Classification method and classification device for pancreatic neuroendocrine tumors

Abstract

The invention discloses a classification method and a classification device for pancreatic neuroendocrine tumors. The classification method comprises the following steps: using a pre-trained segmentation model to segment a plurality of HE full slice images from a plurality of patients; quantizing each segmented HE full slice image to determine a plurality of first feature parameters; based on the cellular morphology and topological features of each HE full-slice image, extracting a plurality of second feature parameters from each HE full-slice image; for each HE full slice image, screening out a target feature from the plurality of first feature parameters and the plurality of second feature parameters, and calculating a pathological omics score; inputting the pathological omics scores and the conventional pathological features of the plurality of HE full slice images into a logistic regression model for training to obtain a trained logistic regression model; and classifying the HE full slice images to be classified by using the logistic regression model. According to the method, pancreatic neuroendocrine tumor G classification can be quickly and accurately realized at low cost.

Description

Technical Field

[0001] This application relates to the field of artificial intelligence, and particularly to a classification method, a classification device, a medium, an electronic device, and a computer program product for pancreatic neuroendocrine tumors. Background Art

[0002] Pancreatic neuroendocrine tumors (PNETs), as a rare tumor originating from peptidergic neurons and neuroendocrine cells in the pancreas, have shown a significant increase in incidence in recent years, posing a serious threat to the health of patients.

[0003] According to the mitotic index and Ki-67 index, PNETs are subdivided into three grades: G1, G2, and G3. Pathological classification is crucial in patient management as it directly affects treatment strategies and is closely related to survival rates. The European Neuroendocrine Tumor Society (ENETS) guidelines recommend radical resection for all PNETs larger than 2 cm and functional PNETs. G1-grade PNETs can achieve good survival prognosis through surgeries such as central pancreatectomy and enucleation. However, due to the high recurrence risk, G3-grade PNETs require adjuvant chemotherapy after resection.

[0004] The World Health Organization classification stipulates that accurate classification of PNETs relies on Ki-67 (MIB-1) immunohistochemistry. Precise assessment of the tumor proliferation index requires manual counting of at least 500 - 2000 tumor cells. This method is labor-intensive, costly, and may produce inconsistent counting results due to the variability of IHC (immunohistochemistry) preparations and differences in pathologist subjectivity. Especially in highly differentiated PNETs, extremely small differences (1 - 5%) may also significantly affect tumor classification, leading to changes in treatment regimens.

[0005] Therefore, the current methods are not only highly subjective but also time-consuming and laborious, vulnerable to differences in pathologists' experience and judgment, resulting in inconsistencies and inaccuracies in classification results. Summary of the Invention

[0006] Embodiments of this application provide a classification method for pancreatic neuroendocrine tumors, a classification device for pancreatic neuroendocrine tumors, a medium, an electronic device, and a computer program product.

[0007] In a first aspect, embodiments of this application provide a classification method for pancreatic neuroendocrine tumors for an electronic device, the classification method comprising:

[0008] Segmentation step: Using a pre-trained segmentation model, segment multiple HE whole-slide images from multiple patients respectively to segment tumor glands and tumor stroma in each HE whole-slide image;

[0009] Quantification step: For each segmented HE whole-slide image, perform quantification to determine a plurality of first feature parameters, where the plurality of first feature parameters include a plurality of tumor gland-related parameters, a plurality of tumor bed-related parameters, and a plurality of tumor stroma ratio-related parameters;

[0010] Extraction step: Based on the cell morphology and topological features of each HE whole-slide image, extract a plurality of second feature parameters from each HE whole-slide image, where the plurality of second feature parameters include a plurality of morphological parameters, a plurality of texture feature parameters, and a plurality of topological feature parameters;

[0011] Screening step: For each HE whole-slide image, screen out target features from the plurality of first feature parameters and the plurality of second feature parameters, and calculate the pathological omics score of each HE whole-slide image;

[0012] Training step: Input the pathological omics scores of multiple HE whole-slide images and the conventional pathological features of multiple HE whole-slide images into a logistic regression model for training to obtain a trained logistic regression model;

[0013] Classification step: Use the trained logistic regression model to classify the HE whole-slide images to be classified.

[0014] In a second aspect, the present invention provides a classification device for pancreatic neuroendocrine tumors, and the classification device includes:

[0015] Segmentation unit: Using a pre-trained segmentation model, segment multiple HE whole-slide images from multiple patients respectively to segment tumor glands and tumor stroma in each HE whole-slide image;

[0016] Quantification unit: For each segmented HE whole-slide image, perform quantification to determine a plurality of first feature parameters, where the plurality of first feature parameters include a plurality of tumor gland-related parameters, a plurality of tumor bed-related parameters, and a plurality of tumor stroma ratio-related parameters;

[0017] Extraction unit: Based on the cell morphology and topological features of each HE whole-slide image, extract a plurality of second feature parameters from each HE whole-slide image, where the plurality of second feature parameters include a plurality of morphological parameters, a plurality of texture feature parameters, and a plurality of topological feature parameters;

[0018] A screening unit that, for each HE whole-slide image, screens out target features from the multiple first feature parameters and the multiple second feature parameters, and calculates the pathomics score for each HE whole-slide image;

[0019] A training unit that inputs the pathomics scores of multiple HE whole-slide images and the conventional pathological features of multiple HE whole-slide images into a logistic regression model for training to obtain a trained logistic regression model;

[0020] A classification unit that uses the trained logistic regression model to classify the HE whole-slide images to be classified.

[0021] In a third aspect, an embodiment of the present invention provides a computer-readable storage medium, on which instructions are stored, and when the instructions are executed on a computer, the computer is caused to execute the classification method for pancreatic neuroendocrine tumors according to any one of the first aspects.

[0022] In a fourth aspect, an embodiment of the present invention provides an electronic device, including: one or more processors; one or more memories; the one or more memories store one or more programs, and when the one or more programs are executed by the one or more processors, the electronic device is caused to execute the classification method for pancreatic neuroendocrine tumors according to the first aspect.

[0023] In a fifth aspect, an embodiment of the present application provides a computer program product, including computer-executable instructions, and the instructions are executed by a processor to implement the classification method for pancreatic neuroendocrine tumors according to the first aspect.

[0024] In the present invention, multiple first feature parameters related to tumor glands and tumor stroma, and multiple second feature parameters related to cell morphology and topological features are extracted from HE whole-slide images, and features with significant correlation are screened out from the multiple first feature parameters and second feature parameters to calculate the pathomics score, and then the pathomics score and the conventional pathological features obtained from the HE whole-slide images are used to train (construct) a classification model. In this way, according to the output of the classification model, the G classification of pancreatic neuroendocrine tumors can be achieved quickly, accurately, and at low cost. BRIEF DESCRIPTION OF THE DRAWINGS

[0025] Figure 1 According to an embodiment of the present application, a flowchart of a classification method for pancreatic neuroendocrine tumors is shown;

[0026] Figure 2 According to an embodiment of the present application, a structural diagram of a classification device for pancreatic neuroendocrine tumors is shown;

[0027] Figure 3According to an embodiment of the present application, a block diagram of an electronic device is shown. Detailed implementation manners

[0028] Illustrative embodiments of the present application include, but are not limited to, a classification method for pancreatic neuroendocrine tumors, a classification device for pancreatic neuroendocrine tumors, a medium, an electronic device, and a computer program product.

[0029] The embodiments of the present application will be further described in detail below with reference to the accompanying drawings.

[0030] Figure 1 A classification method for pancreatic neuroendocrine tumors according to an embodiment of the present application is shown, which is used for an electronic device. Specifically, in the segmentation step S11, a pre-trained segmentation model is used to separately segment multiple HE whole-slide images from multiple patients to segment tumor glands and tumor stroma in each HE whole-slide image.

[0031] Hematoxylin and eosin (HE)-stained histopathological sections of surgically resected tumors from multiple patients are obtained, and scanned into HE whole-slide images (WSIs) at a magnification of ×40 using a scanner, so as to obtain multiple HE whole-slide images.

[0032] Each HE whole-slide image is cropped into multiple small image patches, and a pre-trained segmentation model is used to segment each small image patch to segment tumor glands and tumor stroma in each small image patch, and then the multiple segmented small image patches are integrated together to form a segmented HE whole-slide image.

[0033] Specifically, due to the huge size and resolution of HE whole-slide images, if the whole image is input, it will pose a huge challenge to the processor. Therefore, the high-resolution HE whole-slide images are cropped into multiple smaller image patches, such as 256×256 pixels, and matched with the input size of the segmentation model. To optimize the model performance, image enhancement techniques such as normalization, denoising, and histogram equalization are applied. In addition, strict data verification is also carried out to ensure the consistency of the number of labels and the precise alignment between the mask and the image size.

[0034] After using a pre-trained segmentation model to segment each small image patch, the multiple segmented small image patches are integrated together to form a segmented HE whole-slide image, that is, a segmentation map of a unified and complete full-frame HE whole-slide image is formed. The specific steps are as follows:

[0035] (1) Obtain the position information of each small image patch by recording the starting coordinates.

[0036] (2) Extract the metadata of the HE whole-slide image using, for example, the OpenSlicer software package, and then create a blank matrix that matches the size of the HE whole-slide image. This matrix will be used to store the processing results of each small image patch, and finally construct a comprehensive view of the entire HE whole-slide image.

[0037] (3) Map the small patches to the HE whole-slide image matrix according to the coordinate information. Each small image patch is accurately aligned to ensure no overlap or omission.

[0038] (4) All small image patches are integrated together to achieve the overall segmentation of the target tissue components. The segmentation results from different small image patches are merged to identify specific tissue structures, and finally a segmentation map of the complete HE whole-slide image is formed.

[0039] The pre-trained segmentation model is, for example, the pre-trained semantic segmentation model DeepLabV3+. The training process is as follows:

[0040] Randomly select the HE whole-slide images of some patients from the training set. After being segmented into multiple small image patches, the tumor glands and tumor stroma are labeled by, for example, pathologists. After the labeling process is completed, multiple classes of masks are generated, and a class value is assigned to each pixel: 1 represents tumor glands, 0 represents background, and 2 represents tumor stroma. The labeled small patches and their corresponding masks are exported and stored in PNG format.

[0041] Input the multiple labeled small image patches into the semantic segmentation model DeepLabV3+ for training to obtain the pre-trained segmentation model.

[0042] In the quantization step S12, each segmented HE whole-slide image is quantized to determine multiple first feature parameters. The multiple first feature parameters include multiple tumor gland-related parameters, multiple tumor bed-related parameters, and multiple tumor stroma ratio-related parameters.

[0043] Here, an image processing tool (such as OpenCV 4.8.0) is used to quantize each segmented HE whole-slide image.

[0044] Specifically, to obtain the true area of a specific tissue component, the image processing tool OpenCV 4.8.0 is used. First, the total number of corresponding pixels (P-i) of the component in the segmented HE whole-slide image (i.e., the segmented image) is counted. Then, according to the metadata of the image, the actual lengths represented by each pixel in the horizontal (x) and vertical (y) directions (S-x and S-y) are obtained. Finally, the total number of pixels is multiplied by the pixel spacings in these two directions to calculate the actual area of the component. The specific formula is: S = P-i * S-x * S-y. Ultimately, three sets of quantifiable first feature parameters are obtained, including multiple tumor gland-related parameters (e.g., 8), multiple tumor bed-related parameters (e.g., 26), and multiple tumor stroma ratio (the ratio of tumor glands to stroma within the tumor bed) -related parameters (e.g., 15).

[0045] In the extraction step S13, based on the cell morphology and topological features of each HE whole-slide image, multiple second feature parameters are extracted from each HE whole-slide image. The multiple second feature parameters include multiple morphological parameters, multiple texture feature parameters, and multiple topological feature parameters.

[0046] First, the tumor cells, stromal cells, and lymphocytes in each HE whole-slide image are segmented and classified. Specifically, a pre-trained HoVer-Net model is used to simultaneously perform nuclear segmentation and classification, including tumor cells, stromal cells, and lymphocytes. This model is pre-trained on the PanNuke dataset and separates aggregated units by leveraging the horizontal and vertical distances from nuclear pixels to their centroids. The input HE whole-slide image, and the model outputs a corresponding.json file that contains all the information about nuclear segmentation and classification for each sample.

[0047] Secondly, cell-level feature extraction is performed on the tumor cells, stromal cells, and lymphocytes to extract multiple second feature parameters.

[0048] Specifically, pathologists outline the region of interest (ROI) and generate a corresponding.xml file. Then, the HE whole-slide image,.xml file, and the above-mentioned corresponding.json file are used as inputs. For each sample, feature extraction is performed using WSI Graph (graphical representation) to obtain a folder containing four.csv data files. For each cell type (tumor cells, lymphocytes, and stromal cells), three.csv files store the attributes of all cells of that type, with each cell identified by a unique cell ID (identifier) and its centroid coordinates. Another.csv file stores the edge information of the sample, i.e., each edge represented by connected cell IDs.

[0049] Finally, multiple second feature parameters (e.g., 378) were extracted from each HE whole-slide image. These feature parameters were divided into three categories: (1) morphological feature parameters, which describe the shape and contour of the cell nucleus; (2) texture feature parameters, which characterize the local pixel distribution pattern within the cell nucleus contour; (3) topological feature parameters, which describe the features regarding the relative positions, interaction patterns between cells, and the resulting spatial organization, specifically focusing on the interactions among tumor cells, inflammatory cells, and stromal cells. For each cell type, three types of topological features were extracted: cell-to-cell edge length (e.g., T-I_MinEdgeLength and T-I_MeanEdgeLength, representing the minimum and average edge lengths between tumor cells and inflammatory cells), number of subgraphs (e.g., I-S_Nsubgraph, representing the number of subgraphs between inflammatory cells and stromal cells), centrality and other network properties (e.g., I-S_Closeness and I-S_Betweenness, representing the closeness centrality and betweenness centrality between inflammatory cells and stromal cells).

[0050] In the screening step S14, for each HE whole-slide image, target features were screened out from multiple first feature parameters and multiple second feature parameters, and the pathomics score of each HE whole-slide image was calculated.

[0051] For each HE whole-slide image, LASS regression was used to screen out target features from multiple first feature parameters and multiple second feature parameters, and the pathomics score of each HE whole-slide image was calculated through the LASSO regression equation. The target features are features with significant correlations.

[0052] Specifically, for example, for each HE whole-slide image, multiple first feature parameters (49) and multiple second feature parameters (e.g., 378) as described above could be obtained, a total of 427 feature parameters. Correlation analysis was performed on the 427 feature parameters, and LASS regression was used to further screen out features with significant correlations as target features. For example, 6 target features were screened out. Finally, a pathomics score was generated through the LASSO regression equation. There were significant differences in the pathomics scores between grade G1 and grade G2 / 3.

[0053] It can be understood that a corresponding pathomics score can be calculated for each HE whole-slide image.

[0054] In the training step S15, the pathomics scores of multiple HE whole-slide images and the conventional pathological features of multiple HE whole-slide images were input into a logistic regression model for training to obtain a trained logistic regression model.

[0055] A pathological report can be obtained from the HE whole-slide image, and the pathological report includes conventional pathological features. The conventional pathological features include tumor size, tumor location, number of tumors, tumor burden score (TBS), T stage, N stage, M stage (the 9th edition of the American Joint Committee on Cancer TNM Staging Manual), tumor texture (solid / cystic, where cystic is defined as any proportion containing cystic components), and so on. The calculation method proposed according to previous studies is defined as: TBS2 = (maximum tumor diameter)2 + (number of tumors)2.

[0056] In this embodiment, tumor size, T stage, N stage, M stage, and TBS are selected as parameters for training the logistic regression model.

[0057] In the classification step S16, the trained logistic regression model is used to classify the HE whole-slide image to be classified.

[0058] It can be understood that the trained logistic regression model, as a classification model, can classify the HE whole-slide image to be classified. In this embodiment, the classification model divides the HE whole-slide image into two categories, one is grade G1, and the other is grade G2 / G3.

[0059] In addition, in this embodiment, through various methods such as ROC curve, Kaplan-Meier survival analysis, and decision curve analysis, the performance of the classification model is comprehensively verified, and its clinical practicability is evaluated.

[0060] In the present invention, a plurality of first feature parameters related to tumor glands and tumor stroma, and a plurality of second feature parameters related to cell morphology and topological features are extracted from the HE whole-slide image, and features with significant correlation are screened out from the plurality of first feature parameters and second feature parameters to calculate the pathological omics score, and then the pathological omics score and the conventional pathological features obtained from the HE whole-slide image are used to train (construct) the classification model. Thus, according to the output of the classification model, the G classification of pancreatic neuroendocrine tumors can be achieved quickly, accurately, and at low cost.

[0061] The present invention also provides a classification device 20 for pancreatic neuroendocrine tumors. The classification device 20 includes:

[0062] A segmentation unit 201 that uses a pre-trained segmentation model to separately segment a plurality of HE whole-slide images from multiple patients to segment the tumor glands and tumor stroma in each HE whole-slide image;

[0063] A quantization unit 202 that quantifies each segmented HE whole-slide image to determine a plurality of first feature parameters, where the plurality of first feature parameters include a plurality of tumor gland-related parameters, a plurality of tumor bed-related parameters, and a plurality of tumor stroma ratio-related parameters;

[0064] The extraction unit 203 extracts a plurality of second feature parameters from each HE whole-slide image based on the cell morphology and topological features of each HE whole-slide image, and the plurality of second feature parameters include a plurality of morphological feature parameters, a plurality of texture feature parameters, and a plurality of topological feature parameters;

[0065] The screening unit 204 screens out target features from the plurality of first feature parameters and the plurality of second feature parameters for each HE whole-slide image, and calculates the pathomics score of each HE whole-slide image;

[0066] The training unit 205 inputs the pathomics scores of the plurality of HE whole-slide images and the conventional pathological features of the plurality of HE whole-slide images into a logistic regression model for training to obtain a trained logistic regression model;

[0067] The classification unit 206 classifies the HE whole-slide images to be classified using the trained logistic regression model.

[0068] It can be understood that the segmentation unit 201, the quantification unit 202, the extraction unit 203, the screening unit 204, the training unit 205, and the classification unit 206 can be implemented by Figure 3 the processor 102 in the electronic device 100 having the functions of these modules or units.

[0069] The present invention also provides a computer-readable storage medium, on which instructions are stored, and when the instructions are executed on a computer, the computer is caused to execute Figure 1 the method shown in

[0070] The present invention also provides a computer program product, including computer-executable instructions, and the instructions are executed by the processor 102 to implement Figure 1 the method shown in

[0071] Now referring to Figure 3 , Figure 3 FIG. schematically shows an exemplary electronic device 1400 according to an embodiment of the present invention. In one embodiment, the electronic device 1400 may include one or more processors 1404, a system control logic unit 1408 connected to at least one of the processors 1404, a system memory 1412 connected to the system control logic unit 1408, a non-volatile memory (NVM) 1416 connected to the system control logic unit 1408, and a network interface 1420 connected to the system control logic unit 1408.

[0072] In some embodiments, the processor 1404 may include one or more single-core or multi-core processors. In some embodiments, the processor 1404 may include any combination of a general-purpose processor and a dedicated processor (e.g., a graphics processor, an application processor, a baseband processor, etc.). In embodiments where the electronic device 1400 employs an eNB (Evolved Node B) or an RAN (Radio Access Network) controller, the processor 1404 may be configured to execute various compliant embodiments, such as, for example, the embodiments as Figure 1 shown.

[0073] In some embodiments, the system control logic unit 1408 may include any suitable interface controller to provide any suitable interface to at least one of the processors 1404 and / or any suitable device or component communicating with the system control logic unit 1408.

[0074] In some embodiments, the system control logic unit 1408 may include one or more memory controllers to provide an interface to the system memory 1412. The system memory 1412 may be used to load and store data and / or instructions. In some embodiments, the system memory 1412 of the electronic device 1400 may include any suitable volatile memory, such as a suitable dynamic random access memory (DRAM).

[0075] The non-volatile memory 1416 may include one or more tangible, non-transitory computer-readable media for storing data and / or instructions. In some embodiments, the non-volatile memory 1416 may include any suitable non-volatile memory such as flash memory and / or any suitable non-volatile storage device, such as at least one of an HDD (Hard Disk Drive), a CD (Compact Disc) drive, and a DVD (Digital Versatile Disc) drive.

[0076] The non-volatile memory 1416 may include a portion of the storage resources installed on the device of the electronic device 1400, or it may be accessible by the electronic device but not necessarily part of the electronic device. For example, the non-volatile memory 1416 may be accessed via the network interface 1420 over a network.

[0077] Specifically, the system memory 1412 and the non-volatile memory 1416 may respectively include: a temporary copy and a permanent copy of the instructions 1424. The instructions 1424 may include: when executed by at least one of the processors 1404, causing the electronic device 1400 to implement as Figure 1Instructions of the method shown. In some embodiments, instructions 1424, hardware, firmware, and / or its software components may alternatively / additionally be placed in system control logic unit 1408, network interface 1420, and / or processor 1404.

[0078] Network interface 1420 may include a transceiver for providing a radio interface for electronic device 1400 to communicate with any other suitable devices (such as front-end modules, antennas, etc.) via one or more networks. In some embodiments, network interface 1420 may be integrated with other components of electronic device 1400. For example, network interface 1420 may be integrated with at least one of processor 1404, system memory 1412, non-volatile memory 1416, and a firmware device with instructions (not shown). When at least one of the instructions is executed by processor 1404, electronic device 1400 implements as Figure 1 the method shown.

[0079] Network interface 1420 may further include any suitable hardware and / or firmware to provide a multiple-input multiple-output radio interface. For example, network interface 1420 may be a network adapter, wireless network adapter, telephone modem, and / or wireless modem.

[0080] In one embodiment, at least one of processors 1404 may be packaged with the logic of one or more controllers for system control logic unit 1408 to form a system-in-package (SiP). In one embodiment, at least one of processors 1404 may be integrated with the logic of one or more controllers for system control logic unit 1408 on the same die to form a system-on-chip (SoC).

[0081] Electronic device 1400 may further include: input / output (I / O) device 1432. I / O device 1432 may include a user interface that enables a user to interact with electronic device 1400; the design of the peripheral component interface enables peripheral components to also interact with electronic device 1400. In some embodiments, electronic device 1400 further includes sensors for determining at least one of environmental conditions and location information related to electronic device 1400.

[0082] In some embodiments, the user interface may include, but is not limited to, a display (e.g., liquid crystal display, touch screen display, etc.), speaker, microphone, one or more cameras (e.g., still image camera and / or video camera), flashlight (e.g., light-emitting diode flash), and keyboard.

[0083] Embodiments of the mechanisms disclosed in this application can be implemented in hardware, software, firmware, or combinations of these implementation methods. Embodiments of this application can be implemented as a computer program or program code executed on a programmable system, which includes at least one processor, a storage system (including volatile and non-volatile memories and / or storage elements), at least one input device, and at least one output device.

[0084] The program code can be applied to the input instructions to perform the various functions described in this application and generate output information. The output information can be applied to one or more output devices in a known manner. For the purposes of this application, a processing system includes any system having a processor such as, for example, a digital signal processor (DSP), a microcontroller, an application specific integrated circuit (ASIC), or a microprocessor.

[0085] The program code can be implemented in a high-level procedural language or an object-oriented programming language to communicate with the processing system. When needed, the program code can also be implemented in assembly language or machine language. In fact, the mechanisms described in this application are not limited to the scope of any particular programming language. In either case, the language can be a compiled language or an interpreted language.

[0086] In some cases, the disclosed embodiments can be implemented in hardware, firmware, software, or any combination thereof. The disclosed embodiments can also be implemented as instructions carried or stored on one or more transient or non-transient machine-readable (e.g., computer-readable) storage media, which can be read and executed by one or more processors. For example, the instructions can be distributed via a network or via other computer-readable media. Thus, a machine-readable medium can include any mechanism for storing or transmitting information in a form readable by a machine (e.g., a computer), including but not limited to, a floppy disk, a compact disc, a CD-ROM, a magneto-optical disc, a ROM, a RAM, an EPROM, an EEPROM, a magnetic or optical card, flash memory, or a tangible machine-readable memory for transmitting information (e.g., carrier waves, infrared signals, digital signals, etc.) in electrical, optical, acoustic, or other forms using the Internet. Thus, a machine-readable medium includes any type of machine-readable medium suitable for storing or transmitting electronic instructions or information in a form readable by a machine (e.g., a computer).

[0087] In the accompanying drawings, some structural or method features may be shown in a particular arrangement and / or order. However, it should be understood that such a particular arrangement and / or ordering may not be required. Instead, in some embodiments, these features may be arranged in a manner and / or order different from that shown in the illustrative drawings. Additionally, the inclusion of a structural or method feature in a particular figure does not imply that such a feature is required in all embodiments, and in some embodiments, these features may not be included or may be combined with other features.

[0088] It should be noted that each unit / module mentioned in the device embodiments of this application is a logical unit / module. Physically, a logical unit / module can be a physical unit / module, a part of a physical unit / module, or can be implemented as a combination of multiple physical units / module. The physical implementation manner of these logical units / modules themselves is not the most important. The combination of the functions implemented by these logical units / modules is the key to solving the technical problems proposed in this application. In addition, to highlight the innovative part of this application, the above device embodiments of this application do not introduce units / modules that are not closely related to solving the technical problems proposed in this application. This does not mean that there are no other units / modules in the above device embodiments.

[0089] It should be noted that in the examples and descriptions of this patent, relational terms such as first and second are only used to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Moreover, the term "comprising", "including" or any other variant thereof is intended to cover non-exclusive inclusion, such that a process, method, article or device comprising a series of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such process, method, article or device. Without further limitation, an element defined by the statement "comprising one" does not exclude the presence of additional identical elements in the process, method, article or device comprising the element.

[0090] Although this application has been illustrated and described by reference to certain preferred embodiments thereof, those of ordinary skill in the art should understand that various changes in form and detail may be made thereto without departing from the spirit and scope of this application.

Claims

1. A method for classifying pancreatic neuroendocrine tumors, for use in electronic equipment, characterized in that: The classification method includes: In a segmentation step, a pre-trained segmentation model is used to segment multiple HE whole-slice images from multiple patients, so as to segment the tumor glands and tumor stroma in each HE whole-slice image; a quantification step, quantifying each segmented HE whole slice image to determine a plurality of first characteristic parameters, wherein the plurality of first characteristic parameters include a plurality of tumor gland-related parameters, a plurality of tumor bed-related parameters, and a plurality of tumor-stroma ratio-related parameters; an extraction step, based on the cell morphology and topological features of each HE whole slice image, extracting a plurality of second feature parameters from each HE whole slice image, wherein the plurality of second feature parameters include a plurality of morphological feature parameters, a plurality of texture feature parameters, and a plurality of topological feature parameters; a screening step, for each HE whole slice image, screening out target features from the plurality of first feature parameters and the plurality of second feature parameters, and calculating a pathological omics score for each HE whole slice image; a training step, inputting the pathological omics scores of the plurality of HE whole-slice images and the conventional pathological features of the plurality of HE whole-slice images into a logistic regression model for training, thereby obtaining a trained logistic regression model; In the classification step, the trained logistic regression model is used to classify the HE whole-slice images to be classified.

2. The classification method according to claim 1, characterized in that: In the segmentation step, each HE full-slice image is cropped into a plurality of small image blocks, and each small image block is segmented using the segmentation model to segment out the tumor glands and tumor stroma in each small image block, and the plurality of segmented small image blocks are integrated together to form a segmented HE full-slice image.

3. The classification method according to claim 1, characterized in that: In the quantification step, each segmented HE whole-slice image is quantified using an image processing tool.

4. The classification method according to claim 1, characterized in that: In the extraction step, the tumor cells, stromal cells, and lymphocytes in each HE whole slice image are segmented and classified, and cell-level feature extraction is performed on the tumor cells, the stromal cells, and the lymphocytes to extract the multiple second feature parameters.

5. The classification method according to claim 1, characterized in that: In the screening step, for each HE whole slice image, the target feature is screened out from the multiple first feature parameters and the multiple second feature parameters using LASS regression, and the pathological omics score of each HE whole slice image is calculated using the LASSO regression equation. The target feature is a feature with significant correlation.

6. The classification method according to claim 1, characterized in that: A pathology report is obtained from the HE whole-slice image, wherein the pathology report includes the conventional pathological features.

7. A classification device for pancreatic neuroendocrine tumors, characterized in that: The classification device comprises: A segmentation unit, using a pre-trained segmentation model, respectively segments a plurality of HE whole-slice images from a plurality of patients to segment the tumor glands and tumor stroma in each HE whole-slice image; a quantization unit, which quantifies each segmented HE whole slice image to determine a plurality of first characteristic parameters, wherein the plurality of first characteristic parameters include a plurality of tumor gland-related parameters, a plurality of tumor bed-related parameters, and a plurality of tumor-stroma ratio-related parameters; an extraction unit, extracting a plurality of second feature parameters from each HE whole slice image based on the cell morphology and topological features of each HE whole slice image, wherein the plurality of second feature parameters include a plurality of morphological feature parameters, a plurality of texture feature parameters, and a plurality of topological feature parameters; a screening unit, for each HE whole-slice image, screening out a target feature from the plurality of first feature parameters and the plurality of second feature parameters, and calculating a pathological omics score for each HE whole-slice image; A training unit, inputting the pathological omics scores of the plurality of HE whole-slice images and the conventional pathological features of the plurality of HE whole-slice images into a logistic regression model for training, thereby obtaining a trained logistic regression model; The classification unit uses the trained logistic regression model to classify the HE whole-slice image to be classified.

8. A computer-readable storage medium, characterized in that: The storage medium stores instructions, which, when executed on a computer, cause the computer to execute the method for classifying pancreatic neuroendocrine tumors according to any one of claims 1 to 6.

9. An electronic device, characterized in that: include: one or more processors; One or more memories; the one or more memories store one or more programs, and when the one or more programs are executed by the one or more processors, the electronic device executes the classification method for pancreatic neuroendocrine tumors described in any one of claims 1 to 6.

10. A computer program product comprising computer executable instructions, characterized in that: The instructions are executed by a processor to implement the method for classifying pancreatic neuroendocrine tumors according to any one of claims 1 to 6.

Images