Tumor cell statistical method, device, equipment and medium

Abstract

The invention relates to the technical field of tumor cell statistics. The tumor cell statistical method, device, equipment and medium are provided, and the method comprises the following steps: based on mark information in a digital pathological image, performing tumor core region segmentation and invasion leading edge boundary identification processing, and generating tumor invasion leading edge boundary coordinates; based on the tumor infringement leading edge boundary coordinates and the marking information, performing multi-dimensional quantification processing on tumor cell space distribution characteristics to generate infringement mode characteristic vectors; performing multi-dimensional weighted aggregation processing on the invasion mode feature vector to generate a tumor invasion ability quantitative index; and performing comprehensive evaluation report generation processing based on the tumor invasion capability quantitative index to obtain a quantitative evaluation report so as to improve the spatial heterogeneity characterization capability, enhance the multi-scale spatial modeling precision, optimize the clinical interpretability of the quantitative index and realize the integration of structured report output.

Description

Technical Field

[0001] This invention relates to the field of tumor cell statistical technology, and in particular to tumor cell statistical methods, devices, equipment and media. Background Technology

[0002] With the rapid development of digital pathology technology, the quantitative assessment of tumor invasiveness is playing an increasingly crucial role in precision cancer diagnosis and treatment. Effective quantification of invasiveness can not only predict the risk of metastasis but also guide the development of individualized treatment plans, becoming an important breakthrough in tumor pathology research.

[0003] Traditional techniques rely on manual microscopic observation to quantify tumor cells, but ignore spatial heterogeneity and fail to reflect the true invasion pattern. While existing automated algorithms can identify tumor boundaries, they lack the ability to systematically model multi-scale spatial features. Furthermore, traditional quantitative indicators generate scores through simple aggregation without incorporating the weight differences of biological invasion pathways, resulting in insufficient clinical interpretability. Moreover, the numerical values ​​and visualization results output by existing technologies are isolated from each other and not integrated into a structured clinical report, thus failing to directly support diagnostic and treatment decisions. Summary of the Invention

[0004] Therefore, it is necessary to provide tumor cell statistical methods, devices, equipment and media to address the above-mentioned technical problems, so as to improve the ability to characterize spatial heterogeneity, enhance the accuracy of multi-scale spatial modeling, optimize the clinical interpretability of quantitative indicators, and achieve the integration of structured report output.

[0005] Firstly, this application provides a method for statistical analysis of tumor cells, the method comprising:

[0006] Biological samples containing the tumor-normal tissue interface are specifically stained using immunohistochemistry or multiplex fluorescent labeling techniques, and digital pathological images are generated by image acquisition and processing using a preset scanning device.

[0007] Based on the labeled information in digital pathological images, tumor core region segmentation and invasion front boundary identification are performed to generate tumor invasion front boundary coordinates.

[0008] Based on the boundary coordinates and labeling information of the tumor invasion front, multi-dimensional quantitative processing of the spatial distribution characteristics of tumor cells is performed to generate an invasion pattern feature vector.

[0009] Multi-dimensional weighted aggregation of invasion pattern feature vectors is performed to generate quantitative indicators of tumor invasion capability.

[0010] A comprehensive assessment report is generated based on quantitative indicators of tumor invasion ability, resulting in a quantitative assessment report.

[0011] In one embodiment, based on the coordinates and marker information of the tumor invasion front boundary, multi-dimensional quantification of the spatial distribution characteristics of tumor cells is performed to generate an invasion pattern feature vector, including:

[0012] A band-shaped analysis region is created based on the coordinates of the tumor invasion front boundary, generating the front analysis band;

[0013] The density of tumor cells within the leading edge analysis zone is statistically processed to generate leading edge cell density values.

[0014] A vertical depth partitioning model is generated by establishing vertical depth partitioning based on the coordinates of the tumor invasion front boundary.

[0015] Based on the depth partitioning model, statistical processing of cell distribution in different depth intervals is performed to generate cell density-depth distribution curves.

[0016] Spatial range constraint processing is performed based on the coordinates of the tumor invasion front boundary, and vascular structure recognition processing is performed based on the labeled information to generate a vascular coordinate dataset.

[0017] A three-dimensional buffer region is created around the blood vessel coordinate dataset to generate a microenvironment model around the blood vessels;

[0018] By combining tumor cell localization data from digital pathology images, quantitative analysis of tumor cells within the perivascular microenvironment model is performed to generate a vascular invasion index.

[0019] Multimodal feature fusion processing is performed on the front cell density value, cell density-depth distribution curve and vascular invasion index to generate invasion mode feature vector.

[0020] In one embodiment, statistical processing of cell distribution in different depth ranges is performed based on a depth partitioning model to generate cell density-depth distribution curves, including:

[0021] Based on the tissue density gradient change characteristics in the deep partitioning model, dynamic interval boundary optimization is performed to generate optimized deep partitions.

[0022] Tumor cell density is calculated for each depth interval in the optimized depth partition to generate a discrete depth-density data point set.

[0023] The following formula is used to perform continuous fitting of biological tissue based on a discrete depth-density data point set, generating a cell density-depth distribution curve:

[0024]

[0025] in, This represents the cell density-depth distribution curve. Represents depth coordinates, The Dirac function represents the location of discrete data points. Indicates the first Tumor cell density values ​​in each depth range Indicates the total number of depth partitions. Operators representing the continuity constraints of biological tissues This represents the Gaussian smoothing kernel function. This represents the convolution operator.

[0026] In one embodiment, a three-dimensional buffer region is created around the blood vessel coordinate dataset to generate a microenvironment model around the blood vessels, including:

[0027] Based on the blood vessel coordinate dataset, the diameter distribution and orientation features of the blood vessel are extracted to generate blood vessel morphology feature vectors.

[0028] Multi-scale spatial weight allocation processing is performed based on vascular morphology feature vectors to generate an adaptive three-dimensional buffer region.

[0029] Using the following formula, combined with tumor cell localization data from digital pathology images, a model of the tumor cell-vascular spatial interaction relationship is generated for an adaptive three-dimensional buffer region to produce a perivascular microenvironment model:

[0030]

[0031] in, This represents a model of the perivascular microenvironment. Indicates the coordinates of the centerline of the target blood vessel. Indicates blood vessels An adaptive three-dimensional buffer region Represents three-dimensional spatial coordinates. Indicates position Tumor cell density at the site, Represents the spatial weighting function. Indicates position to blood vessels European distance, Indicates position Compared to blood vessels azimuth angle, This represents the angle-sensitive function.

[0032] In one embodiment, based on the marker information in the digital pathological image, tumor core region segmentation and invasion front boundary identification are performed to generate tumor invasion front boundary coordinates, including:

[0033] Based on the tumor cell-specific expression characteristics in the labeled information, multi-channel feature fusion processing is performed to generate a tumor cell probability heatmap.

[0034] Topological connectivity analysis is performed on the probability heatmap of tumor cells to generate the boundary of the tumor core region;

[0035] Density gradient field calculation is performed based on the boundary of the tumor core region to generate a tumor invasion direction vector field.

[0036] Adaptive boundary tracking is performed along the vector field of tumor invasion direction to generate the coordinates of the tumor invasion front boundary.

[0037] In one embodiment, topological connectivity analysis is performed on the probability heatmap of tumor cells to generate the boundary of the tumor core region, including:

[0038] Homology group persistence calculation is performed based on tumor cell probability heatmap to generate a topological feature persistence graph;

[0039] Feature filtering based on persistent topological feature graphs for pathological constraints is performed to generate a set of target connected components.

[0040] Boundary optimization processing with tissue continuity constraints is performed on the target connected component set to generate the boundary of the tumor core region.

[0041] In one embodiment, the boundary of the tumor core region is obtained using the following formula:

[0042]

[0043] in, Indicates the boundary of the tumor core region. Indicates the first Each target connected component Indicates the total number of target connected components. Indicates boundary With connected components The intersection measure, This represents the boundary smoothness regularization term. This represents the weighting factor for organizational continuity constraints. Denotes the candidate boundary variables to be solved. Indicates the number of the target connected component.

[0044] Secondly, this application also provides a tumor cell counting device, which includes:

[0045] The sample staining and imaging module is used to specifically stain biological samples containing the tumor-normal tissue interface using immunohistochemistry or multiplex fluorescent labeling technology, and to generate digital pathological images by image acquisition and processing through a preset scanning device.

[0046] The tumor boundary delineation module is used to segment the tumor core region and identify the invasion front boundary based on the marker information in the digital pathology image, and generate the coordinates of the tumor invasion front boundary.

[0047] The spatial feature quantization module is used to perform multi-dimensional quantification of the spatial distribution features of tumor cells based on the coordinates and labeling information of the tumor invasion front boundary, and generate an invasion pattern feature vector.

[0048] The invasion capability aggregation module is used to perform multi-dimensional weighted aggregation of invasion pattern feature vectors to generate quantitative indicators of tumor invasion capability.

[0049] The clinical assessment output module is used to generate a comprehensive assessment report based on quantitative indicators of tumor invasiveness, resulting in a quantitative assessment report.

[0050] Thirdly, this application also provides a computer device, including a memory and a processor, wherein the memory stores a computer program, and the processor executes the computer program to implement the steps of any of the methods in the first aspect of this application.

[0051] Fourthly, this application also provides a computer-readable storage medium having a computer program stored thereon, wherein the computer program, when executed by a processor, implements the steps of any of the methods in the first aspect of this application.

[0052] The tumor cell statistical methods, devices, equipment, and media provided in this application include: specifically staining biological samples containing the tumor-normal tissue interface using immunohistochemistry or multiplex fluorescent labeling technology, and generating digital pathological images by acquiring them through a preset scanning device, ensuring the complete preservation of the association information between the tumor and normal tissue; completing the segmentation of the tumor core region and the identification of the boundary coordinates of the invasion front based on the labeling information in the image, providing a spatial positioning basis for capturing the distribution differences of tumor cells in different regions; and then extracting the spatial distribution characteristics of tumor cells and forming an invasion pattern feature vector through multi-dimensional quantitative processing, effectively covering multi-scale spatial information from local density to depth distribution and vascular microenvironment interaction, significantly improving the ability to characterize spatial heterogeneity and the accuracy of multi-scale spatial modeling.

[0053] The multi-dimensional weighted aggregation of invasion pattern feature vectors fully considers the weight differences of biological invasion pathways, making the generated quantitative indicators of tumor invasion ability more consistent with clinical pathological mechanisms and optimizing the clinical interpretability of the indicators. Through comprehensive evaluation report generation, the quantitative indicators are integrated with the analysis logic system to form a structured quantitative evaluation report, achieving efficient connection between analysis results and clinical application scenarios. Attached Figure Description

[0054] To more clearly illustrate the technical solutions in the embodiments or related technologies of this application, the accompanying drawings used in the description of the embodiments or related technologies will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0055] Figure 1 This is a flowchart of a tumor cell counting method in one embodiment of the present invention;

[0056] Figure 2 This is a flowchart illustrating the process of performing topological connectivity analysis on a probability heatmap of tumor cells to generate the boundary of the tumor core region in one embodiment of the present invention.

[0057] Figure 3 This is a structural diagram of a tumor cell counting device according to one embodiment of the present invention. Detailed Implementation

[0058] To make the above-mentioned objects, features, and advantages of this application more apparent and understandable, the specific embodiments of this application will be described in detail below with reference to the accompanying drawings. Many specific details are set forth in the following description to provide a thorough understanding of this application. However, this application can be implemented in many other ways different from those described herein, and those skilled in the art can make similar modifications without departing from the spirit of the application. Therefore, this application is not limited to the specific embodiments disclosed below.

[0059] First, the application scenarios of the embodiments of this application are described. In the embodiments of this application, methods, devices, equipment, and media for tumor cell statistics are provided, applicable to but not limited to scenarios such as quantitative assessment of tumor invasiveness in precision cancer diagnosis and treatment, analysis of tumor cell spatial characteristics in digital pathology diagnosis, and assessment of tumor biological behavior in personalized treatment planning.

[0060] In illustrative purposes, the tumor cell statistical methods, devices, equipment, and media provided in the embodiments of this application can also be applied to other scenarios such as the quantification of cell distribution characteristics in tumor prognosis risk prediction, the standardized tumor cell statistical scenarios in multi-center pathological data sharing, and the tumor invasion pattern analysis scenarios in clinical pathology research. These are merely illustrative examples and do not limit the specific application scenarios.

[0061] like Figure 1 As shown, this application provides a tumor cell statistical method, which includes:

[0062] S101: Specific staining of biological samples containing the tumor-normal tissue interface is performed using immunohistochemistry or multiplex fluorescent labeling techniques, and digital pathological images are generated through image acquisition and processing using a preset scanning device.

[0063] For example, immunohistochemistry or multiplex fluorescent labeling techniques are used to target and bind to biomarkers specifically expressed by tumor cells in biological samples containing the tumor-normal tissue interface, enabling tumor cells to form clear visual distinctions from normal tissue cells. Using a scanning device that meets preset technical standards, a full-field-of-view image capture operation is performed on the specifically stained biological sample. The acquired optical information is processed through the device's built-in signal conversion and data calibration processes to generate digital pathological images.

[0064] S102: Based on the marker information in the digital pathological image, the tumor core region is segmented and the invasion front boundary is identified to generate the coordinates of the tumor invasion front boundary.

[0065] For example, relying on the tumor cell-specific expression features and tissue differentiation identifiers contained in the labeling information in digital pathological images, feature extraction and signal enhancement processing are performed on the labeling information to screen tumor tissue attribute regions. The tumor core region is segmented through tissue continuity and morphological constraints to eliminate interference from non-tumor-related regions. Then, for the boundary between the tumor core region and the surrounding normal tissue, multi-dimensional boundary detection and verification are performed in combination with the cell density gradient changes and tissue boundary features reflected by the labeling information to locate the tumor invasion front position. After spatial coordinate calibration and boundary contour regularization processing, the boundary coordinates of the tumor invasion front are generated.

[0066] S103: Based on the boundary coordinates and labeling information of the tumor invasion front, multi-dimensional quantitative processing of the spatial distribution characteristics of tumor cells is performed to generate an invasion pattern feature vector.

[0067] For example, the core analysis space is defined by the coordinates of the tumor invasion front boundary. Tumor cells in digital pathological images are identified by combining the label information. Based on the spatial distribution of tumor cells within the defined range, features are extracted and quantified from multiple dimensions such as cell density distribution, deep correlation features, and spatial aggregation patterns. The quantification results of each dimension are integrated and feature regularization is performed to generate an invasion pattern feature vector.

[0068] S104: Perform multi-dimensional weighted aggregation on the feature vector of the invasion pattern to generate a quantitative index of tumor invasion capability.

[0069] For example, by combining clinical pathological mechanisms and biological invasion patterns, corresponding biological weight coefficients are assigned to each dimension of the invasion pattern feature vector. The quantitative results of each dimension of the invasion pattern feature vector are then weighted according to the weight coefficients. The weighted multi-dimensional feature results are then systematically aggregated and integrated to eliminate redundant interference between dimensions and enhance the comprehensive expression of core features. The representation form of the aggregated results is standardized through standardization processing to generate a quantitative index of tumor invasion capability.

[0070] S105: Based on the quantitative indicators of tumor invasion ability, a comprehensive assessment report is generated and processed to obtain a quantitative assessment report.

[0071] For example, relying on the quantitative indicators of tumor invasiveness, the characteristic correlation information and quantitative dimension descriptions in the analysis process are integrated. Combined with the standardized presentation specifications of clinical pathology reports, the core assessment data, characteristic interpretation basis and biological significance related to tumor invasiveness are classified and sorted out. A structured report framework covering assessment conclusions, indicator sub-items and data support elements is constructed. Through information hierarchy regularization and expression standardization, the logicality and clinical readability of the report content are ensured. After multi-dimensional information verification and format unification adjustment, a quantitative assessment report is obtained.

[0072] An embodiment of this application provides a tumor cell statistical method, which includes: specifically staining biological samples containing the tumor-normal tissue interface using immunohistochemistry or multiplex fluorescent labeling technology, and generating digital pathological images by acquiring them through a preset scanning device. This ensures the complete preservation of the association information between the tumor and normal tissue. Based on the labeling information in the image, the tumor core region is segmented and the boundary coordinates of the invasion front are identified, providing a spatial positioning basis for capturing the distribution differences of tumor cells in different regions. Furthermore, through multi-dimensional quantitative processing, the spatial distribution characteristics of tumor cells are extracted and an invasion pattern feature vector is formed. This effectively covers multi-scale spatial information from local density to depth distribution and vascular microenvironment interaction, significantly improving the ability to characterize spatial heterogeneity and the accuracy of multi-scale spatial modeling.

[0073] The multi-dimensional weighted aggregation of invasion pattern feature vectors fully considers the weight differences of biological invasion pathways, making the generated quantitative indicators of tumor invasion ability more consistent with clinical pathological mechanisms and optimizing the clinical interpretability of the indicators. Through comprehensive evaluation report generation, the quantitative indicators are integrated with the analysis logic system to form a structured quantitative evaluation report, achieving efficient connection between analysis results and clinical application scenarios.

[0074] In one embodiment, based on the coordinates and marker information of the tumor invasion front boundary, multi-dimensional quantification of the spatial distribution characteristics of tumor cells is performed to generate an invasion pattern feature vector, including:

[0075] (1) Create a band-shaped analysis region based on the boundary coordinates of the tumor invasion front and generate the front analysis band.

[0076] For example, based on the spatial distribution pattern of the boundary coordinates of the tumor invasion front, combined with the tissue extension characteristics of tumor invasion, and referring to the associated range of tumor cell spread, the width standard of the band analysis region is determined, the length range of the band analysis region is determined according to the complete extension trajectory of the tumor invasion front, the boundary outline of the band analysis region is defined by spatial topology delineation rules, the integrity and rationality of the region coverage are verified, the standardized creation of the band analysis region is completed, and the front analysis band is generated.

[0077] The frontal analysis zone includes a strip-shaped spatial region that extends along the coordinates of the tumor invasion frontal boundary, covering the core area of ​​the tumor invasion front and the surrounding potentially invaded related tissue areas.

[0078] (2) Perform density statistical processing on the tumor cells in the front analysis zone to generate front cell density values.

[0079] For example, relying on the specific expression signals unique to tumor cells in the labeling information, tumor cells in the leading edge analysis zone are distinguished from normal tissue cells and non-specific marker interferences. The spatial position of each tumor cell in the leading edge analysis zone is located by a cell recognition algorithm. Invalid data such as overlapping recognition and false positive recognition are excluded. The actual number of effective tumor cells in the leading edge analysis zone is counted. The actual tissue space area of ​​the leading edge analysis zone is calculated. This area is deducted from the non-tissue interstitial area. The correspondence between the number of effective tumor cells and the actual tissue space area is calculated according to the density quantification specification. The tumor cell density statistical processing is completed, and the leading edge cell density value is generated.

[0080] Among them, the frontal cell density value includes the quantitative result of the ratio of the number of effective tumor cells in the frontal analysis zone to the actual tissue space area of ​​the frontal analysis zone.

[0081] (3) Establish vertical depth partitions based on the coordinates of the tumor invasion front boundary to generate a depth partition model.

[0082] For example, taking each spatial point of the tumor invasion front boundary as the starting reference, the extension direction of the partition is clearly defined along the vertical direction of the tumor invasion into normal tissue. The partitioning interval rules are set in combination with the histological invasion hierarchy. The thickness parameters of each partition are adjusted according to the structural density of different tissue types to ensure that the partition boundary matches the actual tissue hierarchy. The coordinate system of each partition is unified through spatial coordinate calibration, and a vertical depth partition with clear hierarchy and well-defined boundaries is constructed. The systematic establishment of vertical depth partition is completed, and a depth partition model is generated.

[0083] The deep partitioning model includes a multi-level vertical deep partitioning system constructed based on the boundary coordinates of the tumor invasion front. It covers tissue partitions corresponding to different invasion depths extending from the tumor invasion front to normal tissue. The histological invasion hierarchy referenced by the deep partitioning model includes the mucosa, submucosa, and muscularis propria.

[0084] (4) Based on the depth partitioning model, perform statistical processing of cell distribution in different depth intervals to generate cell density-depth distribution curves.

[0085] For example, relying on the signal intensity characteristics and specific expression identifiers of the labeled information, tumor cells are identified one by one in each different depth interval of the deep partitioning model, while interference from normal cells and non-specific markers is excluded. The actual number of effective tumor cells in each depth interval is counted, the actual tissue space area of ​​each depth interval is measured, and the ratio of the number of effective tumor cells to the corresponding actual tissue space area in each depth interval is calculated to obtain the tumor cell density of each depth interval. All tumor cell density data are sorted and organized according to the hierarchical order of the depth intervals, and the discrete density data is transformed into a continuous feature curve through curve fitting rules to complete the generation of the cell density-depth distribution curve.

[0086] Among them, the cell density-depth distribution curve includes a continuous visual feature curve that reflects the correlation between different depth intervals and the corresponding tumor cell density.

[0087] (5) Spatial range constraint processing is performed based on the boundary coordinates of the tumor invasion front, and vascular structure recognition processing is performed based on the label information to generate a vascular coordinate dataset.

[0088] For example, using the boundary coordinates of the tumor invasion front as the core reference, and combining the maximum potential range of tumor invasion to set a spatial constraint threshold, a core analysis space including the tumor invasion front and the surrounding potentially affected areas is delineated. Irrelevant tissue areas exceeding this range are excluded through spatial range screening rules. At the same time, relying on the unique biomarker expression characteristics of vascular structures in the labeled information, tubular structure detection algorithms and continuous contour tracking technology are used to identify vascular structures. The continuity, integrity, and morphological rationality of the identified vascular structures are verified. The spatial coordinates of key points of each vascular structure are recorded, and the coordinate information is systematically organized according to the type and number of vascular structures to generate a vascular coordinate dataset.

[0089] The vascular coordinate dataset includes a set of key point spatial coordinates of all validated vascular structures within the associated spatial range of the tumor invasion front. The key point spatial coordinates of the vascular structures include centerline coordinates, vessel wall contour coordinates, etc.

[0090] (6) Create a three-dimensional buffer region around the blood vessel coordinate dataset to generate a microenvironment model around the blood vessels.

[0091] For example, the spatial orientation, diameter, and curvature of each blood vessel structure in the blood vessel coordinate dataset are analyzed. Differentiated three-dimensional buffer zone expansion radii are set according to the blood vessel morphological characteristics to ensure that the three-dimensional buffer zone can fully cover the key area around the blood vessel where tumor cell interactions may occur. Independent three-dimensional buffer zones centered on each blood vessel structure are constructed according to the three-dimensional spatial coordinate system. Rules for handling buffer zone overlap are formulated, and the spatial rationality and coverage integrity of each three-dimensional buffer zone are verified. The three-dimensional buffer zones corresponding to all blood vessel structures are integrated to form a unified spatial model and generate a microenvironment model around the blood vessels.

[0092] The perivascular microenvironment model includes a three-dimensional buffer space system constructed with each vascular structure as the center and according to the differentiated expansion radius. It covers the area around the blood vessel where tumor cells may interact with each other, such as adhesion and invasion. The differentiated three-dimensional buffer area expansion radius is reflected in the fact that large-diameter blood vessels correspond to larger expansion radii. The rules for handling overlapping buffer areas include merging overlapping areas into a unified space.

[0093] (7) Combine the tumor cell localization data in the digital pathological images to perform quantitative analysis on the tumor cells in the perivascular microenvironment model and generate the vascular invasion index.

[0094] For example, spatially calibrated tumor cell localization data are extracted from digital pathological images to clarify the three-dimensional spatial coordinate information of each tumor cell. The three-dimensional spatial coordinates of each tumor cell are matched and compared one by one with the spatial boundary of the perivascular microenvironment model to screen out target tumor cells that are completely or partially located within the perivascular microenvironment model. The total number of target tumor cells is counted, the spatial distribution density of target tumor cells within the perivascular microenvironment model is calculated, and the shortest spatial distance relationship between target tumor cells and the surface of vascular structures is analyzed. The above multi-faceted analysis results are integrated according to preset quantification rules to complete the quantitative analysis and processing of tumor cells within the perivascular microenvironment model and generate a vascular invasion index.

[0095] The vascular invasion index includes a comprehensive quantitative indicator that reflects the number, distribution density, and spatial correlation of tumor cells within the perivascular microenvironment model with vascular structures. The preset quantitative rules include quantity weight, density weight, and distance weight.

[0096] (8) Perform multimodal feature fusion processing on the front cell density value, cell density-depth distribution curve and vascular invasion index to generate invasion mode feature vector.

[0097] For example, the leading edge cell density value is standardized using a uniform quantization scale to eliminate scale differences between different samples. Key feature parameters are extracted from the cell density-depth distribution curve. The vascular invasion index is hierarchically divided and quantified according to clinical and pathological significance. A weighting rule for multimodal features is set. The standardized leading edge cell density value, the extracted curve feature parameters, and the converted vascular invasion index are weighted according to the weighting rule. The feature redundancy elimination algorithm is used to remove duplicate information between different modal features, strengthen the expression of core features, and arrange all weighted feature parameters in a structured manner according to the preset feature sorting rule to complete the systematic integration of multimodal features and generate an invasion mode feature vector.

[0098] The invasion mode feature vector includes a multi-dimensional structured feature set that integrates standardized frontier cell density values, cell density-depth distribution curve feature parameters, and quantified vascular invasion index. The key feature parameters of the cell density-depth distribution curve include the peak position of the curve, the change in the slope of the curve, and the area under the curve. The weight allocation rules for the multimodal features are formulated based on the degree of biological correlation between each feature and the tumor's invasive ability.

[0099] In one embodiment, statistical processing of cell distribution in different depth ranges is performed based on a depth partitioning model to generate cell density-depth distribution curves, including:

[0100] (1) Based on the tissue density gradient change characteristics in the deep partitioning model, dynamic interval boundary optimization is performed to generate optimized deep partitions.

[0101] For example, based on the tissue density gradient change characteristics in the deep partitioning model, the gradual trend of tissue density in the depth direction and the intrinsic relationship with the biological tissue hierarchy are analyzed, key inflection points of tissue density change are identified, and the boundaries of the deep partitions are adjusted in combination with the structural hierarchy of biological tissues so that the tissue density change in each optimized deep partition conforms to the actual gradient law of biological tissues, thus completing the dynamic interval boundary optimization process and generating optimized deep partitions.

[0102] Among them, the optimized depth partitioning includes a set of depth partitions that can reflect the actual changes in tissue density after adjusting the boundaries based on the original depth partitioning model and combining the tissue density gradient change characteristics and biological tissue layers. The structural layers of biological tissues include the mucosa layer, muscle layer, etc.

[0103] (2) Perform tumor cell density calculation on each depth interval in the optimized depth partition to generate a discrete depth-density data point set.

[0104] For example, for each depth interval in the optimized depth partition, tumor cells are identified, normal cells and non-specific markers are excluded, the number of effective tumor cells in each depth interval is counted, the actual tissue space area of ​​each depth interval is measured, the tumor cell density value of each depth interval is calculated according to the tumor cell density calculation specification, and the depth identifier of each depth interval is associated with the corresponding tumor cell density value to generate a discrete depth-density data point set.

[0105] The discrete depth-density data point set includes a discrete data set consisting of the depth identifier of each optimized depth interval and the tumor cell density value within that interval. The tumor cell density calculation specification is a rule for calculating the ratio of the number of tumor cells to the actual tissue space area of ​​the corresponding depth interval.

[0106] (3) Using the following formula, biological tissue continuity fitting is performed based on the discrete depth-density data point set to generate cell density-depth distribution curves:

[0107]

[0108] in, This represents the cell density-depth distribution curve. Represents depth coordinates, The Dirac function represents the location of discrete data points. Indicates the first Tumor cell density values ​​in each depth range Indicates the total number of depth partitions. Operators representing the continuity constraints of biological tissues This represents the Gaussian smoothing kernel function. This represents the convolution operator.

[0109] For example, the biological tissue continuity constraint operator is invoked to apply the continuity rules of biological tissue level to the discrete depth-density data point set. At the same time, a Gaussian smoothing kernel function is introduced to smoothly connect the discrete data points according to the biological tissue continuity requirements through convolution operation, thereby completing the biological tissue continuity fitting process and generating the cell density-depth distribution curve.

[0110] Among them, the cell density-depth distribution curve includes a visualized curve that reflects the continuous correlation between different depths and the corresponding tumor cell densities. It is a continuous feature curve formed after biological tissue continuity constraints and Gaussian smoothing. The continuity rules at the biological tissue level include those that conform to the progressive characteristics of tumor cell invasion.

[0111] In one embodiment, a three-dimensional buffer region is created around the blood vessel coordinate dataset to generate a microenvironment model around the blood vessels, including:

[0112] (1) Based on the blood vessel coordinate dataset, the diameter distribution and orientation features are extracted and processed to generate blood vessel morphology feature vectors.

[0113] For example, based on a vascular coordinate dataset, the distribution and variation of the diameter of each vascular vessel in the spatial dimension are analyzed one by one, covering the gradual change in diameter, local expansion or contraction, and the characteristics such as the curvature of the direction and the branching pattern are analyzed. Key parameters that can quantify these morphologies are extracted, such as the mean diameter, the curvature of the direction, and the number of branch points. The above parameters are then integrated into a vector form in an orderly manner according to the vascular number to complete the extraction and processing of the diameter distribution and direction features, and generate a vascular morphology feature vector.

[0114] The vascular morphology feature vector includes a set of quantized vectors composed of diameter distribution parameters, orientation parameters, branching parameters, etc., for each blood vessel.

[0115] (2) Multi-scale spatial weight allocation processing is performed based on the vascular morphology feature vector to generate an adaptive three-dimensional buffer region.

[0116] For example, based on the vascular morphology feature vector, different spatial weight benchmarks are set for vessels of different diameters, distinguishing between large-diameter and small-diameter types. The complexity of the vessel's orientation is also considered, differentiating between straight and curved orientations, and adjusting the weight coefficients accordingly. A three-dimensional buffer region is constructed based on these weight rules: large-diameter vessels correspond to a larger buffer region, while the shape of the buffer region for vessels with complex orientations is dynamically adjusted to adapt to their orientation. This multi-scale spatial weight allocation process generates an adaptive three-dimensional buffer region.

[0117] The adaptive three-dimensional buffer region includes a set of three-dimensional spatial regions whose range and shape are dynamically adjusted according to the morphological characteristics of blood vessels, such as diameter and orientation, to adapt to the morphology of blood vessels.

[0118] (3) Using the following formula, combined with tumor cell localization data in digital pathological images, tumor cell-vascular spatial interaction relationship modeling is performed on the adaptive three-dimensional buffer region to generate a perivascular microenvironment model:

[0119]

[0120] in, This represents a model of the perivascular microenvironment. Indicates the coordinates of the centerline of the target blood vessel. Indicates blood vessels An adaptive three-dimensional buffer region Represents three-dimensional spatial coordinates. Indicates position Tumor cell density at the site, Represents the spatial weighting function. Indicates position to blood vessels European distance, Indicates position Compared to blood vessels azimuth angle, This represents the angle-sensitive function.

[0121] For example, spatially calibrated tumor cell localization data is extracted from digital pathological images to clarify the three-dimensional spatial coordinate information of each tumor cell. Within an adaptive three-dimensional buffer area, the tumor cell density at each location is statistically analyzed. The Euclidean distance from each location to the center line of the target blood vessel is calculated, and the azimuth angle of the location relative to the target blood vessel is determined. A spatial weighting function is introduced to assign corresponding weights to the Euclidean distance. An angle-sensitive function is used to differentiate the interaction relationship at different azimuth angles. The tumor cell density, the weighted Euclidean distance, and the angle-sensitive azimuth information are integrated and modeled to complete the modeling of the spatial interaction relationship between tumor cells and blood vessels, generating a microenvironment model around the blood vessel.

[0122] Among them, the perivascular microenvironment model includes a comprehensive model that integrates information such as tumor cell density, spatial distance from blood vessels, and azimuth angle to reflect the distribution of tumor cells around blood vessels and their spatial interaction with blood vessels.

[0123] In one embodiment, based on the marker information in the digital pathological image, tumor core region segmentation and invasion front boundary identification are performed to generate tumor invasion front boundary coordinates, including:

[0124] (1) Based on the tumor cell-specific expression characteristics in the labeling information, multi-channel feature fusion processing is performed to generate a tumor cell probability heatmap.

[0125] For example, based on the tumor cell-specific expression features in the labeling information, tumor cell-specific expression signals from multiple channels in the digital pathological image are extracted. The above-mentioned multi-channel features are integrated and fused. Through feature weighting, inter-channel correlation analysis and other methods, a heat map that reflects the distribution probability of tumor cells in the image is generated. Multi-channel feature fusion processing is completed to generate a tumor cell probability heat map.

[0126] Among them, the tumor cell probability heatmap includes a visual heatmap that integrates multi-channel tumor cell-specific expression features and uses color gradients to reflect the distribution probability of tumor cells.

[0127] (2) Perform topological connectivity analysis on the probability heatmap of tumor cells to generate the boundary of the tumor core region.

[0128] For example, the topological structure of connected regions in a probability heatmap of tumor cells is analyzed to identify connected domains that are continuously distributed and conform to the morphological characteristics of tumor tissue. The spatial range of the connected domains is determined, and a closed boundary surrounding the tumor core region is generated by a boundary extraction algorithm. This completes the topological connected domain analysis and generates the boundary of the tumor core region.

[0129] The tumor core region boundary includes the closed boundary that defines the extent of the tumor core tissue surrounding the screened tumor cell connected domains.

[0130] (3) Based on the boundary of the tumor core region, the density gradient field is calculated and processed to generate the tumor invasion direction vector field.

[0131] For example, based on the boundary of the tumor core region, the number of tumor cells at each spatial location within the boundary is counted, the density distribution of tumor cells is calculated, the density gradient change is analyzed, the possible direction of tumor cell invasion into surrounding tissues is determined according to the direction of the density gradient, the above direction information is integrated into a field structure in vector form, the density gradient field calculation is completed, and a tumor invasion direction vector field is generated.

[0132] The tumor invasion direction vector field includes a set of vectors reflecting the invasion direction of tumor cells based on the density gradient distribution within the boundary of the tumor core region, with each vector indicating the invasion direction at the corresponding location.

[0133] (4) Adaptive boundary tracking is performed along the tumor invasion direction vector field to generate the boundary coordinates of the tumor invasion front.

[0134] For example, along the vector field of tumor invasion direction, the tracking step size is adaptively adjusted according to the direction and distribution characteristics of the vector, while adapting to the structural characteristics of the surrounding tissue, to track the front edge position of tumor cells invading normal tissue, record the spatial coordinates of the above position, complete the adaptive boundary tracking process, and generate the boundary coordinates of the front edge of tumor invasion.

[0135] Among them, the boundary coordinates of the tumor invasion front include a set of spatial coordinates that define the frontal position of the tumor invasion into normal tissue, obtained by adaptive tracking of the vector field along the tumor invasion direction.

[0136] like Figure 2 As shown, topological connectivity analysis is performed on the probability heatmap of tumor cells to generate the boundary of the tumor core region, including:

[0137] S201: Based on the probability heatmap of tumor cells, homology group persistence calculation is performed to generate a persistent topological feature graph.

[0138] For example, based on the probability heatmap of tumor cells, the probability gradient characteristics and spatial distribution patterns of tumor cell distribution in the heatmap are analyzed to determine the core parameter settings for homology group calculation. By systematically traversing different probability threshold intervals, the generation timing and disappearance process of connected regions under each probability threshold are tracked in real time. The birth threshold and death threshold corresponding to each connected component are recorded. The birth threshold and death threshold of each connected component are used as paired coordinate points. All coordinate points are classified and organized according to the topological dimension to complete the homology group persistence calculation process and generate a topological feature persistence map.

[0139] The persistent topological feature graph includes a point set graph constructed with the birth threshold of connected components as the horizontal axis and the death threshold as the vertical axis. Each coordinate point corresponds to the topological existence characteristics of a connected component. The topological dimension includes zero-dimensional, one-dimensional, and other dimension types that reflect the structure of connected regions.

[0140] S202: Feature filtering based on persistent topological feature graphs to generate a set of target connected components for pathological constraints.

[0141] For example, based on a persistent topological feature graph and combined with the core morphological constraints of tumor tissue in pathology, such constraints include spatial scale requirements for connected components, length standards for persistent threshold intervals, and specifications for the degree of association with high-probability regions of tumor cells, a multi-dimensional screening standard is set. Each feature point in the persistent topological feature graph is verified one by one according to the screening standard. Feature points corresponding to small connected components, transient noise components, and non-tumor-related components are removed, and valid feature points that conform to the topological features of the core tumor tissue are retained. The valid feature points are back-mapped to the corresponding connected components in the probability heatmap of tumor cells. All valid connected components are integrated to complete the feature screening process for pathological constraints and generate a target connected component set.

[0142] The target connected component set includes a group of connected components that meet the morphological characteristics and topological properties of the core tumor tissue after being screened by pathological constraints. Each connected component corresponds to a continuous high-probability tumor cell distribution area in the tumor cell probability heatmap. The high-probability tumor cell distribution area refers to the area where tumor cell-specific expression signals are concentrated.

[0143] S203: Perform boundary optimization processing on the target connected component set with tissue continuity constraints to generate the boundary of the tumor core region.

[0144] For example, for each connected component in the target connected component set, the initial boundary contour of each connected component is extracted by a boundary extraction algorithm. Based on the principle of tissue continuity constraint, the spatial continuity and morphological rationality of the initial boundary are analyzed. For irregular structures such as broken parts and serrated protrusions in the boundary, a boundary smoothing algorithm is used for morphological correction. The boundaries of adjacent connected components that meet the tissue continuity requirements are fused to fill the boundary gaps, ensuring that the optimized boundary can completely wrap the tumor core tissue. At the same time, an isolated noise point on the boundary is removed by a noise removal algorithm to make the boundary contour consistent with the spatial distribution of the actual tumor tissue. This completes the boundary optimization process for tissue continuity constraint and generates the boundary of the tumor core region.

[0145] Among them, the boundary of the tumor core region includes the closed boundary contour after morphological correction, gap filling and noise removal, which can define the spatial range of the tumor core tissue, and the boundary morphology conforms to the actual continuous distribution characteristics of the tumor tissue. The principle of tissue continuity constraint refers to the biological requirement that the tumor tissue presents a continuous distribution in space without obvious breaks.

[0146] In one embodiment, the boundary of the tumor core region is obtained using the following formula:

[0147]

[0148] in, Indicates the boundary of the tumor core region. Indicates the first Each target connected component Indicates the total number of target connected components. Indicates boundary With connected components The intersection measure, This represents the boundary smoothness regularization term. This represents the weighting factor for organizational continuity constraints. Denotes the candidate boundary variables to be solved. Indicates the number of the target connected component.

[0149] For example, candidate boundary variables are identified to serve as candidate shapes for the boundaries of the tumor core region, providing a foundation for subsequent optimization calculations. The intersection measure between the candidate boundary variables and each target connected component is calculated, and the degree of overlap between the candidate boundary variables and each target connected component is statistically analyzed. Simultaneously, a smoothness regularization term for the candidate boundary variables is calculated to quantify their smoothness characteristics.

[0150] By introducing a weighting factor for tissue continuity constraints and assigning corresponding weights to the smoothness regularization term, the sum of intersection measures is combined with the weighted smoothness regularization term to construct an optimization objective function that integrates multiple constraints. Solving the optimization objective function, the candidate boundary variable that minimizes the objective function value is found. This variable represents the tumor core region boundary that simultaneously satisfies the constraints of matching the target connected components, boundary smoothness, and tissue continuity. Boundary solving is then completed to obtain the boundary of the tumor core region.

[0151] It should be understood that although the steps in the flowcharts of the embodiments described above are shown sequentially according to the arrows, these steps are not necessarily executed in the order indicated by the arrows. Unless explicitly stated herein, there is no strict order restriction on the execution of these steps, and they can be executed in other orders. Moreover, at least some steps in the flowcharts of the embodiments described above may include multiple steps or multiple stages. These steps or stages are not necessarily completed at the same time, but can be executed at different times. The execution order of these steps or stages is not necessarily sequential, but can be performed alternately or in turn with other steps or at least some of the steps or stages of other steps.

[0152] In one embodiment, such as Figure 3 As shown, this application also provides a tumor cell counting device 300, which includes:

[0153] The sample staining and imaging module 301 is used to specifically stain biological samples containing tumor-normal tissue interfaces using immunohistochemistry or multiplex fluorescent labeling technology, and to generate digital pathological images by image acquisition and processing through a preset scanning device.

[0154] The tumor boundary delineation module 302 is used to perform tumor core region segmentation and invasion front boundary identification processing based on the marker information in the digital pathological image, and generate the tumor invasion front boundary coordinates;

[0155] The spatial feature quantization module 303 is used to perform multi-dimensional quantification of the spatial distribution features of tumor cells based on the coordinates and labeling information of the tumor invasion front boundary, and generate an invasion pattern feature vector.

[0156] The invasion capability aggregation module 304 is used to perform multi-dimensional weighted aggregation processing on the invasion mode feature vector to generate a quantitative index of tumor invasion capability.

[0157] The clinical assessment output module 305 is used to generate a comprehensive assessment report based on quantitative indicators of tumor invasiveness, and obtain a quantitative assessment report.

[0158] Specifically, the sample staining and imaging module 301 uses immunohistochemistry or multiplex fluorescent labeling technology to select specific markers for tumor cells, and specifically stains biological samples containing the tumor-normal tissue interface, so that tumor cells and normal tissues exhibit differentiated staining characteristics. The stained biological sample is then imaged using a preset scanning device, converting the tissue morphology and staining signals of the sample into digital image data, completing image acquisition and processing, and generating a digital pathological image.

[0159] Digital pathological images include digital images of biological samples that are acquired by a preset scanning device, contain the tumor-normal tissue interface, and are specifically stained.

[0160] The tumor boundary delineation module 302 extracts features from the marker information in the digital pathological image, performs segmentation of the tumor core region, and identifies the core aggregation region of tumor cells. For the tumor's invasion front into normal tissue, it performs boundary recognition processing to determine the spatial location of the tumor invasion front, records the coordinates of this location, completes the tumor core region segmentation and invasion front boundary recognition processing, and generates the tumor invasion front boundary coordinates.

[0161] Among them, the boundary coordinates of the tumor invasion front include a set of spatial coordinates that define the location of the tumor's invasion front into normal tissue.

[0162] The spatial feature quantification module 303 determines the spatial range for multi-dimensional quantification based on the coordinates of the tumor invasion front boundary, such as creating a front analysis zone and depth partitioning. Simultaneously, based on labeled information, it identifies elements such as tumor cells and vascular structures, and quantifies front cell density, cell density-depth distribution, and vascular invasion correlation features, respectively. It then fuses these features from different dimensions to complete the multi-dimensional quantification of tumor cell spatial distribution features, generating an invasion pattern feature vector.

[0163] The invasion pattern feature vector includes a multi-dimensional feature set that integrates frontal cell density, cell depth distribution, and vascular invasion-related features.

[0164] The invasion capability aggregation module 304 assigns differentiated weights to each dimension of the invasion pattern feature vector based on its biological correlation with tumor invasion capability. The module then performs weighted calculations on each dimension of the features according to these weights, aggregates the weighted features to form a comprehensive quantitative index, and completes the multi-dimensional weighted aggregation process to generate a quantitative index of tumor invasion capability.

[0165] Among them, the quantitative indicators of tumor invasion ability include a comprehensive quantitative indicator that reflects the tumor invasion ability after integrating the weighted values ​​of each dimension of the invasion pattern feature vector.

[0166] The Clinical Assessment Output Module 305, based on quantitative indicators of tumor invasiveness, organizes the sub-data and biological significance of these indicators. Combining this with standardized clinical pathology report formats, it constructs a report framework that includes assessment conclusions, indicator descriptions, and supporting data. The report content undergoes information verification and format standardization to ensure clinical readability and logical coherence, completing the comprehensive assessment report generation process to obtain a quantitative assessment report.

[0167] The quantitative assessment report includes a standardized clinical assessment report that is generated based on quantitative indicators of tumor invasiveness and contains assessment conclusions and data support.

[0168] The spatial feature quantization module 303 is also used for:

[0169] A band-shaped analysis region is created based on the coordinates of the tumor invasion front boundary, generating the front analysis band;

[0170] The density of tumor cells within the leading edge analysis zone is statistically processed to generate leading edge cell density values.

[0171] A vertical depth partitioning model is generated by establishing vertical depth partitioning based on the coordinates of the tumor invasion front boundary.

[0172] Based on the depth partitioning model, statistical processing of cell distribution in different depth intervals is performed to generate cell density-depth distribution curves.

[0173] Spatial range constraint processing is performed based on the coordinates of the tumor invasion front boundary, and vascular structure recognition processing is performed based on the labeled information to generate a vascular coordinate dataset.

[0174] A three-dimensional buffer region is created around the blood vessel coordinate dataset to generate a microenvironment model around the blood vessels;

[0175] By combining tumor cell localization data from digital pathology images, quantitative analysis of tumor cells within the perivascular microenvironment model is performed to generate a vascular invasion index.

[0176] Multimodal feature fusion processing is performed on the front cell density value, cell density-depth distribution curve and vascular invasion index to generate invasion mode feature vector.

[0177] The spatial feature quantization module 303 is also used for:

[0178] Based on the tissue density gradient change characteristics in the deep partitioning model, dynamic interval boundary optimization is performed to generate optimized deep partitions.

[0179] Tumor cell density is calculated for each depth interval in the optimized depth partition to generate a discrete depth-density data point set.

[0180] The following formula is used to perform continuous fitting of biological tissue based on a discrete depth-density data point set, generating a cell density-depth distribution curve:

[0181]

[0182] in, This represents the cell density-depth distribution curve. Represents depth coordinates, The Dirac function represents the location of discrete data points. Indicates the first Tumor cell density values ​​in each depth range Indicates the total number of depth partitions. Operators representing the continuity constraints of biological tissues This represents the Gaussian smoothing kernel function. This represents the convolution operator.

[0183] The spatial feature quantization module 303 is also used for:

[0184] Based on the blood vessel coordinate dataset, the diameter distribution and orientation features of the blood vessel are extracted to generate blood vessel morphology feature vectors.

[0185] Multi-scale spatial weight allocation processing is performed based on vascular morphology feature vectors to generate an adaptive three-dimensional buffer region.

[0186] Using the following formula, combined with tumor cell localization data from digital pathology images, a model of the tumor cell-vascular spatial interaction relationship is generated for an adaptive three-dimensional buffer region to produce a perivascular microenvironment model:

[0187]

[0188] in, This represents a model of the perivascular microenvironment. Indicates the coordinates of the centerline of the target blood vessel. Indicates blood vessels An adaptive three-dimensional buffer region Represents three-dimensional spatial coordinates. Indicates position Tumor cell density at the site, Represents the spatial weighting function. Indicates position to blood vessels European distance, Indicates position Compared to blood vessels azimuth angle, This represents the angle-sensitive function.

[0189] The tumor boundary delineation module 302 is also used for:

[0190] Based on the tumor cell-specific expression characteristics in the labeled information, multi-channel feature fusion processing is performed to generate a tumor cell probability heatmap.

[0191] Topological connectivity analysis is performed on the probability heatmap of tumor cells to generate the boundary of the tumor core region;

[0192] Density gradient field calculation is performed based on the boundary of the tumor core region to generate a tumor invasion direction vector field.

[0193] Adaptive boundary tracking is performed along the vector field of tumor invasion direction to generate the coordinates of the tumor invasion front boundary.

[0194] The tumor boundary delineation module 302 is also used for:

[0195] Homology group persistence calculation is performed based on tumor cell probability heatmap to generate a topological feature persistence graph;

[0196] Feature filtering based on persistent topological feature graphs for pathological constraints is performed to generate a set of target connected components.

[0197] Boundary optimization processing with tissue continuity constraints is performed on the target connected component set to generate the boundary of the tumor core region.

[0198] The tumor boundary delineation module 302 is also used to obtain the boundary of the tumor core region using the following formula:

[0199]

[0200] in, Indicates the boundary of the tumor core region. Indicates the first Each target connected component Indicates the total number of target connected components. Indicates boundary With connected components The intersection measure, This represents the boundary smoothness regularization term. This represents the weighting factor for organizational continuity constraints. Denotes the candidate boundary variables to be solved. Indicates the number of the target connected component.

[0201] In one embodiment, this application also provides a computer device, including a memory and a processor, wherein the memory stores a computer program, and the processor executes the computer program to implement the steps in the above-described method embodiments.

[0202] In one embodiment, this application also provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the steps in the above-described method embodiments.

[0203] For the device embodiments, since they basically correspond to the method embodiments, the relevant parts can be referred to in the description of the method embodiments. The device embodiments described above are merely illustrative. The components described as separate parts may or may not be physically separate, and 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 modules can be selected to achieve the purpose of this disclosure according to actual needs. Those skilled in the art can understand and implement this without creative effort.

[0204] The above-described embodiments are merely illustrative of several implementation methods of the embodiments of this application, and their descriptions are relatively specific and detailed. However, they should not be construed as limiting the scope of the patent application. It should be noted that those skilled in the art can make various modifications and improvements without departing from the concept of the embodiments of this application, and these modifications and improvements all fall within the protection scope of the embodiments of this application.

Claims

1. A method for statistical analysis of tumor cells, characterized in that, The method includes: Biological samples containing the tumor-normal tissue interface are specifically stained using immunohistochemistry or multiplex fluorescent labeling techniques, and digital pathological images are generated by image acquisition and processing using a preset scanning device. Based on the marker information in the digital pathological image, tumor core region segmentation and invasion front boundary identification are performed to generate tumor invasion front boundary coordinates. Based on the tumor invasion front boundary coordinates and the labeling information, multi-dimensional quantification of the spatial distribution characteristics of tumor cells is performed to generate an invasion pattern feature vector. The invasion mode feature vector is subjected to multi-dimensional weighted aggregation processing to generate a quantitative index of tumor invasion capability; A comprehensive assessment report is generated based on the quantitative indicators of tumor invasion ability, resulting in a quantitative assessment report.

2. The tumor cell statistical method according to claim 1, characterized in that, The step of performing multi-dimensional quantification of the spatial distribution characteristics of tumor cells based on the tumor invasion front boundary coordinates and the marker information to generate an invasion pattern feature vector includes: A band-shaped analysis region is created based on the coordinates of the tumor invasion front boundary, generating a front analysis band; The tumor cells within the frontal analysis zone are subjected to density statistical processing to generate frontal cell density values; A vertical depth partition is established based on the coordinates of the tumor invasion front boundary, and a depth partition model is generated; Based on the aforementioned depth partitioning model, statistical processing of cell distribution in different depth ranges is performed to generate cell density-depth distribution curves. Spatial range constraint processing is performed based on the coordinates of the tumor invasion front boundary, and vascular structure recognition processing is performed based on the labeled information to generate a vascular coordinate dataset. A three-dimensional buffer region is created around the aforementioned blood vessel coordinate dataset to generate a microenvironment model around the blood vessels; By combining the tumor cell localization data in the digital pathological images, the tumor cells in the perivascular microenvironment model are quantitatively analyzed to generate a vascular invasion index. The frontal cell density value, the cell density-depth distribution curve, and the vascular invasion index are subjected to multimodal feature fusion processing to generate the invasion mode feature vector.

3. The tumor cell statistical method according to claim 2, characterized in that, The step of performing statistical processing of cell distribution in different depth intervals based on the depth partitioning model to generate cell density-depth distribution curves includes: Based on the tissue density gradient change characteristics in the deep partitioning model, dynamic interval boundary optimization is performed to generate optimized deep partitions. Tumor cell density is calculated for each depth interval in the optimized depth partition to generate a discrete depth-density data point set; The cell density-depth distribution curve is generated by performing a continuous biological tissue fitting process based on the discrete depth-density data point set using the following formula: in, This represents the cell density-depth distribution curve. Represents depth coordinates, The Dirac function represents the location of discrete data points. Indicates the first Tumor cell density values ​​in each depth range Indicates the total number of depth partitions. Operators representing the continuity constraints of biological tissues This represents the Gaussian smoothing kernel function. This represents the convolution operator.

4. The tumor cell statistical method according to claim 2, characterized in that, The process of creating a three-dimensional buffer region around the blood vessel coordinate dataset to generate a microenvironment model around the blood vessels includes: Based on the aforementioned blood vessel coordinate dataset, the diameter distribution and orientation features of the blood vessel are extracted to generate a blood vessel morphology feature vector. Based on the aforementioned vascular morphology feature vector, multi-scale spatial weight allocation processing is performed to generate an adaptive three-dimensional buffer region. Using the following formula, combined with tumor cell localization data from the digital pathology image, the tumor cell-vascular spatial interaction relationship modeling process is performed on the adaptive three-dimensional buffer region to generate the perivascular microenvironment model: in, This represents a model of the perivascular microenvironment. Indicates the coordinates of the centerline of the target blood vessel. Indicates blood vessels An adaptive three-dimensional buffer region Represents three-dimensional spatial coordinates. Indicates position Tumor cell density at the site, Represents the spatial weighting function. Indicates position to blood vessels European distance, Indicates position Compared to blood vessels azimuth angle, This represents the angle-sensitive function.

5. The tumor cell statistical method according to claim 1, characterized in that, The step of segmenting the tumor core region and identifying the invasion front boundary based on the marker information in the digital pathological image, and generating the tumor invasion front boundary coordinates, includes: Based on the tumor cell-specific expression characteristics in the labeled information, multi-channel feature fusion processing is performed to generate a tumor cell probability heatmap. The tumor cell probability heatmap is subjected to topological connectivity analysis to generate the boundary of the tumor core region; Density gradient field calculation is performed based on the boundary of the tumor core region to generate a tumor invasion direction vector field. Adaptive boundary tracking is performed along the tumor invasion direction vector field to generate the boundary coordinates of the tumor invasion front.

6. The tumor cell statistical method according to claim 5, characterized in that, The step of performing topological connected component analysis on the probability heatmap of tumor cells to generate the boundary of the tumor core region includes: Based on the tumor cell probability heatmap, homology group persistence calculation is performed to generate a topological feature persistence map. Based on the aforementioned persistent topological feature graph, feature filtering processing for pathological constraints is performed to generate a target connected component set. Boundary optimization processing with tissue continuity constraints is performed on the target connected component set to generate the boundary of the tumor core region.

7. The tumor cell statistical method according to claim 6, characterized in that, The boundary of the tumor core region is obtained using the following formula: in, Indicates the boundary of the tumor core region. Indicates the first Each target connected component Indicates the total number of target connected components. Indicates boundary With connected components The intersection measure, This represents the boundary smoothness regularization term. This represents the weighting factor for organizational continuity constraints. Denotes the candidate boundary variables to be solved. Indicates the number of the target connected component.

8. A tumor cell counting device, characterized in that, The device includes: The sample staining and imaging module is used to specifically stain biological samples containing the tumor-normal tissue interface using immunohistochemistry or multiplex fluorescent labeling technology, and to generate digital pathological images by image acquisition and processing through a preset scanning device. The tumor boundary delineation module is used to perform tumor core region segmentation and invasion front boundary identification processing based on the marker information in the digital pathological image, and generate tumor invasion front boundary coordinates; The spatial feature quantization module is used to perform multi-dimensional quantification of the spatial distribution features of tumor cells based on the coordinates of the tumor invasion front boundary and the labeling information, and generate an invasion pattern feature vector. The invasion capability aggregation module is used to perform multi-dimensional weighted aggregation processing on the invasion mode feature vector to generate a quantitative index of tumor invasion capability. The clinical assessment output module is used to generate a comprehensive assessment report based on the quantitative indicators of tumor invasiveness, resulting in a quantitative assessment report.

9. A computer device comprising a memory and a processor, wherein the memory stores a computer program, characterized in that, When the processor executes the computer program, it implements the steps of the tumor cell statistical method according to any one of claims 1 to 7.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by a processor, it implements the steps of the tumor cell statistical method according to any one of claims 1 to 7.

Images