A method and system for quantifying the tumor cocoon house structure of an ex vivo lung cancer section digital pathology image

Abstract

This invention provides a method and system for quantifying tumor cocoon structures in digital pathological images of ex vivo lung cancer slides. The method includes: acquiring digital pathological image data to be processed; identifying tumor cells based on cell type annotation information and spatial coordinate information; determining tumor nests formed by the aggregation of tumor cells and using them as core objects; constructing an enclosing shell that is annular, arc-shaped, or partially closed; selecting at least one cell type within the enclosing shell as target cells; defining the core object, target cells, and enclosing shell as tumor cocoon structures; and calculating quantitative indicators of the enclosing ability of target cells and the aggregation degree of core objects. This invention solves the technical problem of lacking feasible quantitative methods to describe such spatial enclosing structures in the prior art by accurately identifying and quantifying tumor cocoon structures.

Description

Technical Field

[0001] This invention relates to a quantitative processing method and system, specifically to a method and system for quantitative processing of tumor cocoon structure in digital pathological images of ex vivo lung cancer slices. Background Technology

[0002] Non-small cell lung cancer (NSCLC) is one of the most common pathological types of lung cancer. Current clinical risk assessment systems primarily rely on TNM staging, histological typing, and certain molecular markers. However, these indicators mainly reflect the degree of tumor progression at the anatomical level or the state of its molecular characteristics, failing to fully reveal the impact of cell spatial distribution patterns within the tumor microenvironment on disease progression and recurrence behavior. Therefore, relying solely on traditional clinical and molecular markers makes it difficult to achieve refined stratification of patient recurrence risk and individualized prognostic assessment.

[0003] The tumor microenvironment influences tumorigenesis and development not only through cellular composition but also through the spatial relationships between these cells. Tertiary Lymphoid Structures (TLS), formed by immune cells (including B cells, T cells, and dendritic cells) and tumor cells, have been shown to participate in tumor progression regulation as independent spatial structures, possessing value in risk assessment and prognostic prediction. In addition to TLS, recent studies have also revealed observable differences in composition and spatial distribution in tissue sections from metastatic / recurrent and non-recurrent NSCLC patients: a type of tumor-associated fibroblast (belonging to stromal cells) forms an irregular spatial enclosure structure around proliferating epithelial cells (belonging to tumor cells).

[0004] Current research lacks a systematic and standardized definition and quantitative analysis system for such spatial enveloping structures. While some studies use terminology to describe the envelopment of tumors by fibrin or fibrotic matrix, these concepts mostly remain at the level of phenomenological description or mechanistic speculation, failing to systematically develop structural biomarkers with clearly defined, quantifiable indicators and standardized analytical procedures specific to NSCLC. In addition, some studies have analyzed immune loop structures, which share some similarities with spatial enveloping structures, identifying them by statistically analyzing whether a cell type forms a regular ring-like distribution around another cell type within a specific radius. Immune loop structures typically rely on relatively regular, symmetrical, and approximately concentric circle spatial distribution patterns, and are only suitable for identifying relatively idealized ring structures. Spatial enveloping structures differ from simple immune loop structures; their core characteristic lies in the formation of a shell-like structure with boundary features by one cell type surrounding another, which may exhibit locally closed or semi-closed morphologies. More importantly, multiple independent enclosing units often exist simultaneously in the same tissue slice, each with different morphological scales and spatial intensity. The complexity of this structure cannot be defined by a single spatial index. Therefore, it is urgent to establish a clear conceptual definition and quantitative analysis system for this spatial enclosing structure. Summary of the Invention

[0005] The purpose of this application is to provide a method and system for quantitative processing of tumor cocoon structure in digital pathological images of ex vivo lung cancer slices, so as to objectively quantify the spatial enclosure ability of irregularly shaped and heterogeneous tumor cocoon structures in the tumor microenvironment.

[0006] According to a first aspect of the present invention, a method for quantifying tumor cocoon structure in digital pathological images of ex vivo lung cancer sections is provided, the method comprising: Acquire digital pathological image data to be processed, the digital pathological image data including cell type annotation information and spatial coordinate information of each cell; Tumor cells are identified based on the cell type annotation information and spatial coordinate information, and tumor nests formed by the aggregation of tumor cells are determined by spatial density clustering algorithm, which are taken as the core objects of tumor cocoon structure. Identify the edge cells of the core object, and based on the edge cells, construct a ring-shaped, arc-shaped, or partially closed surrounding shell around the core object according to a preset distance threshold; At least one cell type within the surrounding shell is selected as the target cell, and the core object, the target cell, and the surrounding shell are defined as a tumor cocoon structure. The distribution characteristics of the target cells within the surrounding shell are calculated as a quantitative indicator of the enclosing ability of the target cells. The ratio of the core object to the total space area is calculated as a quantitative indicator of the aggregation degree of the core object.

[0007] Preferably, the core object is a cluster of tumor cells formed by proliferating epithelial cells; the target cells are tumor-associated fibroblasts; and the surrounding shell includes stromal cells, immune cells, and other microenvironment cell types.

[0008] Preferably, the step of determining the tumor nest includes: clustering the tumor cells using the density-based clustering algorithm DBSCAN to identify tumor cell clusters that form an aggregated state; removing isolated or scattered tumor cells that do not form an aggregated state, and identifying the tumor cell clusters as the core objects.

[0009] Preferably, the step of identifying the edge cells of the core object includes: using a nearest neighbor algorithm to calculate the spatial relationship between each cell in the core object and its surrounding cells, and identifying cells located at the geometric boundary of the core object as edge cells.

[0010] Preferably, the method for calculating the quantitative index of enclosing ability includes: statistically analyzing the proportion of the number of target cells within the enclosing shell to the total number of cells within the enclosing shell; and, in the case of multiple tumor cocoon structures, calculating the statistical summary value of the corresponding proportions of all tumor cocoon structures as the quantitative index of enclosing ability.

[0011] Preferably, the method further includes: traversing all tumor cocoon structures in the digital pathology image to be processed, checking whether there is a unit structure whose surrounding shell region does not contain target cells, if such a unit exists, it is defined as a completely non-enclosed unit, and calculating the proportion of the completely non-enclosed unit to the total tumor cocoon structure units as a special structural state indicator.

[0012] Preferably, the cell type annotation information and the spatial coordinate information of each cell can be obtained by at least one of the following methods: multiplex immunofluorescence, multiplex immunohistochemistry, spatial transcriptomics, spatial proteomics, or digital pathological image analysis.

[0013] According to a second aspect of the present invention, a system for quantifying tumor cocoon structure in digital pathological images of ex vivo lung cancer slides is provided, comprising: The information acquisition module is used to acquire digital pathological image data to be processed, the digital pathological image data including cell type annotation information and spatial coordinate information of each cell; The core object identification module is used to identify tumor cells based on the cell type annotation information and spatial coordinate information, and to determine the tumor nests formed by the aggregation of tumor cells through a spatial density clustering algorithm, which are used as the core objects of the tumor cocoon structure. The surrounding shell construction module is used to identify the edge cells of the core object and, based on the edge cells, construct a surrounding shell in the form of a ring, arc, or partially closed shape around the core object according to a preset distance threshold. The tumor cocoon structure definition module is used to select at least one cell type within the surrounding shell as the target cell, and to define the core object, the target cell and the surrounding shell as a tumor cocoon structure. The quantitative index calculation module is used to calculate the distribution characteristics of the target cells within the surrounding shell as a quantitative index of the enclosing ability of the target cells, and to calculate the ratio of the core object to the total space area as a quantitative index of the aggregation degree of the core object.

[0014] Compared with existing technologies, this invention has the following significant advantages: 1. By identifying and quantifying tumor cocoon structures, the structure is transformed from a conceptual spatial model into a calculable and comparable spatial quantitative indicator, filling the gap in the quantification of spatial topological features; 2. By constructing an enclosing shell and calculating the proportion of target cells, a quantitative indicator reflecting the physical barrier capacity of the tumor microenvironment is provided, thereby improving the depth of biological feature extraction; 3. By combining clinical variables with spatial quantitative indicators to construct a recurrence risk prediction model, the tumor cocoon structure can be used as a new spatial structural biomarker for recurrence risk assessment and prognosis prediction in NSCLC, which not only improves the discriminative ability of risk assessment but also enhances the interpretability and clinical application potential of the prediction results. Attached Figure Description

[0015] Figure 1 This is a schematic diagram of multiplex immunofluorescence imaging.

[0016] Figure 2 A schematic diagram illustrating the calculation of the target cell encirclement ability;

[0017] Figure 3 A schematic diagram illustrating the calculation of the aggregation degree of core objects;

[0018] Figure 4 Schematic diagram illustrating different states of a tumor cocoon structure that is not completely enclosed by a unit;

[0019] Figure 5 A schematic diagram showing the differences in tumor cocoon structure indicators between the relapse and non-relapse groups;

[0020] Figure 6A schematic diagram of the ROC curve for constructing a prognostic assessment model for NSCLC based on tumor cocoon structure parameters;

[0021] Figure 7 This is a schematic diagram of Kaplan–Meier survival curves based on tumor cocoon structure indices for survival analysis. Detailed Implementation

[0022] To facilitate understanding of the core inventive concept of this application, several key terms and background mechanisms involved in this application are defined and explained herein. The Cancer Cocoon Structure (CCS) proposed in this application is a novel structural biomarker used to describe the spatial topological interactions of cells within the tumor microenvironment (TME). In the development and progression of non-small cell lung cancer (NSCLC), tumor cells do not exist in isolation but rather engage in complex physical contacts and chemical signal interactions with surrounding stromal cells, immune cells, and other cells. In ex vivo tissue sections of NSCLC, cancer-associated fibroblasts (CAFs) often form locally closed or semi-closed spatial enclosing structures around proliferative epithelial cells (i.e., tumor nests formed by the aggregation of proliferative epithelial cells). This structure is not simply spatially adjacent or mixed, but rather exhibits a spatial screen enclosing pattern with structural boundaries. This spatial screen enclosing pattern is morphologically similar to the structure of a silkworm cocoon enclosing a pupa; therefore, this application defines it as the cancer cocoon structure.

[0023] The smallest unit of the tumor cocoon structure defined in this application may include a core object, target cells, an enclosing shell, and the spatial enclosing relationship between them. The core object refers to a cluster of tumor cells that exhibits a significant spatial aggregation. In a preferred embodiment of this application, it refers to a cluster of tumor cells identified by a spatial density clustering algorithm of proliferating epithelial cells. Limiting the core object to clustered cell masses rather than isolated single cells ensures that the structure has clear geometric and topological entity properties, thereby eliminating interference from randomly distributed scattered cells in spatial quantification. Target cells refer to specific microenvironment cell types selected for quantitative analysis within the enclosing shell. In this application, target cells are preferably tumor-associated fibroblasts. The enclosing shell refers to a cell zone region located outside the core object, extending outward from the edge of the core object within a predetermined spatial range. This region may contain stromal cells, immune cells, and other microenvironment cell types. The enclosing shell presents a ring-shaped, arc-shaped, or partially closed enclosing structure relative to the core object, forming a spatial barrier-like structure. This spatial enclosure relationship differs from simple nearest neighbor distribution, regular concentric circle distribution, or random distribution structure. Its essential characteristic lies in establishing an irregular enclosure relationship for core objects with irregular shapes.

[0024] Current technologies primarily rely on traditional TNM staging, histological typing, and single molecular markers as indicators. However, these indicators mainly reflect the anatomical progression of the tumor or its overall molecular expression status, completely ignoring the impact of cellular spatial distribution patterns within the tumor microenvironment on disease progression. Especially for early-stage (e.g., stage I, stage II) patients, even with the same TNM stage, the risk of postoperative recurrence exhibits significant heterogeneity due to the vast differences in cellular spatial topology within their tumor microenvironment. Furthermore, existing spatial point pattern analysis methods (such as the Cross K function algorithm) are typically based on the assumption of regular concentric circles or symmetrical rings, only suitable for identifying highly idealized regular ring structures, and unable to effectively identify and quantify highly irregular tumor cocoon structures with significant dimensional heterogeneity.

[0025] To overcome the limitations of the existing technologies, this application proposes a method and system for quantifying tumor cocoon structures in digital pathological images of ex vivo lung cancer slices. This method enables automatic identification and accurate quantification of irregular spatial enclosed structures and constructs a recurrence risk prediction model, thereby providing a scientific basis for individualized prognostic assessment of non-small cell lung cancer (NSCLC) patients.

[0026] Example 1

[0027] To more clearly illustrate the technical solution of this application, a method for quantifying the tumor cocoon structure in digital pathological images of ex vivo lung cancer sections is described in detail below. This method may specifically include the following steps:

[0028] Step 1: Obtain the digital pathological image data to be processed. The digital pathological image data includes cell type annotation information and spatial coordinate information of each cell.

[0029] Preferably, cell type annotation information and spatial coordinate information can be obtained through at least one of multiplex immunofluorescence, multiplex immunohistochemistry, spatial transcriptomics, digital pathological image analysis, or other detection methods that can provide cell protein or gene expression information and cell spatial information.

[0030] For example, to prepare ex vivo tissue sections from patients with non-small cell lung cancer after surgery, multiplex immunofluorescence staining techniques can be used to obtain high-resolution cellular spatial and phenotypic information. For instance, specific antibodies against proteins such as PANCK (epithelial / tumor cell marker), SMA (fibroblast marker), and Ki67 (proliferation marker) can be used to perform multiple rounds of staining and imaging of the sections. After imaging, the whole-slide scan image is processed using digital pathology image analysis software. Image denoising and channel registration are performed, and then a cell segmentation algorithm is used to identify the nucleus and cytoplasmic boundary of each cell, and its geometric center is calculated as the spatial coordinates of the cell. Based on the average fluorescence intensity of each cell in each channel, a reasonable threshold is set to classify cells into different phenotypes. For example, PANCK-positive and Ki67-positive cells are classified as proliferative epithelial cells, and SMA-positive cells are classified as tumor-associated fibroblasts. A data table containing all cell IDs, X coordinates, Y coordinates, and cell type labels is generated.

[0031] Step 2: Identify tumor cells based on the cell type annotation information and spatial coordinate information, and determine the tumor nests formed by the aggregation of tumor cells using a spatial density clustering algorithm, taking them as the core objects of the tumor cocoon structure.

[0032] Preferably, the core object is a cluster of tumor cells formed by proliferating epithelial cells.

[0033] Specifically, since tumors typically grow in nest-like or clump-like forms within tissues, in order to accurately identify these biologically-based tumor cell clusters, the density-based spatial clustering algorithm DBSCAN is used to cluster tumor cells. This identifies tumor cell clusters that have formed aggregates, removes isolated or scattered tumor cells that have not formed aggregates, and groups spatially high-density clusters of cells into one or more clusters. Each successfully identified tumor cell cluster is then identified as a core object, representing an independent tumor nest.

[0034] Step 3: Identify the edge cells of the core object, and using the edge cells as a reference, construct a ring-shaped, arc-shaped, or partially closed surrounding shell around the core object according to a preset distance threshold.

[0035] Specifically, for each core object identified in step two, its precise boundary needs to be determined. This embodiment uses a nearest neighbor algorithm to calculate the spatial relationship between each cell in the core object and its surrounding cells, identifying cells located at the geometric boundary of the core object as edge cells, and calculating the spatial distance from surrounding cells to these edge cells. After identifying the edge cells, a preset spatial range threshold is defined based on the spatial coordinates of these edge cells. Specifically, for all non-core object cells in the tissue sample, the spatial distance from each to the nearest edge cell is calculated. If this distance meets the preset spatial range threshold, the region is defined as the surrounding shell.

[0036] Step 4: Select at least one cell type within the surrounding shell as the target cell, and define the core object, target cell, and surrounding shell as a tumor cocoon structure.

[0037] Preferably, the target cells are tumor-associated fibroblasts, and the surrounding shell contains stromal cells, immune cells, and other microenvironment cell types.

[0038] Specifically, for each smallest unit of a tumor cocoon structure consisting of a core object and an surrounding shell, the target cells within its surrounding shell are first identified.

[0039] Step 5: Calculate the distribution characteristics of the target cells within the surrounding shell as a quantitative indicator of the target cells' enclosing ability. Calculate the ratio of the core object to the total space area as a quantitative indicator of the core object's aggregation degree. Simultaneously, traverse all tumor cocoon structures in the digital pathology image to be processed, checking for the existence of a unit structure whose surrounding shell region does not contain target cells. If such a unit exists, it is defined as a completely non-enclosing unit. Calculate the proportion of completely non-enclosing units to the total number of tumor cocoon structure units as an indicator of special structural state.

[0040] Specifically, the proportion of the target cells within the enclosing shell to the total number of cells within the enclosing shell is statistically analyzed; in the case of multiple tumor cocoon structures, the statistical summary value of the corresponding proportions of all tumor cocoon structures is calculated as a quantitative indicator of enclosing ability; if the digital pathology image data contains multiple smallest units of tumor cocoon structures, the quantitative indicator of enclosing ability of each smallest structural unit is calculated separately, and the sample-level quantitative evaluation index is obtained by calculating the mean, median, or weighted mean.

[0041] A quantitative indicator of the clustering degree of core objects is calculated. For each core object, the area of ​​its convex hull in two-dimensional space is calculated using the Alpha Shape algorithm. This area represents the space occupied by the tumor nest. The spatial density of the core objects is obtained by dividing the total number of all core objects by the sum of the convex hull areas of all core objects, which serves as a quantitative indicator of the clustering degree of core objects.

[0042] Calculate the special structural state index. Traverse the smallest unit of all tumor cocoon structures in the sample, and check if there is a unit structure whose enclosing shell region does not contain the target cell (i.e., the target cell proportion is 0). This state represents that the target cell does not completely enclose the core object. Record whether there is such a special completely unenclosed unit in the sample, or calculate the proportion of such special units in the total number of units, as an index of special structural state.

[0043] Preferably, the tumor cocoon structure indicators are correlated with the patient's clinical variables through quantitative analysis. A recurrence risk prediction model is constructed using a generalized linear regression model to assess the patient's prognostic status, output the patient's recurrence risk probability, and divide the patient into a high-score group and a low-score group. The tumor cocoon structure indicators include quantitative indicators of envelopment capacity, quantitative indicators of aggregation degree, and indicators of specific structural states.

[0044] Specifically, a training set of ex vivo samples from non-small cell lung cancer patients with complete clinical follow-up data (including whether recurrence occurred, the time of recurrence, and survival status) was collected. Using the methods described in steps one through five above, tumor cocoon structure indices for each training set were calculated. Then, these indices were used as independent variables, and the patient's clinical variables were used as dependent variables to construct a recurrence risk prediction model using a generalized linear regression model. The parameters of the recurrence risk prediction model were fitted using methods such as maximum likelihood estimation to obtain the weight coefficients of each indices. The completed recurrence risk prediction model can calculate a recurrence risk probability value (i.e., a risk score) between 0 and 1 for any input data. To evaluate the model's predictive performance, receiver operating characteristic (ROC) curves were plotted and the area under the curve (AUC) was calculated. Furthermore, based on the risk score output by the recurrence risk prediction model, patients could be prognostically stratified using Kaplan-Meier survival analysis to assess the significant difference in progression-free survival (PFS) between the high-score and low-score groups.

[0045] Example 2

[0046] The following is a detailed description of a system for quantifying tumor cocoon structure in digital pathological images of ex vivo lung cancer slides. This system may include:

[0047] The information acquisition module is used to acquire digital pathological image data to be processed, which includes cell type annotation information and spatial coordinate information of each cell.

[0048] Specifically, the information acquisition module receives raw digital pathology image data and transforms it into a structured dataset that can be used for subsequent spatial topology calculations. In some embodiments, the working principle of the information acquisition module may include: acquiring raw digital pathology image data of ex vivo samples of non-small cell lung cancer tissue. This data can originate from any of the following: multiplex immunofluorescence images, multiplex immunohistochemical images, spatial transcriptome sequencing data, spatial proteome sequencing data, or conventional digital pathology image analysis. For image-type data, a preset cell segmentation algorithm is invoked to accurately segment the cells in the image, precisely identifying the nuclear boundary and cytoplasmic boundary of each cell, and calculating the geometric center point of the cell, which is used as the two-dimensional spatial coordinates of the cell, with units that can be pixels or converted to actual micrometers.

[0049] While performing cell segmentation and coordinate extraction, it is also necessary to perform phenotypic classification on each cell to obtain its type annotation information. In some embodiments, the cell type can be determined based on the signal intensity of different fluorescence channels or staining channels, combined with a preset expression threshold. For example, in a multiplex immunofluorescence scenario, channel images containing DNA dyes, epithelial cell marker proteins, tumor-associated fibroblast marker proteins, and proliferation marker proteins are acquired. The average fluorescence intensity of each segmented cell in each channel is calculated. If the fluorescence intensity of a cell in both the PANCK and Ki67 channels is higher than a preset background threshold, the cell can be annotated as a proliferating epithelial cell; if the fluorescence intensity of a cell in the SMA channel is higher than a preset threshold, the cell can be annotated as a tumor-associated fibroblast; and other cells that express only DAPI or other markers can be annotated as other cells.

[0050] The core object identification module is used to identify tumor cells based on the cell type annotation information and spatial coordinate information, and to determine the tumor nests formed by the aggregation of tumor cells through a spatial density clustering algorithm, which are then used as the core objects of the tumor cocoon structure.

[0051] Specifically, the core object identification module works by extracting the spatial coordinates of all cells labeled as tumor cells (such as proliferating epithelial cells) from the input image data. These coordinates are then fed into the DBSCAN clustering algorithm. Through the DBSCAN algorithm, spatially adjacent tumor cells that meet a threshold number are clustered into one or more clusters, while isolated tumor cells that do not meet the density requirements are removed. The remaining clusters are identified as core objects, biologically corresponding to tumor nests with invasive and proliferative capabilities.

[0052] The surrounding shell construction module is used to identify the edge cells of the core object and, based on the edge cells, construct a surrounding shell in the form of a ring, arc, or partially closed shape around the core object according to a preset distance threshold.

[0053] Specifically, the working principle of the enclosing shell region identification module may include: for each core object, identifying the edge cells within that core object using a nearest neighbor algorithm. Edge cells refer to tumor cells located at the outermost periphery of the tumor nest and in direct contact with the external matrix. After identifying the edge cells, using the spatial coordinates of these edge cells as a reference, calculating the spatial distances from other non-core object cells to the nearest edge cells. The region whose spatial distance to the edge cells meets a preset spatial range threshold is defined as the enclosing shell. The preset spatial range threshold can be set according to the specific biological background.

[0054] In certain application scenarios, tumor nests typically grow asymmetrically and irregularly, with numerous bumps and branches at their boundaries. By identifying edge cells and then extending them outwards at equal intervals, an irregular annular or band-shaped shell region that perfectly conforms to the irregular outline of the tumor nest can be constructed. This method avoids the missed or false positives caused by traditional methods using regular circles or concentric circles, and can accurately reflect the spatial interaction between tumor cells and stromal cells at the boundary.

[0055] The tumor cocoon structure definition module is used to select at least one cell type within the surrounding shell as the target cell, and to define the core object, the target cell, and the surrounding shell as a tumor cocoon structure.

[0056] The quantitative index calculation module is used to calculate the distribution characteristics of the target cells within the surrounding shell as a quantitative index of the enclosing ability of the target cells, and to calculate the ratio of the core object to the total space area as a quantitative index of the aggregation degree of the core object.

[0057] Specifically, the quantitative indicator calculation module counts the number of target cells within the smallest unit of each tumor cocoon structure (composed of a core object and its corresponding surrounding shell region) and calculates the proportion of this number to the total number of cells within that shell region. This proportion reflects the local enclosing ability of the target cells for that specific tumor nest. When the sample contains multiple smallest units, the module can also calculate the statistical summary value (such as mean, median, weighted mean, etc.) of the proportions corresponding to all units, serving as a quantitative indicator of enclosing ability.

[0058] The spatial area occupied by each core object is calculated by constructing an Alpha convex hull. Then, the ratio of the total number of all core objects in the sample to the total spatial area occupied by these core objects is used as a quantitative indicator of the degree of core object aggregation (i.e., spatial density).

[0059] Preferably, it also includes a special state recognition module, used to traverse all tumor cocoon structures in the digital pathology image to be processed, check whether there is a unit structure whose surrounding shell region does not contain target cells, if it exists, it is defined as a completely non-enclosed unit, and calculate the proportion of the completely non-enclosed unit to the total tumor cocoon structure units as a special structure state index.

[0060] Preferably, the tumor cocoon structure indicators are correlated with the patient's clinical variables through quantitative analysis, and a generalized linear regression model is used to fit the relationship between the tumor cocoon structure indicators and recurrence risk or survival prognosis. During the training of the recurrence risk prediction model, regularization techniques can be used for feature selection and to prevent overfitting. After training, for new test samples, their quantitative indicators can be input, and the model will automatically output a continuous risk score or recurrence risk probability.

[0061] In specific application scenarios, a recurrence risk prediction model can be constructed based on a multiplex immunofluorescence dataset containing 34 non-small cell lung cancer patients. A generalized linear regression model is constructed by using target cell encirclement ability indicators, core object aggregation degree indicators, and quantitative indicators of specific unit structures as joint predictors. Experimental results show that the model has extremely high accuracy in predicting postoperative recurrence, with an area under the receiver operating characteristic curve (AUC) of approximately 0.825, significantly outperforming models using only a single indicator. Furthermore, based on the predicted risk score, patients can be divided into high-risk and low-risk groups using Kaplan-Meier survival analysis and a Log-rank test can be performed to achieve precise stratification of patient prognosis.

[0062] Example 3

[0063] To further demonstrate the effectiveness and clinical application value of the method described in this application, a detailed explanation is provided below with specific experimental data and accompanying figures. This example is based on postoperative ex vivo tissue samples from 34 patients with non-small cell lung cancer (NSCLC). All samples were collected from patients with early-stage NSCLC who underwent surgical resection and had complete clinicopathological and follow-up data.

[0064] Step S1: Perform multiplex immunofluorescence (mIF) staining on these 34 samples. In this embodiment, the antibodies used include an antibody against PANCK (for labeling epithelial / tumor cells), an antibody against SMA (for labeling tumor-associated fibroblasts), and an antibody against Ki67 (for labeling proliferating cells). After each round of staining, imaging was performed using a high-resolution full-slide scanner.

[0065] Figure 1 This demonstrates a representative field of view in multiplex immunofluorescence imaging. Among them, Figure 1 Images A and B in the diagram show the results of imaging different proteins within the same field of view. Figure 1 In Figure A, the blue portion represents DNA dye (DAPI), the green portion represents the epithelial cell marker protein PANCK, and the yellow portion represents the tumor-associated fibroblast marker protein SMA. Figure 1 In Figure B, the yellow and blue dyes are the same as in Figure A, while the red portion represents the proliferative epithelial cell marker protein Ki67. By performing overlap analysis between the PANCK channels in Figure A and the Ki67 channels in Figure B, the cells in the overlapping area (i.e., cells simultaneously expressing PANCK and Ki67) were precisely identified as proliferative epithelial cells. Figure 1 As can be visually observed, the yellow tumor-associated fibroblasts surround the green proliferative epithelial cells in a localized or semi-closed manner, which is a direct manifestation of the tumor cocoon structure (CCS) defined in this application.

[0066] Step S2: Quantitatively calculate the mIF data of these 34 samples. First, use the DBSCAN clustering algorithm to identify the core objects. Figure 2 The calculation flow for the target cell encirclement ability when proliferating epithelial cells are used as the core object is shown. For example... Figure 2 As shown, blue dots represent the core group (i.e., proliferating epithelial cells), orange dots represent peripheral cells within the core group, red dots represent target cells (i.e., tumor-associated fibroblasts), and gray dots represent other microenvironment cells. The surrounding region is defined by the spatial aggregation contour of the target cells (as shown by the red dashed line, exhibiting an irregular boundary). Using the DBSCAN algorithm, scattered cells that did not form stable aggregation structures were successfully excluded, accurately locating tumor nests.

[0067] Step S3: Calculate the degree of aggregation of the core objects. Figure 3 A summary diagram of the core object aggregation degree calculation is shown in Figure 3. As shown, blue dots represent core objects (proliferating epithelial cells). The dashed box area represents tumor nests composed of core objects; the area occupied by the dashed portion is calculated by constructing an alpha convex hull. It is worth noting that the area included by the dashed portion may include other cells that are not core objects. The total area occupied by the core objects is obtained by summing the areas of all tumor nests on a tissue slice. The ratio of the total number of core objects to the total area is used as a quantitative indicator of the core object aggregation degree.

[0068] In addition, this application also identifies the special states of the smallest structural unit. Figure 4 This demonstrates different states of the smallest unit, the tumor cocoon structure. For example... Figure 4 As shown, Figure 4 The white dots represent other cells, the blue dots represent the core object (proliferating epithelial cells), the gold dots represent the cell types of the target cells surrounding the shell, and the red dots represent other cells surrounding the shell. Figure 4 In this context, A represents a special state where the target cell's cell type does not completely surround the core object (i.e., the number of gold dots inside the shell is 0). Figure 4 In the figure, B represents the state where the target cell type surrounds the core object (i.e., there are gold dots within the shell). The research process found that this special type of completely unenclosed unit (A in Figure 4) specifically exists in samples of NSCLC relapse and has extremely high clinical predictive value.

[0069] Step S4: Based on the calculated tumor cocoon structure indicators of 34 samples, and combined with the patients' clinical information, an in-depth statistical analysis was conducted. Figure 5 The differences in tumor cocoon structure parameters between the relapse and non-relapse groups were demonstrated. Among them, Figure 5 Figure A shows the statistical test results of the distribution and differences of the target cell encirclement ability index between the relapse and non-relapse groups. The results show that the encirclement index in the non-relapse group was significantly higher than that in the relapse group, indicating that the stronger fibroblast encirclement structure may act as a spatial barrier, limiting the migration and spread of tumor cells, thereby reducing the risk of relapse. Figure 5 Figure B in Figure 5 shows the distribution and differences in the clustering degree of core objects between the relapse and non-relapse groups. The results indicate that the clustering degree of core objects in the relapse group was significantly higher than that in the non-relapse group, suggesting that the denser the tumor cells are spatially, the stronger their invasiveness and the higher the risk of recurrence. Figure C in Figure 5 represents the distribution of relapse samples in different minimum unit states. The x-axis represents whether the sample contains a special minimum unit structure (i.e., a minimum unit not completely surrounded by target cells), and the y-axis represents the true relapse rate of the samples in each group. Figure 5 As can be clearly seen in C in this embodiment, all samples containing the special completely non-enclosed unit structure exhibit recurrence characteristics, further confirming the absolute effectiveness of this special structural state as a recurrence prediction indicator.

[0070] Step S5: In order to construct a recurrence risk prediction model, a generalized linear regression model was constructed based on the above indicators, and an ROC curve was plotted to evaluate its predictive efficacy. Figure 6 The ROC curve of the NSCLC model constructed based on the quantitative index of tumor cocoons is shown. Among them, Figure 6 In the figure, A represents the ROC curve of the model built solely based on the target cell encirclement ability, with an AUC (area under the curve) value of 0.687, indicating that a single encirclement ability index already possesses a certain predictive ability. Figure 6 In the figure, B represents the ROC curve of a model built solely based on the aggregation degree of core objects, with an AUC value of 0.718. Figure 6 In the figure, C represents the ROC curve of the composite model constructed based on all tumor cocoon structure indicators (combining envelopment ability, aggregation degree, and special structural state indicators), with an AUC value of 0.825. This result fully demonstrates that by integrating the spatial characteristics of tumor cocoon structure in multiple dimensions, the ability to discriminate the recurrence risk of non-small cell lung cancer can be significantly improved, and it has extremely high clinical application value.

[0071] Step S6: Use the constructed recurrence risk prediction model to perform prognostic assessment and survival analysis on patients. Based on the predicted values ​​(risk scores) generated by the generalized linear regression model, the 34 patients were divided into a high-score group and a low-score group. Figure 7 Kaplan-Meier survival curves based on tumor cocoon structure parameters are presented. The horizontal axis represents follow-up time (months), and the vertical axis represents progression-free survival (PFS). The log-rank test was used to compare the survival differences between the two groups, and the results showed p = 0.00101, which was highly statistically significant. Figure 7 It was observed that the high-scoring group (i.e., patients with a lower predicted risk of recurrence and more intact tumor cocoon structure) had significantly better progression-free survival than the low-scoring group (i.e., patients with a higher predicted risk of recurrence and fragmented tumor cocoon structure or completely unenclosed units). The survival curves of the two groups diverged early in the follow-up period and gradually widened over time, suggesting that the tumor cocoon structure formed between proliferating epithelial cells and CAFs is closely related to the risk of disease progression. This result indicates that the proposed tumor cocoon quantification index can not only accurately distinguish recurrence status but also has the potential value for precise risk stratification of patient prognosis, providing a scientific basis for clinicians to develop individualized follow-up and adjuvant therapy strategies.

[0072] The above description is only a preferred embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any equivalent substitutions or modifications made by those skilled in the art within the scope of the technology disclosed in the present invention, based on the technical solution and concept of the present invention, should be covered within the scope of protection of the present invention.

Claims

1. A method for quantitative processing of tumor cocoon structure in digital pathological images of ex vivo lung cancer sections, characterized in that, The method includes: Acquire digital pathological image data to be processed, the digital pathological image data including cell type annotation information and spatial coordinate information of each cell; Tumor cells are identified based on the cell type annotation information and spatial coordinate information, and tumor nests formed by the aggregation of tumor cells are determined by spatial density clustering algorithm, which are taken as the core objects of tumor cocoon structure. Identify the edge cells of the core object, and based on the edge cells, construct a ring-shaped, arc-shaped, or partially closed surrounding shell around the core object according to a preset distance threshold; At least one cell type within the surrounding shell is selected as the target cell, and the core object, the target cell, and the surrounding shell are defined as a tumor cocoon structure. The distribution characteristics of the target cells within the surrounding shell are calculated as a quantitative indicator of the enclosing ability of the target cells. The ratio of the core object to the total space area is calculated as a quantitative indicator of the aggregation degree of the core object.

2. The method according to claim 1, characterized in that, The core object is a cluster of tumor cells formed by proliferating epithelial cells; the target cells are tumor-associated fibroblasts; and the surrounding shell includes stromal cells, immune cells, and other microenvironment cell types.

3. The method according to claim 1, characterized in that, The steps for identifying the tumor nests include: The tumor cells were clustered using the density-based clustering algorithm DBSCAN to identify clusters of tumor cells in an aggregated state. Isolated or scattered tumor cells that have not formed an aggregate are removed, and the tumor cell clusters are identified as the core objects.

4. The method according to claim 1, characterized in that, The step of identifying the edge cells of the core object includes: using a nearest neighbor algorithm to calculate the spatial relationship between each cell in the core object and its surrounding cells, and identifying cells located at the geometric boundary of the core object as edge cells.

5. The method according to claim 1, characterized in that, The calculation method for the quantitative index of the enclosure capability includes: The proportion of the number of target cells within the surrounding shell to the total number of cells within the surrounding shell is statistically analyzed. In cases involving multiple tumor cocoon structures, the statistical summary value of the corresponding proportions of all tumor cocoon structures is calculated as a quantitative indicator of enclosing ability.

6. The method according to claim 1, characterized in that, The method further includes: traversing all tumor cocoon structures in the digital pathology image to be processed, checking whether there is a unit structure whose surrounding shell region does not contain target cells, if it exists, it is defined as a completely non-enclosed unit, and calculating the proportion of the completely non-enclosed unit to the total tumor cocoon structure units as a special structural state indicator.

7. The method according to claim 1, characterized in that, The cell type annotation information and the spatial coordinate information of each cell can be obtained through at least one of the following methods: multiplex immunofluorescence, multiplex immunohistochemistry, spatial transcriptomics, spatial proteomics, or digital pathological image analysis.

8. The method according to claim 1, characterized in that, The method further includes: performing quantitative correlation analysis between tumor cocoon structure indicators and patients' clinical variables, and constructing a recurrence risk prediction model using a generalized linear regression model to assess the patient's prognostic status; wherein the tumor cocoon structure indicators include quantitative indicators of envelopment capacity, quantitative indicators of aggregation degree, and indicators of special structural states.

9. A system for quantifying tumor cocoon structure in digital pathological images of ex vivo lung cancer slides, characterized in that, include: The information acquisition module is used to acquire digital pathological image data to be processed, the digital pathological image data including cell type annotation information and spatial coordinate information of each cell; The core object identification module is used to identify tumor cells based on the cell type annotation information and spatial coordinate information, and to determine the tumor nests formed by the aggregation of tumor cells through a spatial density clustering algorithm, which are used as the core objects of the tumor cocoon structure. The surrounding shell construction module is used to identify the edge cells of the core object and, based on the edge cells, construct a surrounding shell in the form of a ring, arc, or partially closed shape around the core object according to a preset distance threshold. The tumor cocoon structure definition module is used to select at least one cell type within the surrounding shell as the target cell, and to define the core object, the target cell and the surrounding shell as a tumor cocoon structure. The quantitative index calculation module is used to calculate the distribution characteristics of the target cells within the surrounding shell as a quantitative index of the enclosing ability of the target cells, and to calculate the ratio of the core object to the total space area as a quantitative index of the aggregation degree of the core object.

10. The system according to claim 9, characterized in that, It also includes a special state recognition module, which is used to traverse all tumor cocoon structures in the digital pathology image to be processed, check whether there is a unit structure whose surrounding shell region does not contain target cells. If it exists, it is defined as a completely non-enclosed unit. The proportion of the completely non-enclosed units to the total tumor cocoon structure units is calculated as a special structure state index.

Images