of new immunotherapy strategies. In the industrial field, it can provide efficient and accurate image analysis solutions for pathology laboratories to improve work efficiency and diagnostic accuracy.

[0047] The PD-L1 immunohistochemical section tumor recognition detection method and system described in the present invention, other advantages, objectives and features of the present invention will be partially reflected in the following description, and will also be understood by technicians in this field through research and practice of the present invention. BRIEF DESCRIPTION OF THE DRAWINGS

[0048] The accompanying drawings are used to provide a further understanding of the present invention and constitute a part of the specification. Together with the embodiments of the present invention, they are used to explain the present invention and do not constitute a limitation of the present invention. In the accompanying drawings:

[0049] Figure 1 This is a diagram showing an embodiment of the prediction of a PD-L1 immunohistochemical section tumor recognition detection method described in the present invention.

[0050] Figure 2 This is a diagram showing an embodiment of the prediction of a PD-L1 immunohistochemical section tumor recognition detection method described in the present invention.

[0051] Figure 3 This is a diagram showing an embodiment of the PD-L1 immunohistochemical section tumor recognition and detection method described in the present invention.

[0052] Figure 4 This is a diagram of an embodiment of a PD-L1 immunohistochemical section tumor recognition and detection system according to the present invention.

[0053] Figure 5 This is another embodiment diagram of the PD-L1 immunohistochemical section tumor recognition and detection system described in the present invention. DETAILED DESCRIPTION

[0054] The present invention will be further described in detail below with reference to the accompanying drawings and embodiments so that those skilled in the art can implement the invention according to the instructions. Figure 1-Figure 5 As shown, the present invention provides a PD-L1 immunohistochemical section tumor recognition detection method, comprising:

[0055] S100 collects PD-L1 slices of various organs for clinical pathology experiments, performs high-resolution scanning, pre-identification screening and cloud data storage, and obtains PD-L1 slice data sets of various organs;

[0056] S200, based on the PD-L1 slice datasets of various organs, automatically annotates each tumor cell with a pixel-level mask, automatically generates annotated bounding boxes, and obtains tumor cell region annotated datasets;

[0057] S300, based on the tumor cell region annotation dataset, preprocesses all locations annotated as valid regions, creates a YOLO v8 deep learning model and performs training to gradually adjust the network weights, and outputs the prediction results of the YOLO v8 deep learning model;

[0058] S400, analyzes and calculates whether the prediction evaluation index information of the YOLO v8 deep learning model meets the training target evaluation index, completes the model training, and performs accurate positioning and pixel-level region recognition detection of tumor cells in PD-L1 immunohistochemistry pathology sections.

[0059] The principle and effect of the above technical solution are as follows: the present invention provides a PD-L1 immunohistochemical section tumor recognition and detection method, including: collecting PD-L1 sections of various organs in clinical pathology experiments, performing high-resolution scanning, pre-recognition screening and cloud data storage, and obtaining PD-L1 section data sets of various organs; according to the PD-L1 section data sets of various organs, automatically annotating each tumor cell annotated pixel-level mask, automatically generating annotated bounding boxes, and obtaining tumor cell area annotated data sets; according to the tumor cell area annotated data sets, preprocessing all positions annotated as valid areas, creating a YOLO v8 deep learning model and training to gradually adjust the network weights, and outputting the prediction results of the YOLO v8 deep learning model; analyzing and calculating whether the prediction evaluation index information of the YOLO v8 deep learning model meets the training target evaluation index, completing the model training to accurately locate the tumor cells in the PD-L1 immunohistochemical pathology sections and perform pixel-level area recognition and detection;

[0060] The solution of the present invention uses the YOLO v8 deep learning model to annotate tumor cells in PD-L1 slices to achieve accurate positioning of tumor cells and identification of pixel-level areas. Compared with traditional methods, firstly, through the automated and standardized image analysis process, manual operations and subjective errors can be reduced, and the consistency and reliability of tumor cell positioning and counting can be improved. Secondly, through YOLO The application of the v8 model can achieve fast and accurate tumor cell identification, providing an effective solution for the processing of large-scale image data; finally, by obtaining pixel-level regional information, the morphological characteristics of tumor cells and the PD-L1 expression level can be further analyzed to provide support for the evaluation of tumor immunotherapy and the formulation of individualized treatment plans; various organs include: respiratory organs, digestive organs, nervous organs or motor organs; and the present invention is applicable to a wide range of scenarios in the field of tumor cell identification in PD-L1 immunohistochemical pathological sections; it can be widely used in clinical pathology for quantitative detection and positioning of tumor PD-L1 expression, providing an important reference for the efficacy evaluation of tumor immunotherapy and the formulation of individualized treatment plans; it can also promote an in-depth understanding of the mechanism of tumor immune escape in the field of scientific research, and provide strong support for the development and optimization of new immunotherapy strategies; in the industrial field, it can provide pathology laboratories with efficient and accurate image analysis solutions to improve work efficiency and diagnostic accuracy.

[0061] In one embodiment, S100 includes:

[0062] S101, collect PD-L1 slices of various organs for clinical pathology experiments, and obtain PD-L1 slice data of various organs through high-resolution digital scanning of PD-L1 slices;

[0063] S102, pre-identify and screen the PD-L1 slice data of various organs to obtain the PD-L1 slice data of various organs containing sufficient tumor cell samples and sufficient PD-L1 expression areas;

[0064] S103, storing the PD-L1 slice data of various organs in the cloud, and obtaining the PD-L1 slice data sets of various organs.

[0065] The principle and effect of the above technical solution are: collecting PD-L1 slices of various organs in clinical pathology experiments, and obtaining PD-L1 slice data of various organs through high-resolution digital scanning of PD-L1 slices; pre-identifying and screening the PD-L1 slice data of various organs, and obtaining PD-L1 slice data of various organs containing sufficient tumor cell samples and sufficient PD-L1 expression areas; storing the PD-L1 slice data of various organs in the cloud, and obtaining PD-L1 slice data sets of various organs; the scope of computer-assisted methods in immunohistochemical pathology slice image analysis can be expanded.

[0066] In one embodiment, S200 includes:

[0067] S201, based on the PD-L1 slice datasets of various organs, the PD-L1 slice annotation expert model is used to input the cell morphological characteristics of the PD-L1 slices and the PD-L1 staining results, and output the location of the tumor cells to be annotated and the distribution area of ​​the tumor cells to be annotated;

[0068] S202, according to the position of the tumor cells to be annotated and the distribution area of ​​the tumor cells to be annotated, setting the minimum number of annotated cells for each slice; the minimum number of annotated cells for each slice is 5000;

[0069] S203, automatically annotate each tumor cell annotated pixel-level mask through the pixel-level mask automatic annotation module; automatically generate an annotation boundary box through the boundary marking program module; the boundary marking program module first generates a minimum bounding box according to the center of any tumor cell annotated pixel-level mask of the pixel-level mask automatic annotation module; expand layer by layer from the inside to the outside with the minimum bounding box as the center until the bounding box covers any tumor cell annotated pixel-level mask, and generate an annotation boundary box; in the annotation process, select the boundary box to be annotated using the valid area label, and then annotate all tumor cells in it without omission; select the boundary box for the undetermined area using the ignore area label; obtain the tumor cell area annotation data set.

[0070] The principle and effect of the above technical solution are as follows: according to the PD-L1 slice data set of various organs, the cell morphological characteristics of the PD-L1 slice and the PD-L1 staining results are input through the PD-L1 slice annotation expert model, and the position of the tumor cells to be annotated and the distribution area of ​​the tumor cells to be annotated are output; according to the position of the tumor cells to be annotated and the distribution area of ​​the tumor cells to be annotated, the minimum number of annotated cells per slice is set; the minimum number of annotated cells per slice is 5000; through the pixel-level mask automatic annotation module, each tumor cell is automatically annotated with a pixel-level mask; the boundary marking program module automatically generates an annotation boundary box; the boundary marking program The module first generates a minimum bounding box based on the center of any tumor cell annotation pixel-level mask of the pixel-level mask automatic annotation module; it expands layer by layer from the inside to the outside with the minimum bounding box as the center until the bounding box covers any tumor cell annotation pixel-level mask to generate an annotation bounding box; in the annotation process, the bounding box is selected for the annotation area using the valid area label, and then all tumor cells are annotated without omission; the bounding box is selected for the undetermined area using the ignored area label; a tumor cell area annotation data set is obtained; the deep learning model can be used to enhance image analysis and further improve efficiency and accuracy in various medical image recognition and segmentation tasks.

[0071] In one embodiment, S300 includes:

[0072] S301, based on the tumor cell region labeling dataset, pre-process all the locations marked as valid regions and divide the training image blocks; assign 1 pixel labels to all tumor regions in the training image blocks and give boundaries; assign 255 pixel labels to the pixels in the training image blocks that exceed the valid region or are marked as ignored; assign 0 pixel labels to all other pixels in the training image blocks; during the training process, remove the loss function contributed by all 255 pixels;

[0073] S302, using a single neural network to directly predict the entire image, and output the target prediction bounding box and target prediction category probability, and output the pixel-level segmentation result to create a YOLO v8 deep learning model;

[0074] S303, using the tumor cell region annotation dataset as a training sample, train the YOLO v8 deep learning model; optimize the model parameters through the back propagation algorithm; use the cross entropy loss and bounding box regression loss to measure the difference between the model prediction results and the labels, and gradually adjust the network weights to output the YOLO v8 deep learning model prediction results.

[0075] The principle and effect of the above technical solution are as follows: according to the tumor cell area annotation dataset, all positions marked as valid areas are preprocessed to divide the training image blocks; all tumor areas in the training image blocks are assigned 1 pixel labels and given boundaries; for the pixels in the training image blocks that exceed the valid area or are marked as ignored, 255 pixel labels are assigned; all other pixels in the training image blocks are assigned 0 pixel labels; during the training process, the loss function contributed by all 255 pixels is removed; a single neural network is used to directly predict the entire image, and the target prediction bounding box and target prediction category probability are output, and the pixel-level segmentation results are output to create a YOLO v8 deep learning model; the tumor cell area annotation dataset is used as a training sample to train the YOLO v8 deep learning model; the model parameters are optimized through the back propagation algorithm; the cross entropy loss and bounding box regression loss are used to measure the difference between the model prediction results and the labels, and the network weights are gradually adjusted to output the prediction results of the YOLO v8 deep learning model; a training optimization target detection model is created in tumor cell recognition, and the output prediction results are more accurate.

[0076] In one embodiment, S400 includes:

[0077] S401, using an untrained PD-L1 image independent test dataset, comparing the prediction results of the YOLO v8 deep learning model with the test dataset labels, calculating the model prediction accuracy, model prediction recall, and model prediction F1 score, and obtaining model prediction evaluation indicator information;

[0078] S402, determine whether the model prediction evaluation index information meets the training target evaluation index; when the model prediction evaluation index information does not meet the training target evaluation index, retrain the YOLO v8 deep learning model until the model prediction evaluation index information meets the training target evaluation index; when the model prediction evaluation index information meets the training target evaluation index, complete the training to obtain the PD-L1 image target evaluation YOLO v8 deep learning model; S403, perform actual PD-L1 immunohistochemical pathology section image recognition through the PD-L1 image target evaluation YOLO v8 deep learning model, count tumor cells, extract morphological features and analyze PD-L1 expression levels for the recognition results, and achieve accurate positioning of tumor cells in PD-L1 immunohistochemical pathology sections and pixel-level area recognition and detection.

[0079] The principle and effect of the above technical solution are as follows: through an untrained PD-L1 image independent test data set, the prediction result of the YOLOv8 deep learning model is compared with the test data set label, the model prediction accuracy, model prediction recall and model prediction F1 score are calculated, and the model prediction evaluation index information is obtained; it is determined whether the model prediction evaluation index information meets the training target evaluation index; when the model prediction evaluation index information does not meet the training target evaluation index, the YOLO v8 deep learning model is retrained until the model prediction evaluation index information meets the training target evaluation index; when the model prediction evaluation index information meets the training target evaluation index, the training is completed to obtain the PD-L1 image target evaluation YOLO v8 deep learning model; S403, through the PD-L1 image target evaluation YOLO v8 deep learning model, actual PD-L1 immunohistochemical pathology section image recognition is performed, tumor cell counting, morphological feature extraction and PD-L1 expression level analysis are performed on the recognition results, and accurate positioning and pixel-level area recognition and detection of tumor cells in PD-L1 immunohistochemical pathology sections are achieved; the target positioning accuracy and recognition detection enhancement model performance can be increased.

[0080] The present invention provides a PD-L1 immunohistochemical section tumor recognition and detection system, comprising:

[0081] The PD-L1 slice collection and processing subsystem collects PD-L1 slices of various organs for clinical pathology experiments, performs high-resolution scanning, pre-identification screening and cloud data storage, and obtains PD-L1 slice data sets of various organs;

[0082] The tumor cell mask annotation subsystem automatically annotates the pixel-level mask of each tumor cell based on the PD-L1 slice datasets of various organs, automatically generates annotated bounding boxes, and obtains the tumor cell region annotation dataset;

[0083] The YOLO v8 model creates a training subsystem that pre-processes all locations marked as valid areas based on the tumor cell region annotation dataset, creates a YOLO v8 deep learning model, performs training, and gradually adjusts the network weights, outputting the prediction results of the YOLO v8 deep learning model.

[0084] The prediction results evaluate the tumor recognition and detection subsystem, analyze and calculate whether the prediction evaluation index information of the YOLO v8 deep learning model meets the training target evaluation index, and complete the model training to accurately locate the tumor cells in the PD-L1 immunohistochemistry pathology sections and identify the pixel-level area.

[0085] The principle and effect of the above technical scheme are as follows: the present invention provides a PD-L1 immunohistochemical slice tumor recognition and detection system, including: a PD-L1 slice collection and processing subsystem, which collects PD-L1 slices of various organs in clinical pathology experiments, performs high-resolution scanning, pre-recognition screening and cloud data storage, and obtains PD-L1 slice data sets of various organs; a tumor cell mask annotation subsystem, which automatically annotates each tumor cell with a pixel-level mask according to the PD-L1 slice data sets of various organs, automatically generates an annotation boundary box, and obtains a tumor cell region annotation data set; a YOLO v8 model creation and training subsystem, which preprocesses all positions marked as valid areas according to the tumor cell region annotation data set, creates a YOLO v8 deep learning model and performs training to gradually adjust the network weights, and outputs the prediction results of the YOLO v8 deep learning model; a prediction result evaluation tumor recognition and detection subsystem, which analyzes and calculates whether the prediction evaluation index information of the YOLO v8 deep learning model meets the training target evaluation index, completes model training, and performs accurate positioning and pixel-level region recognition and detection of tumor cells in PD-L1 immunohistochemical pathology slices; the scheme of the present invention uses YOLO The YOLO v8 deep learning model can accurately locate tumor cells and identify pixel-level regions by annotating tumor cells in PD-L1 slices. Compared with traditional methods, firstly, the automated and standardized image analysis process can reduce manual operations and subjective errors, and improve the consistency and reliability of tumor cell positioning and counting. Secondly, the application of the YOLO v8 model can achieve fast and accurate tumor cell recognition, providing an effective solution for the processing of large-scale image data. Finally, by obtaining pixel-level regional information, the morphological characteristics of tumor cells and the PD-L1 expression level can be further analyzed to provide support for the evaluation of tumor immunotherapy and the formulation of individualized treatment plans. The present invention is applicable to a wide range of scenarios in the field of tumor cell recognition in PD-L1 immunohistochemical pathological slices. It can be widely used in clinical pathology for quantitative detection and positioning of tumor PD-L1 expression, providing an important reference for the efficacy evaluation of tumor immunotherapy and the formulation of individualized treatment plans. It can also promote an in-depth understanding of the mechanism of tumor immune escape in the field of scientific research, and provide strong support for the development and optimization of new immunotherapy strategies. In the industrial field, it can provide efficient and accurate image analysis solutions for pathology laboratories to improve work efficiency and diagnostic accuracy.

[0086] In one embodiment, the PD-L1 slice collection and processing subsystem includes:

[0087] The PD-L1 slice acquisition high-resolution scanning subsystem collects PD-L1 slices of various organs for clinical pathology experiments, and obtains PD-L1 slice data of various organs through high-resolution digital scanning of PD-L1 slices;

[0088] The slice data pre-identification and screening subsystem pre-identifies and screens the PD-L1 slice data of various organs, and obtains PD-L1 slice data of various organs that contain sufficient tumor cell samples and sufficient PD-L1 expression areas;

[0089] The PD-L1 slice data cloud storage subsystem stores the PD-L1 slice data of various organs in the cloud and obtains the PD-L1 slice data sets of various organs.

[0090] The principles and effects of the above technical solution are as follows: the PD-L1 slice collection and processing subsystem includes: a PD-L1 slice collection high-resolution scanning subsystem, which collects PD-L1 slices of various organs for clinical pathology experiments, and obtains PD-L1 slice data of various organs through high-resolution digital scanning of PD-L1 slices; a slice data pre-identification and screening subsystem, which pre-identifies and screens PD-L1 slice data of various organs, and obtains PD-L1 slice data of various organs containing sufficient tumor cell samples and sufficient PD-L1 expression areas; a PD-L1 slice data cloud storage subsystem, which performs cloud data storage for PD-L1 slice data of various organs, and obtains PD-L1 slice data sets of various organs; it can expand the scope of computer-assisted methods in immunohistochemical pathology slice image analysis.

[0091] In one embodiment, the tumor cell mask annotation subsystem includes:

[0092] The PD-L1 slice annotation expert model subsystem uses the PD-L1 slice annotation expert model to input the cell morphological characteristics of the PD-L1 slices and the PD-L1 staining results based on the PD-L1 slice datasets of various organs, and outputs the location of the tumor cells to be annotated and the distribution area of ​​the tumor cells to be annotated;

[0093] The subsystem for setting the number of labeled cells sets the minimum number of labeled cells for each slice according to the location of the tumor cells to be labeled and the distribution area of ​​the tumor cells to be labeled; the minimum number of labeled cells for each slice is 5000;

[0094] The mask automatic annotation box selection subsystem automatically annotates each tumor cell annotation pixel-level mask through the pixel-level mask automatic annotation module; automatically generates an annotation bounding box through the boundary marking program module; the boundary marking program module first generates a minimum bounding box according to the center of any tumor cell annotation pixel-level mask of the pixel-level mask automatic annotation module; expands layer by layer from the inside to the outside with the minimum bounding box as the center until the bounding box covers any tumor cell annotation pixel-level mask to generate an annotation bounding box; in the annotation process, the bounding box selection is annotated with the valid area label for the area to be annotated, and then all tumor cells are annotated without omission; the bounding box selection is annotated with the ignore area label for the undetermined area; and the tumor cell area annotation data set is obtained.

[0095] The principle and effect of the above technical solution are as follows: the tumor cell mask annotation subsystem includes: the PD-L1 slice annotation expert model subsystem, which inputs the cell morphological characteristics and PD-L1 staining results of the PD-L1 slices according to the PD-L1 slice data sets of various organs through the PD-L1 slice annotation expert model, and outputs the location of the tumor cells to be annotated and the distribution area of ​​the tumor cells to be annotated; the annotated cell number setting subsystem, which sets the minimum number of annotated cells for each slice according to the location of the tumor cells to be annotated and the distribution area of ​​the tumor cells to be annotated; the minimum number of annotated cells for each slice is 5000; the mask automatic annotation box selection subsystem, which automatically annotates each tumor cell annotated pixel-level mask through the pixel-level mask automatic annotation module ; Automatically generate a labeled bounding box through a boundary marking program module; The boundary marking program module first generates a minimum bounding box according to the center of any tumor cell labeled pixel-level mask of the pixel-level mask automatic labeling module; Expand layer by layer from the inside to the outside with the minimum bounding box as the center until the bounding box covers any tumor cell labeled pixel-level mask to generate a labeled bounding box; In the labeling process, the valid area label is used to select the bounding box for the labeled area, and then all tumor cells are labeled without omission; The ignored area label is used to select the bounding box for the undetermined area; Obtain a tumor cell area labeling data set; The deep learning model can be used to enhance image analysis and further improve efficiency and accuracy in various medical image recognition and segmentation tasks.

[0096] In one embodiment, the YOLO v8 model creation training subsystem includes:

[0097] The labeled region preprocessing subsystem preprocesses all the locations marked as valid regions according to the tumor cell region labeling dataset and divides the training image blocks; assigns 1 pixel labels to all tumor regions in the training image blocks and gives boundaries; assigns 255 pixel labels to pixels in the training image blocks that exceed the valid region or are marked as ignored; assigns 0 pixel labels to all other pixels in the training image blocks; during the training process, removes the loss function contributed by all 255 pixels;

[0098] The single neural network YOLO v8 model subsystem uses a single neural network to make predictions directly on the entire image and output the target prediction bounding box and target prediction class probability, while outputting pixel-level segmentation results to create a YOLO v8 deep learning model;

[0099] The tumor annotation data model training subsystem uses the tumor cell region annotation dataset as training samples to train the YOLO v8 deep learning model; optimizes the model parameters through the back propagation algorithm; uses the cross entropy loss and bounding box regression loss to measure the difference between the model prediction results and the labels, and gradually adjusts the network weights to output the YOLO v8 deep learning model prediction results.

[0100] The principle and effect of the above technical solution are as follows: the YOLO v8 model creation training subsystem includes: the annotation area preprocessing subsystem, which preprocesses all positions marked as valid areas according to the tumor cell area annotation data set, and divides the training image blocks; assigns 1 pixel labels to all tumor areas in the training image blocks and gives boundaries; assigns 255 pixel labels to the pixels in the training image blocks that exceed the valid area or are marked as ignored; assigns 0 pixel labels to all other pixels in the training image blocks; during the training process, removes the loss function contributed by all 255 pixels; the single neural network YOLO v8 model subsystem, which uses a single neural network to directly predict on the entire image, and outputs the target prediction bounding box and target prediction category probability, and outputs the pixel-level segmentation results to create the YOLO v8 deep learning model; the tumor annotation data model training subsystem, which uses the tumor cell area annotation data set as a training sample to train the YOLO v8 deep learning model; optimizes the model parameters through the back propagation algorithm; uses the cross entropy loss and bounding box regression loss to measure the difference between the model prediction result and the label, and gradually adjusts the network weights to output the YOLO v8 deep learning model prediction results; create a training and optimization target detection model for tumor cell recognition, and output more accurate prediction results.

[0101] In one embodiment, the prediction result evaluation tumor recognition detection subsystem includes:

[0102] The model prediction comparison calculation subsystem compares the prediction results of the YOLO v8 deep learning model with the test dataset labels through an independent test dataset of untrained PD-L1 images, calculates the model prediction accuracy, model prediction recall, and model prediction F1 score, and obtains model prediction evaluation indicator information;

[0103] The training target evaluation determination subsystem determines whether the model prediction evaluation index information meets the training target evaluation index; when the model prediction evaluation index information does not meet the training target evaluation index, the YOLO v8 deep learning model is retrained until the model prediction evaluation index information meets the training target evaluation index; when the model prediction evaluation index information meets the training target evaluation index, the training is completed to obtain the PD-L1 image target evaluation YOLO v8 deep learning model;

[0104] The PD-L1 immunohistochemistry analysis image recognition subsystem uses the PD-L1 image target evaluation YOLO v8 deep learning model to perform actual PD-L1 immunohistochemistry pathology slice image recognition, and performs tumor cell counting, morphological feature extraction, and PD-L1 expression level analysis on the recognition results to achieve accurate positioning of tumor cells in PD-L1 immunohistochemistry pathology slices and pixel-level region recognition and detection.

[0105] The principle and effect of the above technical solution are: using the prediction results to evaluate tumor recognition detection, the model prediction comparison calculation subsystem, through the untrained PD-L1 image independent test data set, compares the YOLO v8 deep learning model prediction results with the test data set labels, calculates the model prediction accuracy, model prediction recall and model prediction F1 score, and obtains the model prediction evaluation index information; the training target evaluation determination subsystem determines whether the model prediction evaluation index information meets the training target evaluation index; when the model prediction evaluation index information does not meet the training target evaluation index, the YOLO v8 deep learning model is retrained until the model prediction evaluation index information meets the training target evaluation index; when the model prediction evaluation index information meets the training target evaluation index, the training is completed to obtain the PD-L1 image target evaluation YOLO v8 deep learning model; the PD-L1 immunohistochemistry analysis image recognition subsystem, through the PD-L1 image target evaluation YOLO v8 deep learning model, performs actual PD-L1 immunohistochemical pathology slice image recognition, counts tumor cells, extracts morphological features, and analyzes PD-L1 expression levels on the recognition results, and achieves accurate positioning and pixel-level region recognition and detection of tumor cells in PD-L1 immunohistochemical pathology slices; it can increase target positioning accuracy and recognition detection to enhance model performance.

[0106] Although the embodiments of the present invention have been disclosed as above, they are not limited to the applications listed in the specification and the implementation methods. They can be fully applied to various fields suitable for the present invention. For those familiar with the art, additional modifications can be easily implemented. Therefore, without departing from the general concept defined by the claims and the scope of equivalents, the present invention is not limited to the specific details and the illustrations shown and described herein.

Claims

1. A PD-L1 immunohistochemical section tumor recognition detection method, characterized in that: include: S100 collects PD-L1 slices of various organs for clinical pathology experiments, performs high-resolution scanning, pre-identification screening and cloud data storage, and obtains PD-L1 slice data sets of various organs; S200, based on the PD-L1 slice datasets of various organs, automatically annotates each tumor cell with a pixel-level mask, automatically generates annotated bounding boxes, and obtains tumor cell region annotated datasets; S300, based on the tumor cell region annotation dataset, preprocesses all locations annotated as valid regions, creates a YOLO v8 deep learning model and performs training to gradually adjust the network weights, and outputs the prediction results of the YOLO v8 deep learning model; S400, analyzes and calculates whether the prediction evaluation index information of the YOLO v8 deep learning model meets the training target evaluation index, completes the model training, and performs accurate positioning and pixel-level region recognition detection of tumor cells in PD-L1 immunohistochemistry pathology sections.

2. A PD-L1 immunohistochemical section tumor recognition detection method according to claim 1, characterized in that: S100 includes: S101, collect PD-L1 slices of various organs for clinical pathology experiments, and obtain PD-L1 slice data of various organs through high-resolution digital scanning of PD-L1 slices; S102, pre-identify and screen the PD-L1 slice data of various organs to obtain the PD-L1 slice data of various organs containing sufficient tumor cell samples and sufficient PD-L1 expression areas; S103, storing the PD-L1 slice data of various organs in the cloud, and obtaining the PD-L1 slice data sets of various organs.

3. A PD-L1 immunohistochemical section tumor recognition detection method according to claim 1, characterized in that: S200 includes: S201, based on the PD-L1 slice datasets of various organs, the PD-L1 slice annotation expert model is used to input the cell morphological characteristics of the PD-L1 slices and the PD-L1 staining results, and output the location of the tumor cells to be annotated and the distribution area of ​​the tumor cells to be annotated; S202, setting a minimum number of labeled cells for each slice according to the location of the tumor cells to be labeled and the distribution area of ​​the tumor cells to be labeled; S203, automatically annotate each tumor cell annotated pixel-level mask through the pixel-level mask automatic annotation module; automatically generate an annotation boundary box through the boundary marking program module; the boundary marking program module first generates a minimum bounding box according to the center of any tumor cell annotated pixel-level mask of the pixel-level mask automatic annotation module; expand layer by layer from the inside to the outside with the minimum bounding box as the center until the bounding box covers any tumor cell annotated pixel-level mask, and generate an annotation boundary box; in the annotation process, select the boundary box to be annotated using the valid area label, and then annotate all tumor cells in it without omission; select the boundary box for the undetermined area using the ignore area label; obtain the tumor cell area annotation data set.

4. A PD-L1 immunohistochemical section tumor recognition detection method according to claim 1, characterized in that: S300 includes: S301, based on the tumor cell region labeling dataset, pre-process all the locations marked as valid regions and divide the training image blocks; assign 1 pixel labels to all tumor regions in the training image blocks and give boundaries; assign 255 pixel labels to the pixels in the training image blocks that exceed the valid region or are marked as ignored; assign 0 pixel labels to all other pixels in the training image blocks; during the training process, remove the loss function contributed by all 255 pixels; S302, using a single neural network to directly predict the entire image, and output the target prediction bounding box and target prediction category probability, and output the pixel-level segmentation result to create a YOLO v8 deep learning model; S303, using the tumor cell region annotation dataset as a training sample, train the YOLO v8 deep learning model; optimize the model parameters through the back propagation algorithm; use the cross entropy loss and bounding box regression loss to measure the difference between the model prediction results and the labels, and gradually adjust the network weights to output the YOLO v8 deep learning model prediction results.

5. A PD-L1 immunohistochemical section tumor recognition detection method according to claim 1, characterized in that: S400 includes: S401, using an untrained PD-L1 image independent test dataset, comparing the prediction results of the YOLO v8 deep learning model with the test dataset labels, calculating the model prediction accuracy, model prediction recall, and model prediction F1 score, and obtaining model prediction evaluation indicator information; S402, determining whether the model prediction evaluation indicator information meets the training target evaluation indicator; when the model prediction evaluation indicator information does not meet the training target evaluation indicator, retraining the YOLO v8 deep learning model until the model prediction evaluation indicator information meets the training target evaluation indicator; when the model prediction evaluation indicator information meets the training target evaluation indicator, completing the training to obtain the PD-L1 image target evaluation YOLO v8 deep learning model; S403, through the PD-L1 image target evaluation YOLO v8 deep learning model, actual PD-L1 immunohistochemical pathology section image recognition is performed, and tumor cell counting, morphological feature extraction and PD-L1 expression level analysis are performed on the recognition results to achieve accurate positioning and pixel-level region recognition and detection of tumor cells in PD-L1 immunohistochemical pathology sections.

6. A PD-L1 immunohistochemical section tumor recognition detection system, characterized in that: include: The PD-L1 slice collection and processing subsystem collects PD-L1 slices of various organs for clinical pathology experiments, performs high-resolution scanning, pre-identification screening and cloud data storage, and obtains PD-L1 slice data sets of various organs; The tumor cell mask annotation subsystem automatically annotates the pixel-level mask of each tumor cell based on the PD-L1 slice datasets of various organs, automatically generates annotated bounding boxes, and obtains the tumor cell region annotation dataset; The YOLO v8 model creates a training subsystem, pre-processes all locations marked as valid areas based on the tumor cell region annotation dataset, creates a YOLO v8 deep learning model, performs training, gradually adjusts the network weights, and outputs the prediction results of the YOLOv8 deep learning model; The prediction results evaluate the tumor recognition and detection subsystem, analyze and calculate whether the prediction evaluation index information of the YOLO v8 deep learning model meets the training target evaluation index, and complete the model training to accurately locate the tumor cells in the PD-L1 immunohistochemistry pathology sections and identify the pixel-level area.

7. A PD-L1 immunohistochemical section tumor recognition detection system according to claim 6, characterized in that: The PD-L1 slice collection and processing subsystem includes: The PD-L1 slice acquisition high-resolution scanning subsystem collects PD-L1 slices of various organs for clinical pathology experiments, and obtains PD-L1 slice data of various organs through high-resolution digital scanning of PD-L1 slices; The slice data pre-identification and screening subsystem pre-identifies and screens the PD-L1 slice data of various organs, and obtains PD-L1 slice data of various organs that contain sufficient tumor cell samples and sufficient PD-L1 expression areas; The PD-L1 slice data cloud storage subsystem stores the PD-L1 slice data of various organs in the cloud and obtains the PD-L1 slice data sets of various organs.

8. A PD-L1 immunohistochemical section tumor recognition detection system according to claim 6, characterized in that: The tumor cell mask annotation subsystem includes: The PD-L1 slice annotation expert model subsystem uses the PD-L1 slice annotation expert model to input the cell morphological characteristics of the PD-L1 slices and the PD-L1 staining results based on the PD-L1 slice datasets of various organs, and outputs the location of the tumor cells to be annotated and the distribution area of ​​the tumor cells to be annotated; The subsystem for setting the number of labeled cells sets the minimum number of labeled cells for each slice according to the location of the tumor cells to be labeled and the distribution area of ​​the tumor cells to be labeled; The mask automatic annotation box selection subsystem automatically annotates each tumor cell annotation pixel-level mask through the pixel-level mask automatic annotation module; automatically generates an annotation bounding box through the boundary marking program module; the boundary marking program module first generates a minimum bounding box according to the center of any tumor cell annotation pixel-level mask of the pixel-level mask automatic annotation module; expands layer by layer from the inside to the outside with the minimum bounding box as the center until the bounding box covers any tumor cell annotation pixel-level mask to generate an annotation bounding box; in the annotation process, the bounding box selection is annotated with the valid area label for the area to be annotated, and then all tumor cells are annotated without omission; the bounding box selection is annotated with the ignore area label for the undetermined area; and the tumor cell area annotation data set is obtained.

9. A PD-L1 immunohistochemical section tumor recognition detection system according to claim 6, characterized in that: The YOLO v8 model creation and training subsystem includes: The labeled region preprocessing subsystem preprocesses all the locations marked as valid regions according to the tumor cell region labeling dataset and divides the training image blocks; assigns 1 pixel labels to all tumor regions in the training image blocks and gives boundaries; assigns 255 pixel labels to pixels in the training image blocks that exceed the valid region or are marked as ignored; assigns 0 pixel labels to all other pixels in the training image blocks; during the training process, removes the loss function contributed by all 255 pixels; The single neural network YOLO v8 model subsystem uses a single neural network to make predictions directly on the entire image and output the target prediction bounding box and target prediction class probability, while outputting pixel-level segmentation results to create a YOLO v8 deep learning model; The tumor annotation data model training subsystem uses the tumor cell region annotation dataset as training samples to train the YOLOv8 deep learning model; optimizes the model parameters through the back propagation algorithm; uses the cross entropy loss and bounding box regression loss to measure the difference between the model prediction results and the labels, and gradually adjusts the network weights to output the YOLO v8 deep learning model prediction results.

10. A PD-L1 immunohistochemical section tumor recognition detection system according to claim 6, characterized in that: The tumor recognition detection subsystem for predictive outcome evaluation includes: The model prediction comparison calculation subsystem compares the prediction results of the YOLO v8 deep learning model with the test dataset labels through an independent test dataset of untrained PD-L1 images, calculates the model prediction accuracy, model prediction recall, and model prediction F1 score, and obtains model prediction evaluation indicator information; The training target evaluation determination subsystem determines whether the model prediction evaluation index information meets the training target evaluation index; when the model prediction evaluation index information does not meet the training target evaluation index, the YOLO v8 deep learning model is retrained until the model prediction evaluation index information meets the training target evaluation index; when the model prediction evaluation index information meets the training target evaluation index, the training is completed to obtain the PD-L1 image target evaluation YOLO v8 deep learning model; The PD-L1 immunohistochemistry analysis image recognition subsystem uses the PD-L1 image target evaluation YOLO v8 deep learning model to perform actual PD-L1 immunohistochemistry pathology slice image recognition, and performs tumor cell counting, morphological feature extraction, and PD-L1 expression level analysis on the recognition results to achieve accurate positioning of tumor cells in PD-L1 immunohistochemistry pathology slices and pixel-level region recognition and detection.

Images