A method and device for quantifying tumor neural infiltration based on digital pathology whole slide images, and a medium

Abstract

This invention belongs to the field of medical image analysis and computational pathology, and discloses a method, device, and medium for quantifying tumor nerve invasion based on digital pathological panoramic slices. The method includes: acquiring digital pathological panoramic slices of tumors, performing multi-tissue semantic segmentation, and saving the panoramic segmentation mask; extracting candidate nerve region masks and pathological pixel images through connected components of the mask based on the coordinate mapping relationship between the mask and the panoramic slice; extracting features from the candidate nerve region masks and their pixel images; establishing and applying a classification model for the reliability of candidate nerve regions to filter and obtain reliable nerve region samples; calculating the epithelial wrapping index of a single nerve based on the epithelial tissue mask and nerve region mask corresponding to the reliable nerve region samples, and forming a quantitative score according to the level; and summarizing all nerve regions in the panoramic slice to form a quantitative index of tumor peri-tumor nerve invasion. This invention can achieve automated assessment of the degree of tumor peri-tumor nerve invasion in digital pathological panoramic slices.

Description

Technical Field

[0001] This invention relates to a method, device, and medium for quantifying tumor neural infiltration based on digital pathological panoramic slices, belonging to the field of medical image analysis and computational pathology technology. Background Technology

[0002] Pancreatic ductal adenocarcinoma is one of the most malignant digestive system tumors, often accompanied by high morbidity and mortality. Nerve invasion (PNI) is a pathological feature of pancreatic ductal adenocarcinoma, with an incidence exceeding 70%. Existing research indicates that nerve invasion in tumor tissue is closely related to tumor invasiveness, biological behavior, and patient prognosis. However, in the pathological analysis of pancreatic ductal adenocarcinoma, related diagnoses typically require pathologists to observe pathological slides for extended periods and interpret tumor type and grade based on their professional knowledge and experience. Limited by the subjective nature and relatively poor repeatability of manual identification methods, it is difficult to objectively and reproducibly quantify the spatial relationship between neural structures and tumor tissue. Most existing pathology reports only provide a qualitative conclusion regarding the presence or absence of nerve invasion, lacking a quantitative description of the degree of nerve invasion.

[0003] With the development of digital pathology technology, semantic segmentation models based on computer vision technology have been gradually applied to the automated analysis of tumor slides, laying the foundation for the quantitative assessment of the tumor microenvironment. Patent CN119992552B discloses a seamless semantic segmentation method for digital pathology slides, which can be used for the automatic segmentation of key components of the tumor microenvironment such as epithelium and nerve bundles. However, a quantitative analysis method for PNI (Potential Inclusion Nitrogen) has not yet been established. More importantly, nerve bundles and some tumor stroma components are extremely similar in appearance, leading to a large number of false positives in the segmentation results, making it impossible to form an accurate PNI score. Summary of the Invention

[0004] The purpose of this invention is to provide a method, device, and medium for quantifying tumor nerve invasion based on digital pathological panoramic slides. By constructing quantitative indicators through the spatial morphological characteristics of nerve tissue and epithelial tissue in digital pathological slides, the degree of invasion of nerve regions in pancreatic cancer pathological slides can be automatically identified, providing standardized data support for pathological diagnosis, clinical efficacy prediction, and scientific research analysis, while improving the interpretability and reliability of model predictions.

[0005] To achieve the above objectives, the present invention is implemented using the following technical solution.

[0006] On the one hand, the present invention provides a method for quantifying tumor neural infiltration based on digital pathological panoramic slides, comprising:

[0007] Obtain panoramic digital pathology slice images of tumors and perform multi-tissue category semantic segmentation to obtain whole-slice semantic segmentation results; store the semantic segmentation results as a mask file;

[0008] Based on the mapping relationship between the pixel coordinates of the mask image in the mask file and the pixel coordinates of the digital pathology panoramic slice image, the pixel image corresponding to the candidate neural region in the mask file is cropped from the digital pathology panoramic slice image.

[0009] Feature extraction is performed on the candidate neural regions and their corresponding pixel images in the mask file, and a machine learning model is used for classification to obtain reliable segmentation samples;

[0010] The segmented reliable samples are mapped to a mask file and a digital pathological panoramic slice image through a mapping relationship. The epithelial tissue mask is extracted from the mask file, and the epithelial wrapping index of a single nerve is calculated in combination with the nerve region mask. The epithelial wrapping index is the ratio of the number of radial rays that effectively contact the epithelial tissue out of multiple radial rays emitted from the nerve centroid to the periphery of a single nerve to the total number of rays. It is used to quantify the degree of wrapping of the single nerve structure by the tumor epithelial component in local space.

[0011] Based on the epithelial wrapping index of the single nerve, the degree of epithelial wrapping of the single nerve is classified according to grade, and the infiltration grade of the single nerve is obtained as the quantitative result of nerve infiltration.

[0012] Optionally, the multi-tissue category semantic segmentation is implemented based on an existing multi-tissue category semantic segmentation model that includes neural and epithelial tissue categories;

[0013] The whole-slice semantic segmentation results include neural region segmentation results and epithelial tissue segmentation results.

[0014] Optionally, the process of cropping the pixel image corresponding to the candidate neural region in the mask file from the digital pathology panoramic slice image includes:

[0015] Construct the mapping relationship between the pixel coordinates of the mask image in the mask file and the pixel coordinates of the digital pathology panoramic slide image:

[0016] pixel coordinates of the mask image Mapped to pixel coordinates of digital pathology panoramic slide images The mapping relationship is expressed as:

[0017] ;

[0018] ;

[0019] In the formula, and These are the spatial resolution parameters for digital pathology panoramic slide images and mask images, respectively.

[0020] Obtain the mask of the candidate neural region in the mask file, and use the 8-connected region method to extract the set of pixels of the same type for each candidate neural region to determine the complete range of the corresponding candidate neural region.

[0021] Using the center pixel of each candidate neural region as the center, extract the corresponding candidate neural region mask, and crop the candidate neural region into a binary image of a fixed size pixel and save it. The position in the binary image with a gray value equal to 255 is the neural region.

[0022] Based on the mapping relationship, pixel images corresponding to each candidate neural region are cropped from the digital pathology panoramic slice image.

[0023] Optionally, feature extraction of the candidate neural regions and the corresponding pixel images in the mask file includes: extracting the morphological features of the candidate neural regions and the texture features of the corresponding pixel images;

[0024] The morphological features include the area, roundness, eccentricity, axial length ratio, and boundary Fourier frequency components of the neural region, which are used to characterize the geometric structure and contour features of the neural region and reflect the size, boundary complexity, and other characteristics of the neural region.

[0025] The texture features include grayscale nonuniformity, short run emphasis, long run emphasis, joint entropy, and differential entropy, which are used to characterize the color distribution characteristics and inter-pixel dependencies within the neural region, and to depict the texture complexity and spatial organization characteristics of the neural region.

[0026] Optionally, the machine learning model is a support vector machine model;

[0027] The classification process is as follows:

[0028] The pixel image is manually verified to obtain manually verified reliable segmentation samples and unreliable segmentation samples; unreliable segmentation samples are those whose segmentation results are not neural regions or whose tissue structures are unreasonable, as confirmed by manual verification.

[0029] During the training phase of the support vector machine model, the mapping relationship between neural region features and category labels is learned based on the morphological features, texture features, and manual inspection results.

[0030] During the support vector machine model prediction stage, the trained support vector machine model is used to determine whether a neural region belongs to a reliable segmentation sample.

[0031] Optionally, the calculation process for the epithelial wrapping index of the single nerve includes:

[0032] The segmented reliable samples are mapped to mask files and digital pathological panoramic slide images through a mapping relationship to obtain mask images, which include neural region mask images and epithelial tissue mask images;

[0033] In the mask image, construct an indicator function. When the grayscale value is 255, ,otherwise ;

[0034] The centroid coordinates of the mask image region are obtained by weighted averaging the pixel coordinates based on the indicated function. :

[0035] ;

[0036] ;

[0037] Centered on the centroid coordinates of the mask image for each neural region, within a range of 0° to 360°, according to a preset angle. Constructing radial rays, the angle step size ;

[0038] The minimum circumscribed rectangle extension of each neural region As the search range, the pixel coordinates on the ray are represented as follows:

[0039] ;

[0040] In the formula, The length of the current ray;

[0041] The pixel point on each ray that just exits the neural region is recorded as the outermost boundary point of the neural region. , Boundary points Axis coordinates Boundary points Axis coordinates;

[0042] Starting from the outermost boundary point, continue searching along the corresponding ray direction to detect whether there are any pixels belonging to the epithelial tissue mask region on the ray; when an epithelial tissue pixel is detected... When this occurs, record the ray as a valid contact ray; Boundary points Axis coordinates Boundary points Axis coordinates;

[0043] Based on the number of effective contact rays The epithelial wrapping index of a single nerve was calculated by comparing it with the total number of rays. , .

[0044] Optionally, the process of classifying the degree of epithelial wrapping of a single nerve includes:

[0045] By calculating the upper boundary points of each neural region and epithelial tissue pixels Minimum Euclidean distance between This yields a measure of the minimum distance between neural tissue and epithelial tissue.

[0046] Introduce a pixel distance threshold, when the minimum Euclidean distance When the distance is less than the pixel distance threshold, it indicates that the nerve is a paraepithelial nerve of a tumor, and the minimum Euclidean distance is used. Multiply by the spatial resolution of the mask file to obtain the true distance. ;

[0047] When the actual distance At that time, based on the epithelial wrapping index of a single nerve The degree of epithelial wrapping around a single nerve is precisely divided into five levels:

[0048] Level 1: It was determined to be without epithelial tissue surrounding it;

[0049] Level 2: It was determined to be a mild epithelial tissue encapsulation;

[0050] Level 3: It was determined to be moderate epithelial tissue encapsulation;

[0051] Level 4: It was determined to be severely encased in epithelial tissue;

[0052] Level 5: It was determined to be severely surrounded by epithelial tissue.

[0053] Optionally, the method for quantifying tumor nerve infiltration based on digital pathology panoramic slices further includes: setting overall scoring indicators and local scoring indicators, and assessing the patient's prognosis based on the indicators;

[0054] For each patient, nerves with a non-zero epithelial wrapping index were selected as effective wrapped nerve units, and the number of effective wrapped nerve units was calculated. According to the total wrapping ratio of the effective wrapped neural units The average proportion of the patient's overall nerves that were wrapped was calculated. ;

[0055] For each patient, the neural region with the smallest distance between the nerve tissue and the epithelial tissue was selected as a candidate target. Epithelial wrapping indices of a single nerve were extracted from these candidate targets. The candidates were then sorted in descending order according to these epithelial wrapping indices. The top N targets were selected, and the average epithelial wrapping ratio of these top N targets was calculated as the local nerve wrapping characteristic for that patient. When the number of objects within a candidate target is less than N, the average value of the epithelial wrapping ratio of all objects within the candidate target is calculated.

[0056] Based on the average proportion of nerves encased in the patient's overall body Patients were divided into two groups according to the severity of their illness, and each group was assigned a score based on an overall scoring index. ;when At that time, assign an overall scoring index ;when At that time, assign an overall scoring index ;

[0057] Based on the patient's local nerve encapsulation characteristics Patients were divided into two groups according to the severity of their illness, and each group was assigned a local scoring index. ;when At that time, assign local scoring indicators ;when At that time, assign local scoring indicators ;

[0058] Score the overall rating indicators Compared with local scoring indicators The scores were summed to obtain the degree of nerve infiltration in the patients. ;

[0059] Patients were stratified for risk based on their nerve infiltration scores, and survival data were combined with this data to assess the survival differences between different risk stratification groups, in order to determine whether the quantification of nerve infiltration had a significant ability to differentiate patient prognosis.

[0060] Secondly, the present invention provides a device for quantifying tumor neural infiltration based on digital pathological panoramic slides, comprising:

[0061] The semantic segmentation module is used to: acquire panoramic digital pathology slice images of tumors and perform multi-tissue category semantic segmentation to obtain whole-slice semantic segmentation results; and store the semantic segmentation results as a mask file.

[0062] The neural infiltration quantization module is used to: based on the mapping relationship between the pixel coordinates of the mask image in the mask file and the pixel coordinates of the digital pathological panoramic slice image, crop out the pixel image corresponding to the candidate neural region in the mask file from the digital pathological panoramic slice image;

[0063] Feature extraction is performed on the candidate neural regions and their corresponding pixel images in the mask file, and a machine learning model is used for classification to obtain reliable segmentation samples;

[0064] The segmented reliable samples are mapped to a mask file and a digital pathological panoramic slice image through a mapping relationship. The epithelial tissue mask is extracted from the mask file, and the epithelial wrapping index of a single nerve is calculated in combination with the nerve region mask. The epithelial wrapping index is the ratio of the number of radial rays that effectively contact the epithelial tissue out of multiple radial rays emitted from the nerve centroid to the periphery of a single nerve to the total number of rays. It is used to quantify the degree of wrapping of the single nerve structure by the tumor epithelial component in local space.

[0065] Based on the epithelial wrapping index of the single nerve, the degree of epithelial wrapping of the single nerve is classified according to grade, and the infiltration grade of the single nerve is obtained as the quantitative result of nerve infiltration.

[0066] Thirdly, the present invention provides a computer-readable storage medium having a computer program / instruction stored thereon, which, when executed by a processor, implements the steps of the method for quantifying tumor nerve infiltration based on digital pathological panoramic slices as described in any of the first aspects.

[0067] Beneficial effects

[0068] 1. This invention can construct quantitative indicators based on the spatial morphological characteristics of nerve tissue and epithelial tissue in digital pathological sections, and automatically identify the degree of invasion of nerve regions in pancreatic cancer pathological sections, providing standardized data support for pathological diagnosis, clinical efficacy prediction and scientific research analysis, while improving the interpretability and reliability of model prediction.

[0069] 2. This invention can be integrated into a computer-aided diagnostic (CAD) system and combined with deep learning models such as semantic segmentation. It can visualize the lesion area and achieve automated assessment of the degree of nerve infiltration in the slice through quantitative indicators. Attached Figure Description

[0070] Figure 1 This is a flowchart of the tumor nerve infiltration quantification method based on digital pathological panoramic slides according to the present invention.

[0071] Figure 2 This is a flowchart illustrating the annotation and segmentation of reliable samples in this invention;

[0072] Figure 3 This is a schematic diagram illustrating the quantitative spatial distribution of nerve and epithelial tissues according to the present invention.

[0073] Figure 4 This is a schematic diagram showing the visualization results of the degree to which a single nerve is wrapped in the present invention;

[0074] Figure 5 This is a schematic diagram of the survival probability curves for individuals with different levels of neural infiltration according to the present invention. Detailed Implementation

[0075] The technical solution of the present invention will be described in detail below with reference to the accompanying drawings and specific embodiments. It should be understood that the embodiments of the present invention and the specific features in the embodiments are detailed descriptions of the technical solution of the present invention, rather than limitations thereof. In the absence of conflict, the embodiments of the present invention and the technical features in the embodiments can be combined with each other.

[0076] Example 1

[0077] This embodiment introduces a method for quantifying tumor neural infiltration based on digital pathology panoramic slides, such as... Figure 1 As shown, it includes:

[0078] Obtain panoramic digital pathology slice images of tumors and perform multi-tissue category semantic segmentation to obtain whole-slice semantic segmentation results; store the semantic segmentation results as a mask file;

[0079] Based on the mapping relationship between the pixel coordinates of the mask image in the mask file and the pixel coordinates of the digital pathology panoramic slice image, the pixel image corresponding to the candidate neural region in the mask file is cropped from the digital pathology panoramic slice image.

[0080] Feature extraction is performed on the candidate neural regions and their corresponding pixel images in the mask file, and a machine learning model is used for classification to obtain reliable segmentation samples;

[0081] The segmented reliable samples are mapped to a mask file and a digital pathological panoramic slice image through a mapping relationship. The epithelial tissue mask is extracted from the mask file, and the epithelial wrapping index of a single nerve is calculated in combination with the nerve region mask. The epithelial wrapping index is the ratio of the number of radial rays that effectively contact the epithelial tissue out of multiple radial rays emitted from the nerve centroid to the periphery of a single nerve to the total number of rays. It is used to quantify the degree of wrapping of the single nerve structure by the tumor epithelial component in local space.

[0082] Based on the epithelial wrapping index of the single nerve, the degree of epithelial wrapping of the single nerve is classified according to grade, and the infiltration grade of the single nerve is obtained as the quantitative result of nerve infiltration.

[0083] This embodiment quantifies tumor nerve infiltration based on pancreatic ductal adenocarcinoma. The specific process includes:

[0084] I. Obtaining panoramic pathological slide images and results of multi-tissue category semantic segmentation models

[0085] Digital pathological slides of pancreatic ductal adenocarcinoma and patient clinical data were collected. Multi-tissue category semantic segmentation was performed on the whole-field-of-sight (WSIs) slides. This semantic segmentation was based on an existing multi-tissue category semantic segmentation model, yielding a spatially corresponding whole-slide semantic segmentation result, including pixel-level segmentation results for neural regions and epithelial tissues. These semantic segmentation results were stored as a TIFF file as a mask file. This step can use any semantic segmentation model that includes neural and epithelial tissue categories.

[0086] 2. Post-processing of the semantic segmentation model results: Based on feature extraction and SVM classification, extract reliable segmentation samples.

[0087] The core purpose of this step is to improve the accuracy of neural segmentation by filtering out typical false detections from the deep learning model in the previous step through an additional image feature-based machine learning model.

[0088] Sample collection: based on the spatial resolution parameters of WSIs and mask files. and The mapping relationship between the mask pixel coordinates and the WSIs pixel coordinates is constructed. Specifically, the method includes: mapping the mask pixels... Mapping to WSIs coordinates The mapping relationship can be expressed as ; Ensure that the spatial position of each tissue in the mask is consistent with that of the WSIs.

[0089] Obtain the mask of the candidate neural region from the mask file. Use the octal connected component method to extract the set of pixels of the same type for each candidate neural region and determine the complete range of the corresponding candidate neural region. Take the center pixel of each candidate neural region as the center, extract the mask of the corresponding candidate neural region, and crop the candidate neural region into a binary image of a fixed size of 256×256 pixels and save it. The position of the gray value equal to 255 in the binary image is the neural region.

[0090] The corresponding 256×256 pixel image is cropped from the original WSIs according to the mapping relationship. This is because the resolution of WSIs is 0.25. A 256×256 image size not only ensures the integrity of the neural region but also includes some information about the surrounding nerves.

[0091] The images above were manually verified, and the verification results were divided into reliable segmentation samples and unreliable segmentation samples. Reliable segmentation samples were those whose segmentation results were manually confirmed to be neural regions, while unreliable segmentation samples were those whose segmentation results were not neural regions or whose tissue structures were unreasonable.

[0092] Feature extraction was performed on the 256×256 image, selecting morphological features of candidate neural regions and texture features of the corresponding pixel images. Morphological features include parameters characterizing the geometric structure and contour of the neural regions, such as area, roundness, eccentricity, aspect ratio, and boundary Fourier frequency components, reflecting characteristics like size and boundary complexity. Texture features include parameters characterizing color distribution characteristics and inter-pixel dependencies within the neural regions, such as grayscale non-uniformity, short run emphasis, long run emphasis, joint entropy, and differential entropy, depicting the texture complexity and spatial organization of the neural regions.

[0093] Support Vector Machine (SVM) was selected as the classification model. During the model training phase, based on the aforementioned features and manual verification results, the mapping relationship between neural region features and their class labels was learned. The training set to test set ratio was 7:3, resulting in a final model accuracy of 96.5%. In the subsequent model prediction phase, the trained SVM model was used to determine whether a region belongs to a reliable segmentation sample.

[0094] The specific process for labeling and segmenting reliable samples is as follows: Figure 2 As shown, the method for labeling and segmenting reliable samples described in this embodiment can further eliminate the problem of high positive rate in semantic segmentation models and improve the accuracy of semantic segmentation models.

[0095] III. Constructing epithelial wrapping indices for single nerves from reliably segmented samples

[0096] In each 256×256 mask image, an indicator function is constructed. When the grayscale value is 255, ,otherwise ;

[0097] The centroid coordinates of the mask image region are obtained by weighted averaging the pixel coordinates based on the indicated function. :

[0098] ;

[0099] ;

[0100] The extracted reliable segmentation samples are mapped to the original mask file and WSIs file using coordinate mapping, and the epithelial tissue mask is extracted from the mask file.

[0101] Centered on the centroid coordinates of the mask image for each neural region, within a range of 0° to 360°, according to a preset angle. Construct radial rays, angular step size ;

[0102] The minimum circumscribed rectangle extension of each neural region As the search range, the pixel coordinates on the ray are represented as follows:

[0103] ;

[0104] In the formula, The length of the current ray;

[0105] The pixel point on each ray that just exits the neural region is recorded as the outermost boundary point of the neural region. , Boundary points Axis coordinates Boundary points Axis coordinates;

[0106] Starting from the outermost boundary point, continue searching along the corresponding ray direction to detect whether there are any pixels belonging to the epithelial tissue mask region on the ray; when an epithelial tissue pixel is detected... When this occurs, record the ray as a valid contact ray; Boundary points Axis coordinates Boundary points Axis coordinates;

[0107] Based on the number of effective contact rays The epithelial wrapping index of a single nerve was calculated by comparing it with the total number of rays. , .

[0108] like Figure 3 As shown, the red area represents nerves, the gray area represents epithelial tissue, and the remaining area represents other tissues.

[0109] IV. Derivation and visualization of the degree of epithelial wrapping of a single nerve root

[0110] By calculating the upper boundary points of each neural region and epithelial tissue pixels Minimum Euclidean distance between This method obtains the minimum distance metric between two types of organizations to characterize their spatial proximity. Because semantic segmentation labels are mutually exclusive, each pixel can only belong to one category. A preset pixel distance threshold T=3 is introduced as a calculation reference. This threshold is calculated based on the minimum Euclidean distance. This indicates that the nerve is a paraepithelial nerve of the tumor. Spatial resolution of tif files ( Multiply by ( / pixel) to get the true distance. .

[0111] When the actual distance At that time, based on the epithelial wrapping index of a single nerve The degree of epithelial wrapping around a single nerve is precisely divided into five levels:

[0112] Level 1: It was determined to be without epithelial tissue surrounding it;

[0113] Level 2: It was determined to be a mild epithelial tissue encapsulation;

[0114] Level 3: It was determined to be moderate epithelial tissue encapsulation;

[0115] Level 4: It was determined to be severely encased in epithelial tissue;

[0116] Level 5: It was determined to be severely encased in epithelial tissue;

[0117] The degree of epithelial wrapping around the aforementioned single nerves was mapped to colors, and visualized on digital pathology images using an ASAP interface (Python) via XML files. The specific color mapping rules are as follows:

[0118] Level 1: Light green (#AAFF00), Level 2: Green (#00FF00), Level 3: Yellow (#FFFF00), Level 4: Orange (#FFA500), Level 5: Red (#FF0000). The specific visualization results are as follows: Figure 4 As shown.

[0119] V. Establish overall and local scoring indicators, and assess the patient's prognosis based on these indicators.

[0120] The survival period of pancreatic ductal adenocarcinoma patients corresponding to digital pathological images of pancreatic ductal adenocarcinoma was obtained. The survival period of pancreatic cancer patients is defined as the time period from the date of initial diagnosis of pancreatic ductal adenocarcinoma to the date of death from any cause. For patients whose death date was not recorded, the date of their last follow-up was used as the end point of time.

[0121] Given the highly malignant nature and poor overall prognosis of pancreatic ductal adenocarcinoma, and considering that the vast majority of patients experience endpoint events within 3 years of diagnosis, resulting in a low proportion of long-term survivors, directly using full follow-up events for statistical analysis would lead to a small tail sample size in the long term, potentially causing unstable survival estimates. Therefore, this example sets a fixed upper limit of 3 years for observation events, within which patient prognosis is uniformly assessed.

[0122] For each patient, nerves with a non-zero epithelial wrapping index were selected as effective wrapped nerve units, and the number of effective wrapped nerve units was calculated. According to the total wrapping ratio of the effective wrapped neural units The average proportion of the patient's overall nerves that were wrapped was calculated. ;

[0123] For each patient, the neural region with the smallest distance between the nerve tissue and the epithelial tissue was selected as a candidate target. Epithelial wrapping indices of a single nerve were extracted from these candidate targets. The candidates were then sorted in descending order according to these epithelial wrapping indices. The top 50 targets were selected, and the average epithelial wrapping ratio of these top 50 targets was calculated as the local nerve wrapping characteristic of that patient. When the number of objects in the candidate target is less than 50, the average value of the epithelial wrapping ratio of all objects in the candidate target is calculated.

[0124] Based on the average proportion of nerves encased in the patient's overall body Patients were divided into two groups according to the severity of their illness, and each group was assigned a score based on an overall scoring index. ;when At that time, assign an overall scoring index ;when At that time, assign an overall scoring index ;

[0125] Based on the patient's local nerve encapsulation characteristics Patients were divided into two groups according to the severity of their illness, and each group was assigned a local scoring index. ;when At that time, assign local scoring indicators ;when At that time, assign local scoring indicators ;

[0126] Score the overall rating indicators Compared with local scoring indicators The scores were summed to obtain the degree of nerve infiltration in the patients. ;

[0127] Based on the patient's nerve infiltration score The degree of nerve infiltration in different patients was calculated using the Kaplan-Meier survival method, based on patient survival data. The survival probability curves were obtained, and the log-rank test was used to statistically compare the differences in component survival to determine whether the quantitative results of the neural infiltration had a significant ability to distinguish patient prognosis.

[0128] A Cox proportional hazards model was constructed to assess the strength of the association between the level of infiltration and the risk of patient mortality. By calculating the hazard ratio and its 95% confidence interval, the difference in mortality risk between the high-infiltration group and the low-infiltration group was quantitatively reflected.

[0129] In this embodiment, both the patient's digital pathology images and clinical information are obtained from the publicly available database TCGA. The final score is divided into three groups. The value range is 3, 4, 5, based on the score. Different values ​​of the score will divide patients into different groups of nerve infiltration grades. At that time, it was determined to be a low-risk nerve infiltration. At that time, it was determined to be a medium-risk nerve infiltration, and the score was... At that time, patients were classified as having high-risk nerve infiltration. Analysis showed that the median survival was 652 days in the low-risk nerve infiltration group, 481 days in the medium-risk group, and 297 days in the high-risk group, with an overall median survival of 511 days. The log-rank test results indicated a statistically significant difference in survival curves among the three groups (p=0.014495867). The Kaplan-Mel survival curves for different nerve infiltration severity scores are shown in the figure below. Figure 5 As shown in the figure. The analysis results show that the hazard ratio of the infiltration score is 1.38, the 95% confidence interval is 1.11-1.72, and the corresponding statistical significance level is p=0.00398 (p<0.005), which is statistically significant.

[0130] Example 2

[0131] Based on the same inventive concept as Embodiment 1, this embodiment introduces a device for quantifying tumor neural infiltration based on digital pathological panoramic slides, comprising:

[0132] The semantic segmentation module is used to: acquire panoramic digital pathology slice images of tumors and perform multi-tissue category semantic segmentation to obtain whole-slice semantic segmentation results; and store the semantic segmentation results as a mask file.

[0133] The neural infiltration quantization module is used to: based on the mapping relationship between the pixel coordinates of the mask image in the mask file and the pixel coordinates of the digital pathological panoramic slice image, crop out the pixel image corresponding to the candidate neural region in the mask file from the digital pathological panoramic slice image;

[0134] Feature extraction is performed on the candidate neural regions and their corresponding pixel images in the mask file, and a machine learning model is used for classification to obtain reliable segmentation samples;

[0135] The segmented reliable samples are mapped to a mask file and a digital pathological panoramic slice image through a mapping relationship. The epithelial tissue mask is extracted from the mask file, and the epithelial wrapping index of a single nerve is calculated in combination with the nerve region mask. The epithelial wrapping index is the ratio of the number of radial rays that effectively contact the epithelial tissue out of multiple radial rays emitted from the nerve centroid to the periphery of a single nerve to the total number of rays. It is used to quantify the degree of wrapping of the single nerve structure by the tumor epithelial component in local space.

[0136] Based on the epithelial wrapping index of the single nerve, the degree of epithelial wrapping of the single nerve is classified according to grade, and the infiltration grade of the single nerve is obtained as the quantitative result of nerve infiltration.

[0137] The specific functions of each module described above are explained in the relevant content of the method in Embodiment 1, and will not be repeated here.

[0138] Example 3

[0139] Based on the same inventive concept as other embodiments, this embodiment describes a computer-readable storage medium having a computer program / instruction stored thereon, which, when executed by a processor, implements the steps of the method for quantifying tumor neural infiltration based on digital pathological panoramic slices as described in any of Embodiment 1.

[0140] Those skilled in the art will understand that embodiments of the present invention can be provided as methods, systems, or computer program products. Therefore, the present invention can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, the present invention can take the form of a computer program product embodied on one or more computer-usable storage media (including, but not limited to, disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0141] This invention is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of the invention. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, generate instructions for implementing the flowchart illustrations and / or block diagrams. Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.

[0142] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1 The function specified in one or more boxes.

[0143] These computer program instructions may also be loaded onto a computer or other programmable data processing equipment to cause a series of operational steps to be performed on the computer or other programmable equipment to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable equipment for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.

[0144] The embodiments of the present invention have been described above with reference to the accompanying drawings. However, the present invention is not limited to the specific embodiments described above. The specific embodiments described above are merely illustrative and not restrictive. Those skilled in the art can make many other forms under the guidance of the present invention without departing from the spirit and scope of the claims. All of these forms are within the protection scope of the present invention.

Claims

1. A method for quantifying tumor neural infiltration based on digital pathological panoramic slides, characterized in that, include: Obtain panoramic digital pathology slice images of tumors and perform multi-tissue category semantic segmentation to obtain the semantic segmentation results of the whole slice; Store the semantic segmentation results as a mask file; Based on the mapping relationship between the pixel coordinates of the mask image in the mask file and the pixel coordinates of the digital pathology panoramic slice image, the pixel image corresponding to the candidate neural region in the mask file is cropped from the digital pathology panoramic slice image. Feature extraction is performed on the candidate neural regions and their corresponding pixel images in the mask file, and a machine learning model is used for classification to obtain reliable segmentation samples; The segmented reliable samples are mapped to a mask file and a digital pathological panoramic slice image through a mapping relationship. The epithelial tissue mask is extracted from the mask file, and the epithelial wrapping index of a single nerve is calculated in combination with the nerve region mask. The epithelial wrapping index is the ratio of the number of radial rays that effectively contact the epithelial tissue out of multiple radial rays emitted from the nerve centroid to the periphery of a single nerve to the total number of rays. It is used to quantify the degree of wrapping of the single nerve structure by the tumor epithelial component in local space. Based on the epithelial wrapping index of the single nerve, the degree of epithelial wrapping of the single nerve is classified according to grade, and the infiltration grade of the single nerve is obtained as the quantitative result of nerve infiltration.

2. The method of quantifying tumor neural infiltration based on digital pathology whole-slide according to claim 1, wherein, The multi-tissue category semantic segmentation is based on an existing multi-tissue category semantic segmentation model that includes neural and epithelial tissue categories; The whole-slice semantic segmentation results include neural region segmentation results and epithelial tissue segmentation results.

3. The method for quantifying tumor neural infiltration based on digital pathological panoramic slides according to claim 1, characterized in that, The process of cropping the pixel image corresponding to the candidate neural region in the mask file from the digital pathological panoramic slice image includes: Construct the mapping relationship between the pixel coordinates of the mask image in the mask file and the pixel coordinates of the digital pathology panoramic slide image: pixel coordinates of the mask image Mapped to pixel coordinates of digital pathology panoramic slide images The mapping relationship is expressed as: ； ； In the formula, and These are the spatial resolution parameters for digital pathology panoramic slide images and mask images, respectively. Obtain the mask of the candidate neural region in the mask file, and use the 8-connected region method to extract the set of pixels of the same type for each candidate neural region to determine the complete range of the corresponding candidate neural region. Using the center pixel of each candidate neural region as the center, extract the corresponding candidate neural region mask, and crop the candidate neural region into a binary image of a fixed size pixel and save it. The position in the binary image with a gray value equal to 255 is the neural region. Based on the mapping relationship, pixel images corresponding to each candidate neural region are cropped from the digital pathology panoramic slice image.

4. The method for quantifying tumor neural infiltration based on digital pathological panoramic slides according to claim 1, characterized in that, Feature extraction of the candidate neural regions and their corresponding pixel images in the mask file includes: extracting the morphological features of the candidate neural regions and the texture features of the pixel images corresponding to the candidate neural regions; The morphological features include the area, roundness, eccentricity, axial length ratio, and boundary Fourier frequency components of the neural region, which are used to characterize the geometric structure and contour features of the neural region. The texture features include grayscale nonuniformity, short run emphasis, long run emphasis, joint entropy, and differential entropy, which are used to characterize the color distribution characteristics within neural regions and the inter-pixel dependencies.

5. The method for quantifying tumor neural infiltration based on digital pathological panoramic slides according to claim 4, characterized in that, The machine learning model used is a support vector machine model; The classification process is as follows: The pixel image is manually verified to obtain manually verified reliable segmentation samples and unreliable segmentation samples; During the training phase of the support vector machine model, the mapping relationship between neural region features and category labels is learned based on the morphological features, texture features, and manual inspection results. During the support vector machine model prediction stage, the trained support vector machine model is used to determine whether a neural region belongs to a reliable segmentation sample.

6. The method for quantifying tumor neural infiltration based on digital pathological panoramic slides according to claim 1, characterized in that, The calculation process for the epithelial wrapping index of the single nerve includes: The segmented reliable samples are mapped to mask files and digital pathological panoramic slide images through a mapping relationship to obtain mask images, which include neural region mask images and epithelial tissue mask images; In the mask image, construct an indicator function. When the grayscale value is 255, ,otherwise ; Based on the indicator function, pixel coordinates By performing a weighted average, the centroid coordinates of the mask image region are obtained. : ； ； Centered on the centroid coordinates of the mask image for each neural region, within a range of 0° to 360°, according to a preset angle. Construct radial rays, angular step size ; The minimum circumscribed rectangle extension of each neural region As the search range, the pixel coordinates on the ray are represented as follows: ； In the formula, The length of the current ray; The pixel point on each ray that just exits the neural region is recorded as the outermost boundary point of the neural region. , Boundary points Axis coordinates Boundary points Axis coordinates; Starting from the outermost boundary point, continue searching along the corresponding ray direction to detect whether there are any pixels belonging to the epithelial tissue mask region on the ray; when an epithelial tissue pixel is detected... When this occurs, record the ray as a valid contact ray; Boundary points Axis coordinates Boundary points Axis coordinates; Based on the number of effective contact rays The epithelial wrapping index of a single nerve was calculated by comparing it to the total number of rays. , .

7. The method for quantifying tumor neural infiltration based on digital pathological panoramic slides according to claim 6, characterized in that, The process of classifying the degree of epithelial wrapping around a single nerve includes: By calculating the upper boundary points of each neural region and epithelial tissue pixels Minimum Euclidean distance between This yields a measure of the minimum distance between neural tissue and epithelial tissue. Introduce a pixel distance threshold, when the minimum Euclidean distance When the distance is less than the pixel distance threshold, it indicates that the nerve is a paraepithelial nerve of a tumor, and the minimum Euclidean distance is used. Multiply by the spatial resolution of the mask file to obtain the true distance. ; When the actual distance At that time, based on the epithelial wrapping index of a single nerve The degree of epithelial wrapping around a single nerve is precisely divided into five levels: Level 1: It was determined to be without epithelial tissue surrounding it; Level 2: It was determined to be a mild epithelial tissue encapsulation; Level 3: It was determined to be moderate epithelial tissue encapsulation; Level 4: It was determined to be severely encased in epithelial tissue; Level 5: It was determined to be severely surrounded by epithelial tissue.

8. The method for quantifying tumor neural infiltration based on digital pathological panoramic slides according to claim 7, characterized in that, The method for quantifying tumor nerve infiltration based on digital pathological panoramic slices also includes: setting overall scoring indicators and local scoring indicators, and assessing the patient's prognosis based on the indicators; For each patient, nerves with a non-zero epithelial wrapping index were selected as effective wrapped nerve units, and the number of effective wrapped nerve units was calculated. According to the total wrapping ratio of the effective wrapped neural units The average proportion of the patient's overall nerves that were wrapped was calculated. ; For each patient, the neural region with the smallest distance between the nerve tissue and the epithelial tissue was selected as a candidate target. Epithelial wrapping indices of a single nerve were extracted from these candidate targets. The candidates were then sorted in descending order according to these epithelial wrapping indices. The top N targets were selected, and the average epithelial wrapping ratio of these top N targets was calculated as the local nerve wrapping characteristic for that patient. When the number of objects within a candidate target is less than N, the average value of the epithelial wrapping ratio of all objects within the candidate target is calculated. Based on the average proportion of nerves encased in the patient's overall body Patients were divided into two groups according to the severity of their illness, and each group was assigned a score based on an overall scoring index. ;when At that time, assign an overall scoring index ;when At that time, assign an overall scoring index ; Based on the patient's local nerve encapsulation characteristics Patients were divided into two groups according to the severity of their illness, and each group was assigned a local scoring index. ;when At that time, assign local scoring indicators ;when At that time, assign local scoring indicators ; Score the overall rating indicators Compared with local scoring indicators The scores were summed to obtain the degree of nerve infiltration in the patients. ; Patients were stratified for risk based on their nerve infiltration scores, and survival data were combined with this data to assess the survival differences between different risk stratification groups, in order to determine whether the quantification of nerve infiltration had a significant ability to differentiate patient prognosis.

9. A device for quantifying tumor neural infiltration based on digital pathological panoramic slides, characterized in that, include: The semantic segmentation module is used to: acquire panoramic digital pathology slice images of tumors and perform multi-tissue category semantic segmentation to obtain whole-slice semantic segmentation results; and store the semantic segmentation results as a mask file. The neural infiltration quantization module is used to: based on the mapping relationship between the pixel coordinates of the mask image in the mask file and the pixel coordinates of the digital pathological panoramic slice image, crop out the pixel image corresponding to the candidate neural region in the mask file from the digital pathological panoramic slice image; Feature extraction is performed on the candidate neural regions and their corresponding pixel images in the mask file, and a machine learning model is used for classification to obtain reliable segmentation samples; The segmented reliable samples are mapped to a mask file and a digital pathological panoramic slice image through a mapping relationship. The epithelial tissue mask is extracted from the mask file, and the epithelial wrapping index of a single nerve is calculated in combination with the nerve region mask. The epithelial wrapping index is the ratio of the number of radial rays that effectively contact the epithelial tissue out of multiple radial rays emitted from the nerve centroid to the periphery of a single nerve to the total number of rays. It is used to quantify the degree of wrapping of the single nerve structure by the tumor epithelial component in local space. Based on the epithelial wrapping index of the single nerve, the degree of epithelial wrapping of the single nerve is classified according to grade, and the infiltration grade of the single nerve is obtained as the quantitative result of nerve infiltration.

10. A computer-readable storage medium having a computer program / instructions stored thereon, characterized in that, When the computer program / instruction is executed by the processor, it implements the steps of the method for quantifying tumor neural infiltration based on digital pathological panoramic slices as described in any one of claims 1 to 8.

Images