A lymphoma histological grading method, system, computer device and storage medium

Abstract

The application belongs to the technical field of pathological grading, and particularly relates to a lymphoma histological grading method, which comprises the following steps: screening a plurality of regions of interest rich in diagnostic information in a lymphoma pathological section image; inputting the regions of interest into a trained Hover-Net nuclear segmentation model to output a cell separation image; the cell separation image can mark the regions of different cell types, the positions of each cell and the classification information thereof; inputting the cell separation image into a classifier to output a histological grading result and the number of each cell type. The application improves the diagnostic accuracy and efficiency of follicular lymphoma pathological grading by efficiently recognizing lymphoma pathological section images of different pathological grades through a Hover-Net nuclear segmentation model.

Description

Technical Field

[0001] This invention belongs to the field of pathological grading technology, specifically relating to a method, system, computer equipment, and storage medium for histological grading of lymphoma. Background Technology

[0002] Follicular lymphoma (FL) is a type of non-Hodgkin's B-cell lymphoma. In recent years, the incidence of lymphoma has been rising, and it is now among the top ten malignant tumors. The diagnosis of FL relies primarily on histopathological examination. Pathologists determine the histological grade of FL by counting centroblasts in at least 10 high-power fields (HPF). Based on the number of centroblasts in each HPF, FL is classified into three histological grades: Grade 1: ≤5 centroblasts per field; Grade 2: 6-15 centroblasts per field; Grade 3: >15 centroblasts per field. Within Grade 3, if a small number of centroblasts are observed, it is classified as 3a; if centroblasts are completely absent and there are a large number of centroblasts, it is classified as 3b. Grades 1 and 2 are considered low-grade, while Grade 3 is considered high-grade. This histological grading plays a crucial role in guiding patient treatment plans and prognosis.

[0003] According to existing literature, the transformation rate of high-grade FL is significantly increased compared to low-grade FL, suggesting that higher pathological grades may be a predictive indicator of histologic transformation in FL. Specifically, higher FL grades may indicate increased invasiveness, making them more prone to transformation. In particular, grade 3b FL shows high similarity to diffuse large B-cell lymphoma in terms of gene expression profiles, further supporting the correlation between high-grade FL and histologic transformation (HT). Therefore, accurate pathological grading is crucial for guiding treatment decisions in clinical practice.

[0004] The pathological grading of FL (Fluorescent Leaf) primarily relies on the number of centroblasts observed under high power fields. Pathologists manually count the number of centroblasts in 10 high-power fields (2 mm²) to determine the pathological grade. However, the current histological grading of FL depends on the pathologist's microscopic counting and interpretation, which is highly subjective, inconsistent, and inefficient.

[0005] In traditional medical practice, histopathological grading of tissue sections (FL) is primarily based on detailed observation and analysis of tissue sections by pathologists using microscopes. However, the effectiveness of this method is limited by the magnification and field of view of the microscope, typically requiring physicians to invest a significant amount of time in carefully analyzing and understanding subtle changes on the sections. Traditional tissue section analysis methods rely heavily on the physician's professional knowledge and clinical experience, thus introducing a degree of subjective judgment. Therefore, even experienced pathologists may arrive at different diagnoses for the same tissue section, highlighting the potential limitations of this method in terms of consistency and objectivity.

[0006] With the rapid development of advanced imaging systems and WSI scanner technology, the field of pathology is undergoing a major transformation. Traditional microscopic slide reading methods are gradually being replaced by WSI-based digital diagnosis. In this technological revolution, computer-aided analysis has shifted from primarily relying on qualitative judgment to more precise quantitative analysis. Utilizing deep learning (DL) models to efficiently integrate and analyze large amounts of data not only improves the efficiency of pathological diagnosis but also makes the diagnostic process more objective and standardized, significantly enhancing the consistency of results. This method effectively reduces subjective differences in judgment among different pathologists during the diagnostic process. Currently, artificial intelligence has been applied to differential diagnosis, subtype identification, grading, and prognostic assessment. Using AI methods for digital pathology processing can rival the diagnostic accuracy of experienced pathologists and meet users' higher performance requirements. These advancements not only represent significant progress in medical technology but also provide new possibilities for future pathological diagnosis.

[0007] In the field of pathology research, artificial intelligence (AI) technology, especially deep learning algorithms, has become an important tool for improving diagnostic accuracy and efficiency. Currently, DL models have been successfully applied to various aspects of tumor diagnosis, including automated diagnosis, quantitative image evaluation, and detailed feature analysis, demonstrating the potential of AI technology in improving the quality and depth of tumor diagnosis and treatment. However, due to the diversity of lymphoma classifications and the complexity of diagnosis, the application of AI in lymphoma pathology diagnosis is still relatively limited. Summary of the Invention

[0008] The purpose of this invention is to provide a method, system, computer device, and storage medium for the histological grading of lymphoma, which solves the potential limitations in terms of consistency and objectivity of relying on manual judgment.

[0009] This invention is achieved through the following technical solution: This invention discloses a histological grading method for lymphoma, comprising the following steps: S1. Filter multiple regions of interest rich in diagnostic information in lymphoma pathological slide images; S2. Input the region of interest into the trained Hover-NET kernel segmentation model and output the cell segmentation image; Among them, the cell segmentation image can identify regions of different cell types, the location of each cell and its classification information; S3. Input the cell segmentation image into the classifier, and output the histological grading results and the number of each cell type.

[0010] Furthermore, in S1, the tool used to screen multiple regions of interest rich in diagnostic information is the OpenHI annotation system.

[0011] Furthermore, in S2, the training process of the Hover-NET kernel segmentation model is as follows: 2.1 Obtain pathological slide images of clinically diagnosed grade 3b lymphoma with typical morphology as the original image set of central blast cells; We acquired pathological slide images of clinically diagnosed low-grade lymphoma with typical morphology, and some pathological slide images of low-grade lymphoma with lighter staining, to form the original image set of the central cells. The original dataset consists of the original image set of the central mother cell and the original image set of the central cell. 2.2. The original dataset is labeled to identify the effective tumor region and the interfering cell region, thus obtaining the labeled dataset; 2.3. Cropping the effective tumor region of each image in the labeled dataset to obtain multiple image patches; 2.4 Image patches obtained from pathological slides of 2b-level lymphoma constitute the central blast cell dataset, and image patches obtained from pathological slides of low-level lymphoma constitute the central cell dataset. These two datasets are divided into training set, validation set, and test set. The Hover-NET kernel segmentation model is trained using the training set until the validation set is qualified and the test set meets the requirements. Then the Hover-NET kernel segmentation model is considered to have passed the training.

[0012] Furthermore, in section 2.2, the tool used to annotate the original dataset is the OpenHI annotation system.

[0013] Furthermore, in S3, the classifier can identify three types of cell image patches, which are divided into central blast cells, central cells, and other cells. The preset threshold in the classifier is the baseline value of the number of central mother cells; After the classifier completes the counting of central blast cells, it compares the count with a preset threshold. If the number of central blast cells exceeds the threshold, the image is classified as high-grade lymphoma tissue; otherwise, the image is classified as low-grade lymphoma tissue.

[0014] Furthermore, when the image is classified as high-grade lymphoma tissue, the output is the number 1; When the image is classified as low-grade lymphoma tissue, the output number is 0.

[0015] This invention also discloses a lymphoma histological grading system, comprising: The filtering module is used to filter multiple regions of interest rich in diagnostic information in lymphoma pathology slide images; The kernel segmentation module is used to input the region of interest into the trained Hover-NET kernel segmentation model and output a cell segmentation image; Among them, the cell segmentation image can identify regions of different cell types, the location of each cell and its classification information; The output module is used to input cell segmentation images into the classifier and output histological grading results and the number of each cell type.

[0016] The present invention also discloses a computer device, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to implement the steps of the lymphoma histological grading method.

[0017] The present invention also discloses a computer-readable storage medium storing a computer program that, when executed by a processor, implements the steps of the lymphoma histological grading method.

[0018] The present invention also discloses a computer program product, including a computer program / instructions that, when executed by a processor, implement the steps of the lymphoma histological grading method.

[0019] Compared with the prior art, the present invention has the following beneficial technical effects: This invention discloses a method for histological grading of lymphoma. First, a region of interest (ROI) is selected and input into the Hover-NET nuclear segmentation model. The model outputs a cell segmentation image, which clearly identifies regions of different cell types, the location of each cell, and its classification information. However, since there is no direct output, a classifier is also designed. The cell segmentation image is input into the classifier, which then outputs the histological grading results and the number of each cell type. This invention aims to improve the diagnostic accuracy and efficiency of pathological grading of follicular lymphoma by using the Hover-NET nuclear segmentation model to efficiently identify lymphoma pathological slides at different pathological grades. The development of this model provides pathologists with an accurate classification basis, aiming to reduce their workload and significantly decrease the risk of misdiagnosis and missed diagnosis due to subjective judgment, thereby improving the standardization of follicular lymphoma grading.

[0020] Unlike previous patch studies, this study focuses on cellular-level identification and uses the Hover-NET nuclear segmentation model; By applying this technology to the pathological grading of lymphoma, a new path has been explored for the application of machine learning models in medical image analysis.

[0021] Furthermore, when constructing the dataset, the dataset for the central cells used pathological slide images of low-grade lymphoma, while the original image set for the central blast cells used 3b high-grade follicular lymphoma. This was because cells in 3a grade FL are generally more mixed, so they were not selected.

[0022] When selecting effective tumor regions, the regions of interfering cells are also marked to facilitate subsequent training of the Hover-NET kernel segmentation model.

[0023] Furthermore, the classifier can identify the three cell counts, but when making a grading decision, it only focuses on the number of central blast cells. When the number of central blast cells exceeds a threshold, the image is classified as high-grade lymphoma tissue; otherwise, it is classified as low-grade lymphoma tissue. Outputting the three cell counts is to allow doctors to better understand the specific situation of the lymphoma tissue. Attached Figure Description

[0024] Figure 1 This is a flowchart of a lymphoma histological grading method according to the present invention; Figure 2 The training process of the Hover-NET kernel segmentation model of this invention; Figure 3 The results of ROI region annotation for grade 3b follicular lymphoma; Figure 4The annotation results are for tumor follicles; Figure a is a schematic diagram of a digital section of a single FL case; Figure b is a schematic diagram after the annotator marked all tumor follicles with annotation lines. Figure 5 The results are shown in Figure 1; Figure 2 shows an enlarged image of one of the tumor follicles; Figure 3 shows the cropping process of this enlarged tumor follicle, which is cropped into 512x512 pixel blocks for training. Figure 6 This is a schematic diagram of the image changes in the low-grade lymphoma identification process, with green representing central cells; where A is the input low-grade lymphoma pathological slide image; B is a typical region; C is a magnified view of a typical region in B; and D is the output cell segmentation image. Figure 7 This is a schematic diagram of the image changes in the high-grade (3a) FL identification process. Red represents the central blast cells, blue represents non-tumor cells, and green represents central cells. Among them, A is the input high-grade lymphoma pathological slide image; B is a typical region; C is a magnified view of a typical region in B; and D is the output cell segmentation image. Figure 8 The diagram shows the image changes during the high-grade (3b) lymphoma identification process. Red represents the central blast cells and blue represents non-tumor cells. Among them, A is the input high-grade lymphoma pathological slide image; B is a typical region; C is a magnified view of a typical region in B; and D is the output cell segmentation image. Figure 9 The performance curves of the three datasets in the test for block-level binary classification discrimination are shown. Figure 10 The output images for the classifier are shown in Figure a, where Figure a is the cell segmentation image of high-level FL and Figure b is the cell segmentation image of low-level FL. Detailed Implementation

[0025] To make the objectives, technical solutions, and advantages of the present invention clearer, the following detailed description is provided in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative of the invention and are not intended to limit the invention; that is, the described embodiments are only a part of the embodiments of the present invention, and not all of them.

[0026] The components described and illustrated in the accompanying drawings and embodiments of this invention can be arranged and designed in various different configurations. Therefore, the detailed description of the embodiments of the invention provided in the following drawings is not intended to limit the scope of the claimed invention, but merely to illustrate one selected embodiment of the invention. All other embodiments obtained by those skilled in the art based on the accompanying drawings and embodiments of this invention without inventive effort are within the scope of protection of this invention.

[0027] It should be noted that the terms “comprising,” “including,” or any other variations are intended to cover non-exclusive inclusion, such that a process, element, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to the process, element, method, article, or apparatus.

[0028] like Figure 1 As shown, this invention discloses a method for histological grading of lymphoma, comprising the following steps: S1. Filter multiple regions of interest rich in diagnostic information in lymphoma pathological slide images; S2. Input the region of interest into the trained Hover-NET kernel segmentation model and output the cell segmentation image; Among them, the cell segmentation image can identify regions of different cell types, the location of each cell and its classification information; S3. Input the cell segmentation image into a trained classifier and output the histological grading results and the number of each cell type.

[0029] During the training of the Hover-NET kernel segmentation model, this invention includes the following parts.

[0030] 1. Selection of clinical slides The criteria are as follows: (1) a clear diagnosis of follicular lymphoma or reactive hyperplasia of lymph nodes, with no other concurrent diagnoses; (2) completeness of the tissue sections, excluding biopsy samples with insufficient tissue volume; (3) good quality of the sections; and (4) relatively complete clinical and laboratory examination data in the electronic medical record. The microscopic diagnosis of all selected cases was confirmed by two senior pathologists, and cases that were difficult to diagnose were evaluated by a third professional lymphoma pathologist. In addition, the pathological histological grading criteria for FL were based on the "Chinese Guidelines for the Diagnosis and Treatment of Follicular Lymphoma".

[0031] 2. Acquisition of WSI (Whole Digital Slice Image) In this study, the Jiangfeng KF-PRO-005 scanner was used to scan all H&E-stained histopathological sections from the experimental samples. During the scanning process, the objective lens magnification was adjusted to 40x to ensure comprehensive coverage of all key stained areas on the sections. The scanning system utilized its high-precision algorithm to automatically identify and locate multiple clear focus points on the sections to ensure image clarity and accuracy. When encountering unsatisfactory focus points, the operator would intervene and manually adjust to supplement and improve the focus point settings. The scanner began to automatically focus and scan progressively, continuously capturing a series of partially overlapping 40x magnified square image areas. These individual image blocks were then precisely and synchronously stitched together by the scanner's built-in software to construct a complete and detailed morphological information image of the section, referred to as the focus map. After the scanning and stitching process was completed, the resulting focus map was stored in SVS format, forming a complete whole-slide image (WSI). Each WSI was strictly named and categorized according to its corresponding section number and diagnostic name to facilitate subsequent data retrieval and analysis.

[0032] 3. Construct three datasets with different compositions. Dataset 1: Obtain pathological slide images of clinically diagnosed grade 3b lymphomas with typical morphology as the original image set of central blast cells; The pathological slide images used for the training set were all of high quality and morphologically typical. Since cells in FL of grade 3a are generally mixed, they were not selected when choosing the training set.

[0033] Acquire pathological slide images of clinically diagnosed low-grade lymphomas with typical morphology as the original image set of the central cells; The original image set of the central mother cell and the original image set of the central cell constitute dataset 1.

[0034] Dataset 2: such as Figure 2 As shown, based on dataset 1, some lightly stained pathological slide images of low-grade lymphoma were added to the original image set of central cells as an expanded dataset.

[0035] Dataset 3: Based on Dataset 2, the training data has been color normalized.

[0036] The cell nucleus segmentation model was used to test the above three datasets. The specific accuracy of high / low grade lymphoma pathological slide image recognition is shown in Tables 1-3.

[0037] Table 1

[0038] The WHO classification criteria (150) are used to test the binary discrimination performance of variables, see [reference needed]. Figure 9 This indicates that dataset 2 has high prediction accuracy, which is a statistical method. The area under the receiver operating characteristic (ROC) curve (AUC) provides a more accurate performance evaluation metric. If the AUC value is between 0.5 and 0.7, the predictive performance of the cell nucleus segmentation model is considered poor; an AUC value of 0.7 to 0.8 indicates that the model has a moderate level of predictive ability; 0.8 to 0.9 indicates that the model has good predictive performance; and an AUC value of 0.9 or higher indicates that the model's predictive performance has reached an excellent level. This figure is to illustrate that the cell nucleus segmentation model of this invention has very good predictive ability, as detailed in Table 3.

[0039] Table 2

[0040] Table 3

[0041] The data analysis revealed that the Hover-NET kernel segmentation model could not recognize some lightly stained grade 1-2 FL pathological slide images. Therefore, this invention added some lightly stained grade 1-2 FL pathological slide images to the dataset. After learning from these slides, the Hover-NET kernel segmentation model improved its recognition performance for low-level FL.

[0042] When dataset 2 was modeled and analyzed using the number of centroblasts as the test variable, its optimal cutoff value was 159, which is very close to the cutoff value of 150 recommended by the WHO guidelines, further confirming the predictive accuracy of dataset 2.

[0043] Therefore, Dataset 2 was used as the dataset for training the cell nucleus segmentation model. A different approach was taken in the research design compared to most other studies: the slices were not uniformly re-stained. This decision was based on considerations of the model's generalization ability; it was believed that using raw slice samples from the real world, rather than artificially processed samples, would make the dataset of this invention more adaptable and generalizable.

[0044] 4. Image annotation The dataset is labeled to identify the effective tumor regions and interfering cell regions, resulting in a labeled dataset.

[0045] like Figure 3 As shown, after annotating one of the pathological slide images, the red area represents the target cells and the black area represents the interfering cells.

[0046] like Figure 4 Figure a shows a schematic diagram of a WSI case of FL. After the annotator marked all the neoplastic follicles with annotation lines, the result was as follows. Figure 4 The image shown in Figure b.

[0047] Specifically, all annotation work was completed using the OpenHI annotation system. This system supports multi-user collaborative or crowdsourced annotation, enabling multiple experts or teams to work together, significantly improving the quality and efficiency of annotation. The OpenHI framework possesses pixel-level precision annotation capabilities, which are crucial for histopathological image analysis requiring high accuracy. Furthermore, OpenHI supports standardized, semantically rich annotation, meaning that annotations are not merely image labels but also include detailed information about tissues, diseases, etc., providing significant value for research and education. This framework can effectively handle large histopathological image datasets and is suitable for large-scale research projects. Its flexible design allows for the annotation of entire slide images from different sources and tumor types.

[0048] According to diagnostic guidelines, the area of ​​each high-power field is set to 2 mm², while in the annotation system used in this invention, the size of an image patch is 256x256 pixels. After conversion, 10 high-power fields are equivalent to 40 regions of interest. Therefore, when performing external testing on each WSI case, 40 regions of interest are annotated to identify and count tumor cells, and based on this, perform histological grading to improve the accuracy and effectiveness of grading.

[0049] 5. Image cropping The labeled dataset is cropped and extracted to obtain several image patches. For example... Figure 2 As shown, the images were incorporated into the training set, validation set, and internal test set in an 8:1:1 ratio. Detailed information on the number of image patches in each group is provided in Table 4.

[0050] The kernel segmentation results of the model are interpreted, and the Hover-NET model is used to achieve three-class classification of cells at the image patch level, specifically including central mother cells (positive) – C3, central cells (negative) – C2, and other cells (unclassified) – C1. Therefore, this invention classifies image patches in the dataset into these three categories based on biological characteristics.

[0051] Table 4 Information on the number of graphic blocks in each group

[0052] This allocation is designed to support subsequent neural network training and internal testing. During internal testing, positive cells identified by the model are compared for consistency with manually labeled target cells. Only when the consistency rate reaches or exceeds 95% will the model proceed to external testing. The external test set primarily consists of cases already definitively diagnosed with follicular lymphoma.

[0053] like Figure 5 Figure a shows a labeled region of interest. Figure 5 The region of interest shown in Figure a is cropped to obtain the following result: Figure 5 The image blocks shown in Figure b.

[0054] 6. Cell nucleus segmentation model (Hover-NET) Cell nucleus segmentation is a fundamental task in digital pathology analysis for studying cell characterization. Hover-NET is an innovative convolutional neural network specifically designed for cell nucleus segmentation and classification in multi-histological images. It has demonstrated superior performance in deep learning model applications, particularly in medical image analysis and pathological image processing. This network not only plays a foundational role in H&E staining digital pathology analysis for studying cell characterization but also plays a central role in the precise segmentation and classification of cell nuclei. This refined analysis is crucial for understanding cellular heterogeneity in histological images, especially in identifying and classifying different cell types such as tumor cells and immune cells. This refined analysis helps pathologists gain a deeper understanding of cellular heterogeneity in histological images, thus playing a key role in cancer diagnosis and treatment.

[0055] Hover-NET's architecture integrates the feature extraction capabilities of pre-activated residual networks with network branches specifically designed for segmentation and classification tasks, achieving efficient multi-task learning. This model is particularly well-suited for processing digital slices, employing a "tile-based approach" that effectively addresses the challenges of large-scale WSI (Web Sample Interface), overcoming the difficulties of handling large and highly complex images. Furthermore, Hover-NET utilizes an innovative method, leveraging the distance information from cell nucleus pixels to the centroid to improve the segmentation accuracy of overlapping cell nuclei, and optimizes the learning process for different task objectives through a composite loss function (including regression loss for segmentation and cross-entropy loss for classification). Notably, Hover-NET can simultaneously perform cell nucleus segmentation and cell type classification, achieved through multi-task learning; that is, the model learns how to segment cell nuclei and identify their types at the same time.

[0056] In terms of performance evaluation, Hover-NET introduces panoptic quality (PQ), a comprehensive evaluation metric. Performance on multiple datasets shows that it significantly outperforms other advanced kernel segmentation methods. These technological innovations make Hover-NET a highly promising tool in the field of digital pathology, providing crucial support for oncology research.

[0057] 7. Training the Hover-NET kernel segmentation model The core feature of Hover-NET is its use of vertical and horizontal encoded information to address cell clustering issues, thereby achieving efficient nuclear segmentation. Nevertheless, overlapping nuclei in lymphoma slices still pose a significant challenge, as a large number of merged nuclei can affect the accuracy and sensitivity of nuclear size analysis.

[0058] This invention reveals that the segmentation performance of the model is not ideal when processing WSIs with high nuclear overlap or thick slices. Particularly in these complex WSIs, the Hover-NET nuclear segmentation model fails to accurately segment each cell nucleus. To address this issue, this invention provides a customized optimization of the Hover-NET nuclear segmentation model.

[0059] The optimization process involves precisely labeling each cell nucleus within three randomly selected regions of interest (ROIs) using human annotations, and then retraining the model with this labeled data to improve its kernel segmentation accuracy. This optimization significantly improves the model's performance, as evidenced by the improved Dice coefficient. The Dice coefficient is a widely used metric for evaluating the performance of AI-based cell segmentation algorithms. Before the optimization, the Hover-NET kernel segmentation model had a Dice coefficient of 0.78, which was increased to 0.98 after optimization. This significant improvement ensures accurate segmentation of cell nuclei within each ROI, providing a reliable foundation for subsequent kernel segmentation.

[0060] These classified images are then fed into a classifier to identify and differentiate between the three types of cell images, such as Figure 10 In the output image of figure a, if there are many central mother cells, it is identified as a high-level FL (Francrine Pathogen); low-level FLs are not as... Figure 10 If the number of cells in the center of the output image is large, it is identified as a low-level FL.

[0061] This invention introduces a threshold-based classification mechanism. The classifier employs a threshold-based discrimination method, where the number of cells (denoted as x) is determined by the cell information output by the cell nucleus segmentation model. The classifier counts the number of cells and compares the number of central blast cells with a preset threshold (cutoff, denoted as C). When the number of central blast cells exceeds the threshold C, i.e., x>C, the system automatically classifies the image as high-grade lymphoma tissue; otherwise, it classifies it as low-grade lymphoma tissue.

[0062] Specifically, when the image is classified as high-grade lymphoma tissue, the output number is 1; when the image is classified as low-grade lymphoma tissue, the output number is 0.

[0063] The features and performance of the present invention will be further described in detail below with reference to application examples.

[0064] Application Example 1 When a picture is selected as follows Figure 6 When examining the pathological slide image of low-grade lymphoma shown in Figure A, first select 40 regions of interest (ROIs) rich in diagnostic information from the lymphoma pathological slide image; each region is 256×256 pixels. Will as Figure 6 The regions of interest shown in Figures B and C are input into a trained Hover-NET kernel segmentation model, and the output is as follows: Figure 6 The cell segmentation image shown in Figure D; Among them, the cell segmentation image can identify regions of different cell types, the location of each cell and its classification information; The cell segmentation image is input into a trained classifier, which outputs the number of each cell type, including 1 centroblast, 26,556 centroblasts, and 3 non-tumor cells; the histological grading result is low grade.

[0065] Application Example 2 When a picture is selected as follows Figure 7 When examining the high-grade lymphoma pathological section image shown in Figure A, first select 40 typical regions from the lymphoma pathological section image, each region being 256×256 pixels; Typical regions (such as images B and C in Figure 7) are input into a trained Hover-NET kernel segmentation model, and the output is as follows: Figure 7 The cell segmentation image shown in Figure D; Among them, the cell segmentation image can identify regions of different cell types, the location of each cell and its classification information; The cell segmentation image was input into a trained classifier, which output the number of each cell type: 7183 centroblasts, 6973 centroblasts, and 667 non-tumor cells; the histological grading result was grade 3a.

[0066] Application Example 3 When a picture is selected as follows Figure 8 When examining the high-grade lymphoma pathological section image shown in Figure A, first select 40 typical regions from the lymphoma pathological section image, each region being 256×256 pixels; Typical areas (such as) Figure 8 The inputs (Figures B and C) are fed into the trained Hover-NET kernel segmentation model, and the output is as follows: Figure 8 The cell segmentation image shown in Figure D; Among them, the cell segmentation image can identify regions of different cell types, the location of each cell and its classification information; The cell segmentation image was input into a trained classifier, which output the number of each cell type: 14,528 centroblasts, 28 centroblasts, and 281 non-tumor cells; the histological grading result was grade 3b.

[0067] This invention also discloses a lymphoma histological grading system, comprising: The filtering module is used to filter multiple regions of interest rich in diagnostic information in lymphoma pathology slide images; The kernel segmentation module is used to input the region of interest into the trained Hover-NET kernel segmentation model and output a cell segmentation image; Among them, the cell segmentation image can identify regions of different cell types, the location of each cell and its classification information; The output module is used to input cell segmentation images into the classifier and output histological grading results and the number of each cell type.

[0068] The lymphoma histological grading method of the present invention can be implemented in a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, the present invention can be implemented as a computer program product on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0069] If the lymphoma histological grading method of the present invention is implemented as a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, all or part of the processes in the methods of the above embodiments of the present invention can also be implemented by a computer program instructing related hardware. The computer program can be stored in a computer-readable storage medium, and when executed by a processor, it can implement the steps of the various method embodiments described above. The computer program includes computer program code, which can be in the form of source code, object code, executable files, or certain intermediate forms. The computer-readable storage medium includes permanent and non-permanent, removable and non-removable media, and information storage can be implemented by any method or technology. Information can be computer-readable instructions, data structures, program modules, or other data. It should be noted that the content contained in the computer-readable medium can be appropriately added or subtracted according to the requirements of legislation and patent practice in the jurisdiction. For example, in some jurisdictions, according to legislation and patent practice, computer-readable media do not include electrical carrier signals and telecommunication signals. The computer storage medium can be any available medium or data storage device that a computer can access, including but not limited to magnetic storage (e.g., floppy disk, hard disk, magnetic tape, magneto-optical disk (MO)), optical storage (e.g., CD, DVD, BD, HVD), and semiconductor storage (e.g., ROM, EPROM, EEPROM, non-volatile memory (NAND FLASH), solid-state drive (SSD)).

[0070] In an exemplary embodiment, a computer device is also provided, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to implement the steps of the lymphoma histological grading method. The processor may be a central processing unit (CPU), or other general-purpose processors, digital signal processors (DSPs), application-specific integrated circuits (ASICs), field-programmable gate arrays (FPGAs), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc.

[0071] The present invention also discloses a computer program product, including a computer program / instructions that, when executed by a processor, implement the steps of the lymphoma histological grading method.

[0072] 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 it. Although the present invention has been described in detail with reference to the above embodiments, those skilled in the art should understand that modifications or equivalent substitutions can still be made to the specific implementation of the present invention. Any modifications or equivalent substitutions that do not depart from the spirit and scope of the present invention should be covered within the protection scope of the claims of the present invention.

Claims

1. A histological grading method for lymphoma, characterized in that, Includes the following steps: S1. Filter multiple regions of interest rich in diagnostic information in lymphoma pathological slide images; S2. Input the region of interest into the trained Hover-NET kernel segmentation model and output the cell segmentation image; Among them, the cell segmentation image can identify regions of different cell types, the location of each cell and its classification information; S3. Input the cell segmentation image into the classifier, and output the histological grading results and the number of each cell type; In S2, the training process of the Hover-NET kernel segmentation model is as follows: 2.1 Obtain pathological slide images of clinically diagnosed grade 3b lymphoma with typical morphology as the original image set of central blast cells; We acquired pathological slide images of clinically diagnosed low-grade lymphoma with typical morphology, and some pathological slide images of low-grade lymphoma with lighter staining, to form the original image set of the central cells. The original dataset consists of the original image set of the central mother cell and the original image set of the central cell. 2.

2. The original dataset is labeled to identify the effective tumor region and the interfering cell region, thus obtaining the labeled dataset; 2.

3. Cropping the effective tumor region of each image in the labeled dataset to obtain multiple image patches; 2.4 Image patches obtained from pathological slides of 2b-level lymphoma constitute the central blast cell dataset, and image patches obtained from pathological slides of low-level lymphoma constitute the central cell dataset. These two datasets are divided into training set, validation set, and test set. The Hover-NET kernel segmentation model is trained using the training set until it passes validation on the validation set and meets the requirements when tested on the test set. In this case, the Hover-NET kernel segmentation model is considered successfully trained.

2. The method for histological grading of lymphoma according to claim 1, characterized in that, In S1, the tool used to filter multiple regions of interest rich in diagnostic information is the OpenHI annotation system.

3. The method for histological grading of lymphoma according to claim 1, characterized in that, In section 2.2, the OpenHI annotation system was used to annotate the original dataset.

4. The method for histological grading of lymphoma according to claim 1, characterized in that, In S3, the classifier can identify three types of cell image patches, which are divided into central blast cells, central cells, and other cells. The preset threshold in the classifier is the baseline value of the number of central mother cells; After the classifier completes the count of central blast cells, it compares the count with a preset threshold. If the number of central blast cells exceeds the threshold, the image is classified as high-grade lymphoma tissue; otherwise, the image is classified as low-grade lymphoma tissue.

5. A method for histological grading of lymphoma according to claim 4, characterized in that, When the image is classified as high-grade lymphoma tissue, output the number 1; When the image is classified as low-grade lymphoma tissue, the output number is 0.

6. A lymphoma histological grading system implementing the lymphoma histological grading method as described in any one of claims 1 to 5, characterized in that, include: The filtering module is used to filter multiple regions of interest rich in diagnostic information in lymphoma pathology slide images; The kernel segmentation module is used to input the region of interest into the trained Hover-NET kernel segmentation model and output a cell segmentation image; Among them, the cell segmentation image can identify regions of different cell types, the location of each cell and its classification information; The output module is used to input cell segmentation images into the classifier and output histological grading results and the number of each cell type.

7. A computer 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 computer program, it implements the steps of the lymphoma histological grading method as described in any one of claims 1 to 5.

8. A computer-readable storage medium storing a computer program, characterized in that, When the computer program is executed by the processor, it implements the steps of the lymphoma histological grading method as described in any one of claims 1 to 5.

9. A computer program product comprising a computer program / instructions, characterized in that, When the computer program / instructions are executed by the processor, they implement the steps of the lymphoma histological grading method as described in any one of claims 1 to 5.

Images