Tissue imaging processing method, system, and computer-readable recording medium

Abstract

To relate to the field of image processing technology and provide a method and system for processing tissue images on whole-slide imaging.SOLUTION: A method includes: obtaining a target image; obtaining a dataset; enhancing the staining components of the dataset; using the enhanced dataset for federated learning to construct a tumor segmentation model, the dataset including several test images, and the test images being immunohistochemical images; inputting the target image into the tumor segmentation model to determine a tumor area within the target image; performing cell membrane staining detection on the tumor area; classifying the tumor area based on the integrity of the cell membrane and the level of staining; and grading based on the classification result.SELECTED DRAWING: Figure 1

Description

[Technical Field]

[0001] This disclosure relates to the field of image processing technology, and more particularly to methods and systems for processing tissue images in whole-slide imaging. [Background technology]

[0002] The rapid development of precision medicine and personalized medicine is driving the demand for accurate diagnostic tests. These tests are crucial for implementing new therapies such as targeted therapies. With existing technologies, processing target tissue images relies heavily on manual identification that depends on human experience to determine gene expression levels within the images. This method is generally inaccurate, making it an urgent task to improve the accuracy of image processing. [Overview of the project] [Problems that the invention aims to solve]

[0003] From this perspective, this disclosure provides a method and system for processing tissue images in whole-slide imaging, which aims to optimize the method and evaluate the level or expression of human epidermal growth factor receptor 2 (HER2) in target tissue images more accurately than the prior art. [Means for solving the problem]

[0004] In a first embodiment, the Disclosure provides a specific embodiment of a method for processing tissue images from whole slide imaging, the method comprising: (a) acquiring a target image; (b) acquiring a dataset, highlighting the staining components of the dataset; using the highlighted dataset for federated learning to construct a tumor segmentation model, the dataset including a test image, the test image being an immunohistochemical (IHC) image; (c) inputting the target image into the tumor segmentation model to determine the tumor region of the target image; (d) performing cell membrane staining detection on the tumor region, further classifying the tumor region based on the integrity of the cell membrane and the level of staining of the cell membrane, and grading based on the classification results.

[0005] In some embodiments, “acquiring a dataset and highlighting the staining components of the dataset” further includes acquiring the test image of the dataset, performing color separation on the test image, acquiring the color base and staining intensity of the pigments in the test image, randomly scaling and randomly translating the staining intensity of the pigments to highlight the staining intensity of the pigments, randomly rotating the color base of the pigments to acquire the enhanced color base of the pigments, and integrating the enhanced color base and enhanced staining intensity of the pigments to complete the highlighting of the staining components.

[0006] In some embodiments, "using the enhanced datasets for federated learning to construct the tumor segmentation model" further includes preprocessing several enhanced datasets and inputting the preprocessing results into a federated learning server to construct the tumor segmentation model. The preprocessing is used to reduce the amount of data input into the federated learning server.

[0007] In some embodiments, “preprocessing several enhanced datasets and inputting the preprocessing results into a federated learning server to construct the tumor segmentation model” further includes constructing a local model and a global model based on the learning rules of the federated learning server, training the local model on one or more enhanced datasets to obtain the parameters of the local model, inputting the parameters of the local model into the federated learning server, training the global model via a clustering pipeline, training the global model on several enhanced datasets, inputting the parameters of the trained global model into the federated learning server to construct the tumor segmentation model.

[0008] In some embodiments, (d) "performing cell membrane staining detection on the tumor region and classifying the tumor region" further includes (d1) detecting the nuclei of the tumor region to determine the location of the nuclei of the tumor region, (d2) performing cell membrane detection based on the location of the nuclei to determine the location of the cell membranes corresponding to the nuclei, (d3) classifying the cell membranes based on the staining level to obtain a pre-classification result, and (d4) classifying the tumor region based on the pre-classification result and the integrity of the cell membranes.

[0009] In some embodiments, (d2) "performing cell membrane detection based on the position of the nuclei and determining the position of the cell membrane corresponding to the nuclei" further includes expanding the segmentation mask of the nuclei based on the position of the nuclei and determining the position of the cell membrane corresponding to the nuclei.

[0010] (d3) "Classifying the cell membrane based on the staining level and obtaining a pre-classification result" further includes performing color separation on each pixel of the cell membrane, converting it to a representation in hematoxylin (H&E)-diaminobenzidine (DAB), classifying each pixel of the cell membrane according to a preset threshold based on the staining intensity of the diaminobenzidine channel, obtaining the classification level for each pixel, and obtaining the pre-classification result based on the classification level for each pixel.

[0011] In some embodiments, (d4) “classifying the tumor region based on the pre-classification results and the integrity of the cell membrane” further includes obtaining the integrity of the cell membrane using a skeletonization algorithm and refining the pre-classification results based on the integrity of the cell membrane in order to classify the tumor region.

[0012] In some embodiments, rating based on classification results involves evaluating the level of HER2 representation in the target image using a predetermined standard, such as the ASCO CAP 2018 guidelines.

[0013] In a second embodiment, the Disclosure also provides a specific embodiment of a tissue image processing system relating to whole slide imaging, the system comprising a memory, a processor, and a computer program stored in the memory, wherein the processor implements the tissue image processing method of the first embodiment by executing the computer program. .

[0014] In a third embodiment, the Disclosure also provides a specific embodiment of a computer program product, which includes a computer program / instruction, which, when executed by a processor, performs the tissue imaging processing method of the first embodiment. [Effects of the Invention]

[0015] Compared with the prior art, the invention of the present application has at least the following beneficial effects or unexpected results: 1. The expression of HER2 in the target image can be accurately evaluated. 2. The target images obtained from different facilities can be optimized, and cross-facility model learning can be realized without compromising the privacy of users.

Brief Description of the Drawings

[0016] [Figure 1] FIG. 1 is a schematic diagram of a full-loading slide tissue image processing method according to an embodiment of the present invention. [Figure 2] FIG. 2 is another schematic diagram of a full-loading slide tissue image processing method according to an embodiment of the present invention. [Figure 3] FIG. 3 is another schematic diagram of a full-loading slide tissue image processing method according to an embodiment of the present invention.

Modes for Carrying Out the Invention

[0017] One or more specific embodiments are shown by way of example in the accompanying drawings and are not limiting. The accompanying drawings are not to scale unless otherwise disclosed. The present disclosure should be understood by those of ordinary skill in the art in light of the following detailed description of the preferred specific embodiments and with reference to the accompanying drawings.

[0018] The above features, technical features, advantages, and their realizations of the present disclosure will be further described below in a clear and easy-to-understand manner in conjunction with examples of preferred specific embodiments.

[0019] To more clearly explain the technical solutions of the embodiments of this application, specific embodiments will be described below with reference to the accompanying drawings. The drawings described below represent only a portion of the embodiments of this application. Those skilled in the art can obtain other drawings and other embodiments based on these drawings without any ingenuity. Modifications and improvements made without departing from the spirit of this application are protected within the scope of this application.

[0020] To simplify the drawings, only the parts relevant to the corresponding embodiments are schematically shown. These do not represent the actual structure of the product. Furthermore, to make the drawings concise and easy to understand, some components having the same structure or function are only partially schematicly depicted. In reality, there may be more or fewer components having the same structure or function.

[0021] In this disclosure, unless otherwise expressly stated, ordinal numbers such as “first” and “second” are used to distinguish related objects and do not indicate relative importance or order. “Plural” includes two or more, as do other quantifiers. “ / ” is used to express a relationship between related objects and indicates an “or” relationship. “And / or” includes any combination relationship between related objects, such as “a and / or b,” and is used to describe a relationship between related objects. “a and / or b” includes “a alone,” “b alone,” or “a and b.” “One or more” or “at least one” of plural objects means any combination of any one or more objects, such as “one or more of a1, a2, and a3” or “at least one of a1, a2, and a3.” "One or more of a1, a2, and a3" or "at least one of a1, a2, and a3" includes "a1 alone", "a2 alone", "a3 alone", "a1 and a2", "a1 and a3", "a2 and a3", or "a1, a2, and a3".

[0022] Recent advances in precision oncology, such as immunotherapy and antibody-drug conjugates, offer significant potential for extending the survival time of cancer patients. To formulate appropriate treatment strategies, these advances often involve meticulous investigation of patient information, including high-throughput sequencing and companion diagnostics. Such tests greatly benefit from artificial intelligence (AI) and deep learning (DL) models that provide quantitative details of biomarker expression.

[0023] In the treatment of metastatic breast cancer, human epidermal growth factor receptor 2 (HER2) plays a crucial role in protein-targeted therapy. Trastuzumab (brand name: Herceptin), a HER2-targeted monoclonal antibody, has matured as a first-line treatment for HER2-positive patients, significantly reducing the risk of disease progression and death compared to chemotherapy alone. HER2 expression is assessed using immunohistochemical (IHC) staining, and in situ hybridization testing when the former is ambiguous. However, recent studies have shown that clinical interpretation of HER2 may be impaired by low consistency in pathologist judgments, and that interpathological agreement can be improved with the help of several artificial intelligence algorithms. The urgent need for AI in HER2 expression analysis is clear.

[0024] Several challenges remain in developing AI systems for HER2 expression assessment. Firstly, deep learning models integrated into such systems must be data-driven, but medical images are often scarce resources for a single institution. More importantly, patient privacy policies prohibit inter-institutional data sharing, making it impossible to train deep learning models using conventional methods. Fortunately, the newly emerging federative learning (FL) allows healthcare institutions to collaboratively develop deep learning models while maintaining patient privacy. Specifically, the FL strategy involves multiple institutions training models in parallel using their respective data, passing model parameters to each other during iterations. This approach has been shown to significantly improve the generalizability of various deep learning models for medical imaging.

[0025] Secondly, while FL applications have been successful in computed tomography and magnetic resonance imaging, the performance of DL models specifically for histological images may be hampered by inter-institutional imaging variations. Histological images, such as those requiring IHC staining for HER2 expression assessment, consist of color channels convolved by pixel-level intensities of hematoxylin (H&E) and diaminobenzidine (DAB). The color components of these stains are heavily influenced by laboratory conditions during sample preparation. Pathologists clearly observe variability in staining color, but DL models that strictly use color channels as input are largely undetectable. Consequently, trained models may underperform and be biased towards partial data sources.

[0026] Please refer to Figure 1, which shows a schematic diagram of a whole slide tissue imaging method provided in one embodiment of this disclosure. In one embodiment, as shown in Figure 1, a whole slide tissue imaging method for evaluating the expression level of HER2 in a target image includes the following steps:

[0027] S100 (Step a): Acquire target images / target tissue images, preferably whole-slide imaging (WSI) immunohistochemical imaging.

[0028] S200 (Step b): Acquire the dataset, highlight the staining components of the dataset, and use the highlighted dataset for federative learning to build a tumor segmentation model.

[0029] Of particular note is that the datasets obtained from different institutions are first subjected to a staining component augmentation (SC augmentation). The augmented datasets are then used to build a tumor segmentation model using federative learning techniques. Federative learning is a machine learning technique that specifically involves training an algorithm on multiple distributed edge devices or servers with local data samples. This approach differs significantly from traditional centralized machine learning methods that upload all local datasets to a single server, and from more classical distributed methods that generally assume local data samples are uniformly distributed. However, the federative learning presented here enables multiple participants to build a common, robust machine learning model without sharing data, thereby addressing critical issues such as data privacy, data security, data access rights, and heterogeneous data access. Therefore, using federative learning techniques allows for the training of reliable models on user-collected data while ensuring user data privacy. The dataset includes multiple test images, which are immunohistochemical images.

[0030] Specifically, using datasets obtained from National Taiwan University Hospital (NTUH) and the University of Warwick, we perform local staining component expansion, upload the enhanced dataset to a federated learning server, and construct a tumor segmentation model through federated learning operations. The tumor segmentation model can automatically detect the location and boundaries of tumors in medical images, segment tumor regions from normal regions, and quantify tumor size, volume, and features, helping physicians quickly and accurately identify tumors in patients.

[0031] S300 (Step c): The target image is input into the tumor segmentation model to determine the tumor region of the target image. Since the target image or target tissue image includes not only the tumor region but also normal regions, the purpose of this disclosure is to evaluate the expression level of human epidermal growth factor receptor 2 (HER2). Therefore, it is necessary to pre-determine the tumor region within the target image or target tissue image. Although this disclosure only reveals the use of a method for evaluating the expression level of HER2, those skilled in the art can understand the necessary technical means from this disclosure and further apply this method to evaluate the expression levels of other genes or proteins. Therefore, evaluating the expression levels of other genes in a target tissue image using the method disclosed herein should also be considered within the scope of this disclosure.

[0032] S400 (Step d): Cell membrane staining detection is performed on the tumor area, and the tumor area is classified based on the integrity and staining level of the cell membrane, and a rating is given based on the classification result. Specifically, the tumor area can be subdivided into negative, faint, weak, and strong based on the staining result. Cell membrane integrity is divided into complete and incomplete. Based on these combinations, seven classification results are obtained: negative, faint-incomplete, faint-complete, weak-incomplete, weak-complete, strong-incomplete, and strong-complete.

[0033] The whole-slide image processing method disclosed herein constructs a tumor segmentation model by expanding staining components, improving the model's compatibility with target images from different institutions and accurately presenting tumor regions. It then classifies HER2 levels in the target image by performing cell membrane staining detection on the tumor regions. This method accurately evaluates HER2 expression levels in target images and optimizes target images obtained from different institutions, enabling cross-institutional model learning without compromising user privacy.

[0034] Please refer to Figure 2, which shows another schematic diagram of a whole slide image processing method provided in one embodiment of the present disclosure. In one embodiment, the step of “acquiring a dataset and highlighting staining components” further includes the following steps:

[0035] S210: Retrieve test images from the dataset, perform color separation on the test images, and obtain the color base and staining intensity of the pigments.

[0036] Generally, color expansion is an enhancement technique that broadens the data distribution range during model training by simulating various histological images based on the source image. Due to differences in staining intensity and pigment components between institutions, color expansion is necessary to minimize these differences. Ideally, pixels in an image are stained by only two pigments, such as H&E and DAB in IHC images. Color expansion techniques use these two pigments to produce enhanced images that better fit the distribution of histological data compared to general RGB image enhancement. Specifically, a test image is first acquired from the dataset, and color decomposition is performed on the test image. For a given input IHC image, color expansion calibrates its H&E and DAB channels by color decomposition. Image X is decomposed into a staining color base S and corresponding staining intensity A: T(X) = A * S. Here, T transforms the image into optical density space, and S consists of a background vector orthogonal to unit length row vectors representing the H&E and DAB color components.

[0037] S220: To obtain the enhanced staining intensity, randomly scale and translate the staining intensity;

[0038] Next, the intensity A of each pigment spot j and pixel i ij is scaled by a random coefficient α j and shifted by a random offset β j : A’ ij = A ij * α j + β j .

[0039] S230: Randomly rotate the color base of the pigment to obtain the enhanced color base of the pigment.

[0040] Furthermore, the color component S of each pigment spot j j is affected. If (1, φ(S j ), θ(S j )) are the spherical coordinates of S j , then the angular displacement: φ(S’ j ) = φ(S j ) + δφ j , θ(S’ j ) = θ(S j ) + δθ j is randomly rotated by. The angles φ(S’ j ) and θ(S’ j ) are constrained between 0 and π / 2, so the perturbed color component S j remains in the valid optical density space. This perturbation randomly changes the color gamuts of H&E and DAB around the original IHC image. All scalings α j , translations β j [[ID= 56]], rotation parameters δφ j and δθ j [[ID= 60]]are randomly sampled from a uniform distribution with a certain range.

[0041] S240: Integrate the enhanced staining intensity and the enhanced color base to complete the enhancement of the staining components.

[0042] After integration, the enhanced image is as follows: X’ = T -1The result is (A'*S'), and the enhancement of the staining component is complete. The effectiveness of the staining component enhancement technique can be evaluated using the Camelion17 dataset. The Camelion17 dataset contains thousands of high-resolution breast cancer tissue slice images from various cases, some containing cancerous areas and others containing normal tissue. Researchers can use these images to train and test deep learning models to automatically detect and identify cancerous areas, helping physicians make faster and more accurate diagnoses and treatment plans. Through this dataset, it can be verified that the staining component enhancement technique in this application can fill distribution gaps in histological images and improve the stability and generalization of metastasis classification models trained using associative learning.

[0043] In one embodiment, the step of “using a federated learning-enhanced dataset to build a tumor segmentation model” further includes the following steps: preprocessing multiple enhanced datasets and inputting the preprocessing results into a federated learning server to build a tumor segmentation model. Preprocessing is used to reduce the amount of data input to the federated learning server.

[0044] Specifically, preprocessing multiple enhanced datasets does not require uploading the original data (i.e., the multiple enhanced datasets) to the federated learning server. Instead, mapping results and updates related to the original data are registered to the federated learning server. Compared to directly uploading the original data to the federated learning server, uploading processed data reduces the load on the server while maintaining the security of the original data, thus avoiding the leakage of some private content contained in the original data.

[0045] In one embodiment, the step of "preprocessing multiple enhanced datasets and inputting the preprocessing results into a federated learning server to construct a tumor segmentation model" further includes the following steps:

[0046] Step "Build local and global models based on the learning rules of the federated learning server"

[0047] Step "Train the local model using multiple emphasis datasets to obtain the parameters of the local model." Specifically, these multiple emphasis datasets can be obtained from different sources, such as the aforementioned NTUH and Warwick University. The local model is then trained locally based on these datasets to obtain the parameters of the local model, which are weights, biases, etc., related to the original data.

[0048] The step is to "input the parameters of the local model into the federated learning server and train the global model through the clustering pipeline." In federated learning, the clustering pipeline refers to a set of processes and methods for clustering data in a distributed environment. The federated learning server is responsible for inputting the parameters of the local model, then aggregating the parameters of the local model to obtain a global clustering model. Both the local model and the global model are based on federated learning rules.

[0049] The first step involves "training a global model using multiple enhanced datasets and inputting the parameters of the trained global model into a federated learning server to build a tumor segmentation model." Specifically, after training the global model on the federated learning server, the global model is distributed and returned to each local site. Each local site retrains the global model based on multiple enhanced datasets and inputs the parameters obtained during training into the federated learning server. This learning process is repeated until the global model trained by the federated learning server meets certain pre-configured requirements and stops. The trained global model becomes the required tumor segmentation model.

[0050] Please refer to Figure 3, which shows another schematic diagram of a whole slide image processing method provided in one embodiment of the present application. In one embodiment, step S400 (step d) "perform cell membrane staining detection on the tumor region and classify the tumor region based on the integrity of the cell membrane and the level of staining" further includes the following steps:

[0051] S410 (Step d1): Detect nuclei in the tumor region and determine their locations. First, nuclei are detected in the IHC image. Scalar features are calculated for each pixel by combining hue, saturation, value, and RGB intensity variance. Based on these scalar features, a nuclear segmentation mask is generated using Otsu thresholding. Furthermore, the nuclear segmentation is divided into segmentation masks for each nucleus by two consecutive segmentation processes including connected component analysis, local peak detection, and a watershed algorithm. Uncovered nuclear components are further filtered based on their size and roundness.

[0052] S420 (Step d2): Cell membrane detection is performed based on the nuclear position to determine the location of the cell membrane corresponding to the nucleus. Specifically, the nuclear segmentation mask is expanded based on the nuclear position to determine the location of the cell membrane corresponding to the nucleus. For nuclei detected from the IHC image, the segmentation mask of each nucleus is first expanded to determine its membrane region. The degree of expansion is dynamically calculated based on approximating the cell membrane as a stack of hexagons of circular objects.

[0053] S430 (Step d3): Classify cell membranes based on staining levels and obtain pre-classification results. Specifically, color separation is performed on each pixel of the cell membrane and converted to an H&E-DAB representation; each pixel of the cell membrane is classified based on the staining intensity in the DAB channel according to a preset threshold, and the classification level of each pixel is obtained; pre-classification results are obtained based on the classification level of each pixel. Each nuclear membrane region is classified based on its staining level. Pixels within the membrane region are converted to an H&E-DAB representation by color separation. Then, based on three preset thresholds for the DAB channel, they are divided into negative, faint, weak, and strong levels, which are the results of the pre-classification. Each membrane is classified as having the strongest staining level with a sufficient number of pixels in the membrane region.

[0054] S440 (Step d4): Classify tumor regions based on pre-classification results and cell membrane integrity. Specifically, cell membrane integrity is obtained using a skeletonization algorithm. The pre-classification results are refined based on cell membrane integrity to classify tumor regions. Cell membrane regions are further classified based on staining integrity, resulting in classifications of negative, faint-incomplete, faint-complete, weak-incomplete, weak-complete, strong-incomplete, and strong-complete. For each membrane, the skeletonization algorithm is applied to pixels corresponding to the staining level of the membrane region and one level weaker. The skeleton contour is then processed by a closing operation. If the skeleton contour is closed and encloses the main region of the membrane, the membrane is classified as complete; otherwise, it is considered incomplete.

[0055] In one embodiment, classification results are graded according to preset criteria. These preset criteria include the ASCO CAP 2018 guidelines. The whole-slide image processing method includes determining the level of HER2 in the target image based on the ASCO CAP 2018 guidelines and the classification results. Specifically, the American Society of Clinical Oncology (ASCO) and the College of American Pathologists (CAP) jointly published guidelines on cancer diagnosis and treatment in 2018, providing physicians and pathologists with the latest diagnostic and treatment recommendations. The obtained classification results can be used to determine the HER2 level of the target image as 0, 1+, 2+, or 3+ according to the guidelines.

[0056] Based on the same technical concept, the disclosure also provides embodiments of tissue image processing systems relating to whole-slide imaging, comprising memory, a processor, and a computer program stored in the memory, wherein the processor executes the computer program to carry out the steps of the whole-slide image processing method in the embodiment.

[0057] The whole-slide image processing system disclosed herein constructs a tumor segmentation model by expanding staining components, increasing affinity with target images from different institutions and accurately presenting tumor regions. It then classifies the HER2 level of the target image by performing cell membrane stain detection on the tumor region. This system accurately evaluates the HER2 level in the target image and optimizes target images obtained from different institutions, enabling cross-institutional model learning without compromising user privacy.

[0058] Based on the same technical concept, this disclosure further provides a computer program product (e.g., a computer-readable recording medium) including a computer program / instruction, which, when executed by a processor, carries out steps of the whole slide image processing method in the embodiment.

[0059] In the embodiments described above, the descriptions of each embodiment focus on different aspects. Parts not described or documented in detail in one embodiment can be referenced to the relevant descriptions of other embodiments. Furthermore, the embodiments described above can be freely combined as needed.

[0060] [Cross-reference with related applications] This application claims priority to Provisional Application No. 63 / 550,591 filed on 6 February 2024, the contents of which are incorporated herein by reference in their entirety.

Claims

1. (a) Acquire a target image, (b) Obtain a dataset, enhance the staining components of the dataset, and use the enhanced dataset for federative learning to construct a tumor segmentation model. The dataset includes test images, and these test images are immunohistochemical images. (c) Input the target image into the tumor segmentation model and determine the tumor region of the target image, (d) Perform cell membrane staining detection on the tumor region, and further classify the tumor region based on the integrity of the cell membrane and the level of staining of the cell membrane. (e) Grade based on the classification results, Obtaining the aforementioned dataset and highlighting the staining components of the aforementioned dataset is, The test image is obtained from the dataset, color separation is performed on the image, and the color base and staining intensity of the pigments in the image are obtained. The staining intensity of the dye is randomly scaled and randomly shifted to enhance the staining intensity of the dye. The color base of the aforementioned pigment is randomly rotated to obtain the enhanced color base of the aforementioned pigment. The enhanced color base and enhanced dyeing intensity of the aforementioned dye are integrated to complete the enhancement of the dyeing component. This further includes, Tissue imaging processing method.

2. Using the enhanced dataset for federated learning to construct the tumor segmentation model is The enhanced dataset is preprocessed, and the preprocessing results are input to a federated learning server to construct the tumor segmentation model. This further includes, The tissue imaging processing method according to claim 1.

3. Preprocessing the enhanced dataset and inputting the preprocessing results into a federated learning server to construct the tumor segmentation model is: Based on the learning rules of the aforementioned federated learning server, a local model and a global model are constructed. To obtain multiple parameters of the local model, the local model is trained on the enhanced dataset. The parameters of the local model are input to the federated learning server, and the global model is trained via a clustering pipeline. The global model is trained on the enhanced dataset, the parameters of the trained global model are input to the federative learning server, and the tumor segmentation model is constructed. This further includes, The tissue imaging processing method according to claim 2.

4. (d) is, (d1) In order to determine the location of the nuclei in the tumor region, nuclear detection is performed on the tumor region. (d2) Based on the position of the nucleus, cell membrane detection is performed to determine the position of the cell membrane corresponding to the nucleus, (d3) Classify the cell membranes based on the staining level of the cell membranes and obtain the pre-classification results. (d4) Classify the tumor region based on the pre-classification results and the integrity of the cell membrane. This further includes, The tissue imaging processing method according to claim 1.

5. (d2) is, Based on the position of the nucleus, the segmentation mask of the nucleus is enlarged to determine the position of the cell membrane corresponding to the nucleus. This further includes, The tissue imaging processing method according to claim 4.

6. (d3) is, Color separation is performed on each pixel of the cell membrane, and the representation is converted to hematoxylin-diaminobenzidine. Based on the staining intensity of the diaminobenzidine channel, each pixel of the cell membrane is classified according to a preset threshold, and the classification level of each pixel is obtained. A pre-classification result is obtained based on the classification level of each pixel. This further includes, The tissue imaging processing method according to claim 4.

7. (d4) is, The integrity of the cell membrane is obtained using a skeletonization algorithm. In order to classify the tumor region, the pre-classification results are refined based on the integrity of the cell membrane. This further includes, The tissue imaging processing method according to claim 4.

8. Memory and Processor and The computer program stored in the aforementioned memory, Includes, The processor implements the tissue imaging processing method according to any one of claims 1 to 7 by executing the computer program. Tissue imaging processing system.

9. Including computer programs / instructions, When the computer program / instruction is executed by the processor, it performs the tissue imaging processing method according to any one of claims 1 to 7. Computer-readable recording media for tissue imaging processing.

Images