A cancer survival prediction method and device based on whole slide pathological images

Abstract

The application discloses a cancer survival prediction method and device based on whole-slice pathological images, and relates to the technical field of medical artificial intelligence and digital pathological image analysis. The method comprises the following steps: acquiring whole-slice pathological images and patient survival information, and performing pretreatment; performing cell detection on the whole-slice pathological images to obtain cell structure data; dividing the whole-slice pathological images into a plurality of image regions, and determining regions of interest; screening the cell structure data in each region of interest to construct a cell subset; sorting the cell structure data according to the distribution relationship of cells in space to construct a cell structure sequence; constructing a cell structure modeling model, performing feature modeling on the cell structure sequence, and obtaining tissue structure features capable of representing the microstructure of tumor tissues; and inputting the tissue structure features into a survival prediction model to predict the survival risk of the patient. The application can improve the accuracy and stability of cancer survival prediction.

Description

Technical Field

[0001] This invention relates to the field of medical artificial intelligence and digital pathological image analysis technology, and in particular to a method and device for predicting cancer survival based on whole-section pathological images. Background Technology

[0002] Cancer is a major disease threatening human life and health, and predicting patient survival time is crucial for clinical prognosis assessment, treatment planning, and healthcare resource allocation. Survival prediction, through modeling and analyzing the time from diagnosis or treatment initiation to the occurrence of clinical events, can be used to assess a patient's risk level and disease progression, thus providing important decision-making references for clinicians. Therefore, how to accurately predict the survival risk of cancer patients has become an important research question in the fields of medical artificial intelligence and digital healthcare.

[0003] With the development of digital pathology technology, traditional glass slides are gradually being replaced by digital scanning, resulting in high-resolution whole-slide pathology images. Whole-slide pathology images can fully present the cellular structure and spatial distribution information of tumor tissue, providing a rich data source for studying the tumor microenvironment and disease progression. Therefore, using whole-slide pathology images to conduct cancer survival prediction research has gradually become an important research direction in digital pathology analysis.

[0004] Existing cancer survival prediction technologies based on whole-section pathological images typically employ the following technical process: First, acquire the patient's whole-section pathological image and preprocess the image; then, divide the whole-section pathological image into several image blocks and extract feature representations at the image block level; finally, model the extracted image features using machine learning or deep learning models to predict the patient's survival risk or survival time.

[0005] In the aforementioned existing technologies, image patches are typically used as basic modeling units. A holistic representation of the whole-section pathological image is obtained by aggregating features from multiple image patches, and survival prediction is performed accordingly. This type of method reduces the computational complexity of directly modeling whole-section pathological images to some extent and can be trained even in the absence of fine annotations. Therefore, it has been widely used in digital pathology survival analysis research.

[0006] However, from the perspective of tumor tissue structure and pathological analysis, cancer development is closely related to cellular structural characteristics and their spatial distribution. Modeling based solely on image block-level features is insufficient to accurately describe the spatial relationships between tumor cells, immune cells, and stromal cells, and also fails to fully reflect the structural characteristics of the tumor microenvironment. With the development of cell detection technologies, some studies have begun to attempt to extract cellular structural information from whole-section pathological images and build models based on cellular-level data to improve the expressive power of survival prediction models.

[0007] Although cell-level modeling can theoretically depict tissue structure information with greater precision, it still has significant shortcomings in practical applications, mainly in the following aspects: (1) The number of cells is huge, and the computational cost of modeling is high.

[0008] Whole-section pathological images typically contain a large number of cells. Modeling all cells directly would significantly increase computational complexity and storage overhead, thus affecting model training and inference efficiency.

[0009] (2) The distribution of cell types is uneven, and key cell information is easily overlooked.

[0010] In tumor tissue, the number of different types of cells varies greatly. If there is a lack of effective screening or sampling mechanisms, the model is easily affected by the dominant cells during training, thereby reducing its ability to identify key cellular structural information.

[0011] (3) Long-range spatial relationships between cells are difficult to model effectively.

[0012] Cellular structures in tumor tissues often exhibit complex spatial relationships, with some cells potentially having cross-regional spatial dependencies. Existing methods struggle to effectively capture these long-range spatial relationships while maintaining computational efficiency during the modeling process.

[0013] Therefore, how to fully utilize the cellular structure information in whole-slice pathological images and effectively model the spatial relationships between cells while ensuring computational efficiency, thereby improving the accuracy of cancer survival prediction, remains a problem that urgently needs to be solved in the current technological field. Summary of the Invention

[0014] To address the aforementioned problems, the present invention aims to provide a method and apparatus for cancer survival prediction based on whole-section pathological images. By screening and structurally modeling the cellular structure information in the whole-section pathological images, the method retains important cellular structural features related to survival prediction while reducing computational complexity, and effectively models the spatial relationships between cells, thereby improving the accuracy and stability of cancer survival prediction.

[0015] To solve the above-mentioned technical problems, the present invention provides the following technical solution: On the one hand, a cancer survival prediction method based on whole-section pathological images is provided, the method comprising the following steps: S1. Obtain whole-section pathological images and patient survival information, and perform preprocessing; S2. Perform cell detection on the whole-section pathological images and represent all cells as cell structure data composed of cell spatial coordinates and cell attribute information; S3. Divide the whole-section pathological image into multiple image regions according to a preset spatial scale, and determine the region of interest by analyzing the differences in cell structure data in different image regions; S4. Filter the cell structure data in each region of interest, select representative cells, and construct a cell subset for subsequent modeling. S5. After obtaining the filtered cell subset, sort the cell structure data according to the spatial distribution relationship of cells and construct the cell structure sequence. S6. Construct a cell structure modeling model, perform feature modeling on cell structure sequences, and obtain tissue structure features that can characterize the microstructure of tumor tissue; S7. Input the tissue structure features output by the cell structure modeling model into the survival prediction model to predict the patient's survival risk.

[0016] Optionally, in step S1, the whole-section pathological image is obtained by high-resolution scanning of the tissue section using a scanning device; the patient survival information includes the patient's survival time and censoring status information. Preprocessing of whole-section pathological images includes: background removal, image standardization, and tissue region identification.

[0017] Optionally, step S2 specifically includes: Cell detection was performed on the preprocessed whole-section pathological images to identify all cells in the images; Obtain the spatial coordinates of each cell in the image; Each cell type is identified based on its morphological characteristics, which include size, shape, and texture features. Based on the spatial coordinates and corresponding attribute information of each cell, all cells in the whole-section pathological image are represented as cell structure data composed of cell spatial coordinates and cell attribute information. The attribute information includes cell type and cell morphology characteristics.

[0018] Optionally, step S3 specifically includes: The whole-section pathological image is divided into multiple image regions according to a preset spatial scale, and each image region contains cell structure data within a certain range; The number of cells, cell types, and spatial distribution characteristics of cells in each image region are statistically analyzed. By analyzing the differences in cell structure data in different image regions, regions with important tissue structure information are identified and determined as regions of interest. The region of interest corresponds to a region where tumor cells are concentrated or where the tumor microenvironment structure is complex.

[0019] Optionally, step S4 specifically includes: After obtaining the region of interest, the cell structure data within each region of interest are screened. During the screening process, the importance of cells is assessed based on cell type, cell distribution in local spatial structure, and spatial relationship between cells. Based on the cell importance assessment results, cells that can represent local tissue structural features are preferentially retained to construct a cell subset for subsequent modeling.

[0020] Optionally, step S5 specifically includes: When constructing cell structure sequences, a sequencer strategy based on space-filling curves is introduced to determine the arrangement order of cell structure data according to the spatial proximity relationship between cells, so that the cell structure sequence can preserve the spatial proximity relationship between cells to the greatest extent.

[0021] Optionally, step S6 specifically includes: A cell structure modeling model is constructed. The modeling process learns the spatial proximity relationships between cells and the impact of different cell structure sequences on cancer survival risk, thereby obtaining tissue structure features that can characterize the microstructure of tumor tissue. Specifically, G representative central cell points are selected from the cell subset; For each central cell point, K-nearest neighbor search is used in two-dimensional coordinate space to select its K nearest neighbor cells, forming G local cell sets. Geometric features are extracted from each local cell set using a PointNet encoder to generate the first tissue structure features; Meanwhile, by projecting the set of central cell points into the same feature space through two layers of multilayer perceptron and the GELU activation function, the second tissue structure feature is obtained.

[0022] Optionally, step S7 specifically includes: The first and second tissue structure features output from the cell structure modeling model are input into the survival prediction model Mamba to predict the patient's survival risk.

[0023] On the other hand, a cancer survival prediction device based on whole-section pathological images is provided for implementing the method described in any of the above embodiments, the device comprising: The acquisition module is used to acquire whole-slice pathological images and patient survival information, and to perform preprocessing. The detection module is used to perform cell detection on whole-section pathological images, representing all cells as cell structure data composed of cell spatial coordinates and cell attribute information; The segmentation module is used to divide the whole-slice pathological image into multiple image regions according to a preset spatial scale, and to determine the region of interest by analyzing the differences in cell structure data in different image regions. The filtering module is used to filter the cell structure data in each region of interest, select representative cells, and construct a cell subset for subsequent modeling. The sorting module is used to sort the cell structure data according to the spatial distribution relationship of cells after obtaining the filtered cell subset, and construct the cell structure sequence. The modeling module is used to construct cell structure modeling models, perform feature modeling on cell structure sequences, and obtain tissue structure features that can characterize the microstructure of tumor tissue. The prediction module is used to input the tissue structure features output by the cell structure modeling model into the survival prediction model to predict the patient's survival risk.

[0024] On the other hand, an electronic device is provided, the electronic device comprising: processor; A memory storing computer-readable instructions, which, when loaded and executed by the processor, implement the steps of the cancer survival prediction method based on whole-slice pathological images as described above.

[0025] On the other hand, a computer-readable storage medium is provided, wherein program code is stored in the computer-readable storage medium, the program code being invoked by a processor to execute the steps of the cancer survival prediction method based on whole-slice pathological images described above.

[0026] The beneficial effects of the technical solution provided by this invention include at least the following: (1) This invention performs regional analysis on the cellular structure information in whole-section pathological images and constructs regions of interest. Based on this, cells are screened, thereby significantly reducing the number of cells involved in modeling while retaining key tissue structure information. Compared with methods that directly model all cells, this invention can effectively reduce computational complexity and storage overhead, and improve the efficiency of model training and prediction processes.

[0027] (2) This invention enables the model to focus more on important cellular structural information related to cancer survival prediction by screening and structuring cellular structure data. By retaining representative cellular structures, the interference of redundant information on the model learning process can be effectively reduced, thereby improving the model's ability to express the microstructural features of tumor tissue.

[0028] (3) This invention constructs a cell structure sequence representation, enabling the originally disordered cell spatial data to be modeled in a structurally ordered form, thereby better preserving the spatial relationship information between cells. By modeling the spatial structural relationship of cells, the complex spatial organizational structure inside tumor tissue can be more accurately depicted, improving the model's ability to express the structure of the tumor microenvironment.

[0029] (4) This invention achieves cancer survival risk prediction based on tissue microstructure information by performing feature modeling on cell structure sequences and combining it with a survival prediction model. Compared with traditional survival prediction methods based on image patch features, this invention can make fuller use of cell structure information in pathological images, thereby improving the accuracy and stability of cancer survival prediction.

[0030] Therefore, the method proposed in this invention can make full use of the cellular structure information in whole-slice pathological images while ensuring computational efficiency, and can more accurately predict the survival risk of cancer patients, thus having good application value. Attached Figure Description

[0031] To more clearly illustrate the technical solutions in the embodiments of the present invention, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0032] Figure 1 This is a flowchart of a cancer survival prediction method based on whole-section pathological images provided in an embodiment of the present invention; Figure 2 This is a schematic diagram illustrating the implementation process of the cancer survival prediction method based on whole-section pathological images provided in this embodiment of the invention. Figure 3 This is a schematic diagram of the structure of the cancer survival prediction device based on whole-slice pathological images provided in an embodiment of the present invention. Detailed Implementation

[0033] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some, not all, of the embodiments of the present invention. All other embodiments obtained by those skilled in the art based on the described embodiments of the present invention without creative effort are within the scope of protection of the present invention.

[0034] This invention provides a cancer survival prediction method based on whole-section pathological images. By extracting, filtering, and structurally modeling cellular structure information from whole-section pathological images, it achieves the prediction of survival risk for cancer patients. (Reference) Figure 1 and Figure 2 As shown, the processing flow of this method may include the following steps: S1. Obtain whole-section pathological images and patient survival information, and perform preprocessing.

[0035] In this step, we first acquire whole-section pathological images and patient survival information. The whole-section pathological images are obtained by high-resolution scanning of the tissue sections using a scanning device, which has high resolution and contains rich tissue structural information; the patient survival information includes the patient's survival time and censoring status information.

[0036] Because whole-section pathological images typically have large image sizes and contain a significant amount of background, image preprocessing is necessary. This preprocessing includes background removal, image normalization, and tissue region identification to obtain pathological image data containing only valid tissue regions, thereby reducing the impact of irrelevant information on subsequent analysis.

[0037] Specifically, the initial dataset can be represented as .in For the number of patients, per sample Defined as a triple . Indicates the first Original whole-section pathological images of the patients. For censored indicator variables, This corresponds to the survival time (in months).

[0038] S2. Perform cell detection on the whole-section pathological images and represent all cells as cell structure data composed of cell spatial coordinates and cell attribute information.

[0039] In this step, cell detection is performed on the preprocessed whole-section pathological image to identify all cells in the image; the spatial coordinates of each cell in the image are obtained; and each cell type is identified based on cell morphological characteristics, including size, shape, and texture features. Through this method, based on the spatial coordinates of each cell and its corresponding attribute information, all cells in the whole-section pathological image can be represented as cell structure data composed of cell spatial coordinates and cell attribute information. The attribute information includes cell type and cell morphological characteristics.

[0040] Specifically, each patient sample is represented as a triple. ,in Indicates inclusion A point cloud of detected cells, For censored indicator variables, This represents the discretized survival tag. Each cell point... From two-dimensional spatial coordinates Category tags (Corresponding to tumor cells, immune cells and other cells respectively) and composed of 18-dimensional morphological or texture features.

[0041] This cellular structure dataset reflects the spatial distribution of tumor cells, immune cells, and other cells in tissues, providing foundational data for subsequent structural modeling.

[0042] S3. Divide the whole-section pathological image into multiple image regions according to a preset spatial scale, and determine the region of interest by analyzing the differences in cell structure data in different image regions.

[0043] Since whole-section pathological images typically contain a large number of cells, directly modeling all cells would result in high computational complexity. Therefore, it is necessary to perform regional analysis on the cell structure data.

[0044] In this step, the whole-section pathological image is first divided into multiple image regions according to a preset spatial scale. Each image region contains cellular structure data within a certain range. Then, the number of cells, cell type composition, and spatial distribution characteristics of cells within each image region are statistically analyzed. By analyzing the differences in cellular structure data across different image regions, regions with important tissue structure information can be identified and designated as regions of interest (ROIs). These ROIs correspond to areas where tumor cells are concentrated or where the tumor microenvironment structure is complex; these areas have higher analytical value in cancer survival prediction.

[0045] Specifically, after obtaining the regional representation of the whole-slice pathological image, the image regions are screened through survival correlation modeling to determine the region of interest. Indicates the patient The initial point cloud data, where each row corresponds to one cell, includes its spatial coordinates, type label, and 18-dimensional morphological and texture features. For each image patch... Calculate the proportion of tumor cells within it; if this proportion exceeds a preset judgment threshold... If the image region is identified as a tumor region, then that region is included within the region of interest. The formula for determining the tumor region is:

[0046] in, Represents image blocks The number of tumor cells in the body Represents image blocks The total number of cells inside, This is a preset threshold (usually 0.8). To identify tumor microenvironment regions, the algorithm iteratively expands from the labeled tumor image region to its eight neighboring regions. If a certain adjacent image region... If a cell meets the preset cell diversity criteria, it is marked as a region of interest. The formula for the diversity determination value is:

[0047] in, Indicates cell type, For image blocks The middle belongs to the type The number of cells, This indicates the total number of cells within the image patch. A preset diversity threshold (e.g., 0.5) is used. This criterion ensures that the selected region of interest exhibits cellular heterogeneity, reflects immune activity, and has prognostic value.

[0048] S4. Filter the cell structure data in each region of interest, select representative cells, and construct a cell subset for subsequent modeling.

[0049] In this step, after obtaining the region of interest, the cell structure data within each region of interest are screened. During the screening process, the importance of cells is evaluated based on cell type, cell distribution in local spatial structure, and spatial relationship between cells. Based on the cell importance evaluation results, cells that can represent the characteristics of local tissue structure are preferentially retained to construct a cell subset for subsequent modeling.

[0050] This screening process can significantly reduce the number of cells while retaining cellular structural information that plays an important role in cancer survival prediction, thereby reducing computational complexity while ensuring that key structural information is not lost.

[0051] Specifically, after identifying the region of interest, the corresponding cells are extracted to construct a dimensionality-reduced point cloud representation. For patients The collection of all cells within a region of interest. Within each region of interest, a representative cell is selected for each cell type (tumor cells, immune cells, stromal cells); this cell is the one closest to the geometric center of the image region.

[0052] in, Indicates in the image region The closest to the geometric center The types are The strategy significantly reduces cell number while still preserving key biologically significant signals.

[0053] S5. After obtaining the filtered cell subset, sort the cell structure data according to the spatial distribution of cells and construct the cell structure sequence.

[0054] After obtaining the filtered subset of cells, the cell structure data needs to be represented in a structured manner. This invention sorts the cells according to their spatial distribution and constructs a cell structure sequence representation.

[0055] In this step, a sequencer strategy based on space-filling curves is introduced when constructing cell structure sequences. The order of cell structure data is determined according to the spatial proximity between cells, ensuring that the cell structure sequences preserve these spatial proximity relationships to the greatest extent possible. This step transforms the originally disordered cell spatial structure data into cell structure sequence data with a structural order, enabling subsequent models to more effectively learn the spatial relationships between cells.

[0056] Specifically, a sequencer strategy based on space-filling curves is introduced to achieve the sorting of cell structure data. Space-filling curves can map points in multidimensional space to a one-dimensional sequence while preserving spatial proximity to a certain extent. Among them, Hilbert curves are widely used in point cloud and spatial data sequencer tasks due to their good locality preservation properties. In addition, to further enhance the model's robustness to spatial structure, a Trans-Hilbert sequencer path is introduced as a supplement. By transforming the original spatial coordinates before performing Hilbert sorting, a sequence perspective different from the original Hilbert path can be generated. Finally, the structural features of cell point clouds are learned from different spatial traversal orders through a dual-path sequencer mechanism, thereby reducing the structural bias caused by a single sequence order.

[0057] S6. Construct a cell structure modeling model, perform feature modeling on cell structure sequences, and obtain tissue structure features that can characterize the microstructure of tumor tissue.

[0058] After constructing the cell structure sequence, a cell structure modeling model is built. The modeling process learns the spatial proximity relationships between cells and the impact of different cell structure sequences on cancer survival risk, obtaining tissue structure features that characterize the microstructure of tumor tissue. By modeling the cell structure sequence, the spatial dependencies between cells can be effectively captured, thus more accurately describing the relationship between tumor tissue structure and patient survival risk.

[0059] Specifically, from cell subsets Selected from A representative central cell point:

[0060] in, Indicates the first Each center point contains its two-dimensional spatial coordinates. Cell type tags (representing tumor cells, immune cells, and stromal cells respectively), and 18-dimensional morphological or textural features.

[0061] For each central cell point In a two-dimensional coordinate space, a K-nearest neighbor search is used to select the closest neighbor. Each neighboring cell constitutes A local collection of cells:

[0062] Among them, the A local cell collection Indicates the center point Centered The nearest neighbor cell, The feature structure is the same as the center point, which is also a 21-dimensional vector.

[0063] In the feature embedding stage, the PointNet encoder independently encodes each point using a shared multilayer perceptron, enabling the extraction of point-level features regardless of point order. It exhibits good permutation invariance and is suitable for the unordered structure of cell point clouds. Through the PointNet encoder, each cell is mapped to a high-dimensional feature vector to characterize its comprehensive attribute representation.

[0064] Furthermore, the PointNet encoder is used to extract geometric features from each local cell set, generating an initial neighborhood-level feature representation, i.e., the first tissue structure feature:

[0065] Simultaneously, the central cell point set is obtained through two layers of multilayer perceptron and the GELU activation function. Projecting onto the same feature space yields the second organizational structure feature:

[0066] S7. Input the tissue structure features output by the cell structure modeling model into the survival prediction model to predict the patient's survival risk.

[0067] In this step, the first tissue structure feature output by the cell structure modeling model is... Second organizational structure characteristics Input the survival prediction model Mamba to predict the patient's survival risk.

[0068] Survival prediction models assess a patient's future survival risk based on the relationship between tissue structure characteristics and patient survival information, and output corresponding survival probabilities or risk scores. This method can predict the prognostic risk of cancer patients, thereby providing clinicians with auxiliary decision-making information and improving the accuracy of cancer prognostic assessment.

[0069] In this embodiment of the invention, cellular structure information from whole-section pathological images is utilized to predict the survival risk of cancer patients through regional analysis of cell structures, screening of key cells, and structured modeling. Specifically, this includes: (1) This invention proposes a cell structure modeling method based on whole-section pathological images. By extracting cell spatial location and cell attribute information from whole-section pathological images, cell structure data that can reflect the microstructural characteristics of tumor tissue is constructed, providing a structured data basis for subsequent survival prediction.

[0070] (2) This invention divides the whole-section pathological image into regions and constructs regions of interest based on the number, type and spatial distribution characteristics of cells in the region, thereby identifying regions containing key tissue structure information and providing a basis for subsequent cell screening.

[0071] (3) The present invention screens cells in the region of interest and selects representative cells to form a cell subset based on cell category information and cell distribution in the spatial structure, thereby significantly reducing the number of cells while retaining important cell structure information related to cancer survival prediction.

[0072] (4) This invention presents a structured representation of the screened cell structure data and constructs a cell structure sequence based on the spatial proximity relationship between cells, so that the originally disordered cell spatial data can be modeled in a structurally ordered form, thereby better preserving the spatial relationship information between cells.

[0073] Compared with existing survival prediction methods based on image patch features, this invention models the tumor tissue structure at the cellular level, thereby making fuller use of the microstructural information of the tumor tissue and improving the accuracy and stability of survival prediction.

[0074] Accordingly, embodiments of the present invention also provide a cancer survival prediction device based on whole-section pathological images, such as... Figure 3 As shown, the device includes: The acquisition module 201 is used to acquire whole-slice pathological images and patient survival information, and to perform preprocessing. The detection module 202 is used to perform cell detection on whole-slice pathological images and represent all cells as cell structure data composed of cell spatial coordinates and cell attribute information. The segmentation module 203 is used to divide the whole-slice pathological image into multiple image regions according to a preset spatial scale, and to determine the region of interest by analyzing the differences in cell structure data in different image regions. The filtering module 204 is used to filter the cell structure data in each region of interest, select representative cells, and construct a cell subset for subsequent modeling. The sorting module 205 is used to sort the cell structure data according to the spatial distribution relationship of cells after obtaining the filtered cell subset, and construct the cell structure sequence. Modeling module 206 is used to construct a cell structure modeling model, perform feature modeling on cell structure sequences, and obtain tissue structure features that can characterize the microstructure of tumor tissue. The prediction module 207 is used to input the tissue structure features output by the cell structure modeling model into the survival prediction model to predict the patient's survival risk.

[0075] For ease of explanation, Figure 3 Only the main components of the device are shown. The device of this embodiment can be used to perform... Figure 1 The technical solutions of the method embodiments shown are similar in principle and in effect, and will not be described again here.

[0076] In an exemplary embodiment, the present invention also provides an electronic device, the electronic device comprising: processor; A memory storing computer-readable instructions, which, when loaded and executed by the processor, implement the steps of the cancer survival prediction method based on whole-slice pathological images as described above.

[0077] In an exemplary embodiment, the present invention also provides a computer-readable storage medium storing at least one instruction, which is loaded and executed by a processor to implement the steps of the cancer survival prediction method based on whole-slice pathological images described above. For example, the computer-readable storage medium may be a ROM, random access memory (RAM), CD-ROM, magnetic tape, floppy disk, or optical data storage device, etc.

[0078] It should be noted that, in this document, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or terminal device that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or terminal device. Unless otherwise specified, an element defined by the phrase "comprising..." does not exclude the presence of other identical elements in the process, method, article, or terminal device that includes said element.

[0079] The use of terms such as "an embodiment," "an embodiment," "an exemplary embodiment," and "some embodiments" in the specification indicates that the described embodiment may include a specific feature, structure, or characteristic, but not every embodiment necessarily includes that specific feature, structure, or characteristic. Furthermore, when a specific feature, structure, or characteristic is described in connection with an embodiment, implementing such a feature, structure, or characteristic in conjunction with other embodiments (whether explicitly described or not) should be within the knowledge of those skilled in the art.

[0080] It should be understood that, in various embodiments of the present invention, the order of the above-mentioned process numbers does not imply the order of execution. The execution order of each process should be determined by its function and internal logic, and should not constitute any limitation on the implementation process of the embodiments of the present invention.

[0081] In the embodiments provided by this invention, it should be understood that the disclosed devices, apparatuses, and methods can be implemented in other ways. For example, the apparatus embodiments described above are merely illustrative. For instance, the division of units is only a logical functional division, and in actual implementation, there may be other division methods. For example, multiple units or components may be combined or integrated into another device, or some features may be ignored or not executed. Furthermore, the coupling or direct coupling or communication connection shown or discussed may be through some interfaces; the indirect coupling or communication connection between devices or units may be electrical, mechanical, or other forms.

[0082] The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the units can be selected to achieve the purpose of this embodiment according to actual needs.

[0083] In addition, the functional units in the various embodiments of the present invention can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit.

[0084] If the aforementioned functions are implemented as software functional units and sold or used as independent products, they can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of this invention, or the part that contributes to the prior art, or a part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of this invention. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0085] This invention encompasses any substitutions, modifications, equivalent methods, and solutions made within the spirit and scope of this invention. To provide the public with a thorough understanding of this invention, specific details are described in detail in the following preferred embodiments; however, those skilled in the art will fully understand the invention even without these details. Furthermore, to avoid unnecessary misunderstanding of the essence of this invention, well-known methods, processes, procedures, components, and circuits are not described in detail.

[0086] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A cancer survival prediction method based on whole-section pathological images, characterized in that, Includes the following steps: S1. Obtain whole-section pathological images and patient survival information, and perform preprocessing; S2. Perform cell detection on the whole-section pathological images and represent all cells as cell structure data composed of cell spatial coordinates and cell attribute information; S3. Divide the whole-section pathological image into multiple image regions according to a preset spatial scale, and determine the region of interest by analyzing the differences in cell structure data in different image regions; S4. Filter the cell structure data in each region of interest, select representative cells, and construct a cell subset for subsequent modeling. S5. After obtaining the filtered cell subset, sort the cell structure data according to the spatial distribution relationship of cells and construct the cell structure sequence. S6. Construct a cell structure modeling model, perform feature modeling on cell structure sequences, and obtain tissue structure features that can characterize the microstructure of tumor tissue; S7. Input the tissue structure features output by the cell structure modeling model into the survival prediction model to predict the patient's survival risk.

2. The cancer survival prediction method based on whole-section pathological images according to claim 1, characterized in that, In step S1, the whole-section pathological image is obtained by high-resolution scanning of the tissue section using a scanning device; the patient survival information includes the patient's survival time and censoring status information. Preprocessing of whole-section pathological images includes: background removal, image standardization, and tissue region identification.

3. The cancer survival prediction method based on whole-section pathological images according to claim 1, characterized in that, Step S2 specifically includes: Cell detection was performed on the preprocessed whole-section pathological images to identify all cells in the images; Obtain the spatial coordinates of each cell in the image; Each cell type is identified based on its morphological characteristics, which include size, shape, and texture features. Based on the spatial coordinates and corresponding attribute information of each cell, all cells in the whole-section pathological image are represented as cell structure data composed of cell spatial coordinates and cell attribute information. The attribute information includes cell type and cell morphology characteristics.

4. The cancer survival prediction method based on whole-section pathological images according to claim 1, characterized in that, Step S3 specifically includes: The whole-section pathological image is divided into multiple image regions according to a preset spatial scale, and each image region contains cell structure data within a certain range; The number of cells, cell types, and spatial distribution characteristics of cells in each image region are statistically analyzed. By analyzing the differences in cell structure data in different image regions, regions with important tissue structure information are identified and determined as regions of interest. The region of interest corresponds to a region where tumor cells are concentrated or where the tumor microenvironment structure is complex.

5. The cancer survival prediction method based on whole-section pathological images according to claim 1, characterized in that, Step S4 specifically includes: After obtaining the region of interest, the cell structure data within each region of interest are screened. During the screening process, the importance of cells is assessed based on cell type, cell distribution in local spatial structure, and spatial relationship between cells. Based on the cell importance assessment results, cells that can represent local tissue structural features are preferentially retained to construct a cell subset for subsequent modeling.

6. The cancer survival prediction method based on whole-section pathological images according to claim 1, characterized in that, Step S5 specifically includes: When constructing cell structure sequences, a sequencer strategy based on space-filling curves is introduced to determine the arrangement order of cell structure data according to the spatial proximity relationship between cells, so that the cell structure sequence can preserve the spatial proximity relationship between cells to the greatest extent.

7. The cancer survival prediction method based on whole-section pathological images according to claim 1, characterized in that, Step S6 specifically includes: A cell structure modeling model is constructed. The modeling process learns the spatial proximity relationships between cells and the impact of different cell structure sequences on cancer survival risk, thereby obtaining tissue structure features that can characterize the microstructure of tumor tissue. Specifically, G representative central cell points are selected from the cell subset; For each central cell point, K-nearest neighbor search is used in two-dimensional coordinate space to select its K nearest neighbor cells, forming G local cell sets. Geometric features are extracted from each local cell set using a PointNet encoder to generate the first tissue structure features; Meanwhile, by projecting the set of central cell points into the same feature space through two layers of multilayer perceptron and the GELU activation function, the second tissue structure feature is obtained.

8. The cancer survival prediction method based on whole-section pathological images according to claim 7, characterized in that, Step S7 specifically includes: The first and second tissue structure features output from the cell structure modeling model are input into the survival prediction model Mamba to predict the patient's survival risk.

9. A cancer survival prediction device based on whole-section pathological images, the device being used to implement the method as described in any one of claims 1 to 8, characterized in that, The device includes: The acquisition module is used to acquire whole-slice pathological images and patient survival information, and to perform preprocessing. The detection module is used to perform cell detection on whole-section pathological images, representing all cells as cell structure data composed of cell spatial coordinates and cell attribute information; The segmentation module is used to divide the whole-slice pathological image into multiple image regions according to a preset spatial scale, and to determine the region of interest by analyzing the differences in cell structure data in different image regions. The filtering module is used to filter the cell structure data in each region of interest, select representative cells, and construct a cell subset for subsequent modeling. The sorting module is used to sort the cell structure data according to the spatial distribution relationship of cells after obtaining the filtered cell subset, and construct the cell structure sequence. The modeling module is used to construct cell structure modeling models, perform feature modeling on cell structure sequences, and obtain tissue structure features that can characterize the microstructure of tumor tissue. The prediction module is used to input the tissue structure features output by the cell structure modeling model into the survival prediction model to predict the patient's survival risk.

10. An electronic device, characterized in that, The electronic device includes: processor; A memory storing computer-readable instructions that, when loaded and executed by the processor, implement the method as described in any one of claims 1 to 8.

Images