CN-121998985-A - Tumor nerve infiltration quantification method, device and medium based on digital pathological panoramic section

CN121998985ACN 121998985 ACN121998985 ACN 121998985ACN-121998985-A

Abstract

The invention belongs to the field of medical image analysis and computational pathology, and discloses a tumor nerve infiltration quantification method, device and medium based on digital pathology panoramic sections, which comprises the steps of obtaining the digital pathology panoramic sections of tumors, carrying out multi-tissue category semantic segmentation and storing panoramic segmentation masks; extracting candidate nerve region masks and pathological pixel images through connected regions of the masks according to the coordinate mapping relation between the masks and the panoramic slice, extracting features of the candidate nerve region masks and the pixel images thereof, establishing and applying a classification model of the reliability of the candidate nerve region, filtering to obtain a reliable nerve region sample, calculating epithelium wrapping indexes of single nerves according to epithelium tissue masks and nerve region masks corresponding to the reliable nerve region sample and forming quantization scores according to grades, and summarizing all nerve regions in the panoramic slice to form quantitative indexes of peripheral nerve infiltration around tumors. The invention can realize the automatic evaluation of the tumor peripheral nerve infiltration degree of the digital pathological panoramic section.

Inventors

YU HAIRONG
JIAO YIPING
ZHANG XINYUAN

Assignees

南京信息工程大学

Dates

Publication Date: 20260508
Application Date: 20260409

Claims (10)

1. The tumor nerve infiltration quantification method based on the digital pathological panoramic section is characterized by comprising the following steps of: Obtaining a tumor digital pathological panoramic slice image and carrying out multi-tissue category semantic segmentation to obtain a full-slice semantic segmentation result; Cutting out a pixel image corresponding to a candidate nerve region in the mask file from the digital pathological panoramic slice image according to the mapping relation between the pixel coordinates of the mask image of the mask file and the pixel coordinates of the digital pathological panoramic slice image; Extracting features of the candidate neural areas in the mask file and pixel images corresponding to the candidate neural areas, and classifying by using a machine learning model to obtain a segmentation reliable sample; Mapping the segmented reliable samples into a mask file and a digital pathological panoramic slice image through a mapping relation, extracting an epithelial tissue mask from the mask file, and calculating an epithelial wrapping index of a single nerve by combining the nerve region mask; And according to the epithelial wrapping index of the single nerve, grading the epithelial wrapping degree of the single nerve, and obtaining the infiltration grade of the single nerve as a nerve infiltration quantification result.
2. The method for quantifying tumor neuroinfiltration based on digital pathology panorama slicing according to claim 1, wherein the multi-tissue class semantic segmentation is implemented based on an existing multi-tissue class semantic segmentation model comprising a neural class and an epithelial tissue class; The full-slice semantic segmentation result comprises a neural region segmentation result and an epithelial tissue segmentation result.
3. The method for quantifying tumor nerve infiltration based on digital pathological panoramic slicing according to claim 1, wherein the process of clipping the pixel image corresponding to the candidate nerve region in the mask file from the digital pathological panoramic slicing image comprises: constructing a mapping relation between mask image pixel coordinates of a mask file and digital pathological panoramic slice image pixel coordinates: Coordinates of pixels of the mask image Mapping to digital pathological panoramic slice image pixel coordinates The mapping relationship is expressed as: ; ; in the formula, And Spatial resolution parameters of the digital pathological panoramic slice image and the mask image are respectively calculated; Obtaining a mask of a candidate neural region in a mask file, extracting the same-class pixel set of each candidate neural region by adopting an eight-connected-domain method, and determining the complete range of the corresponding candidate neural region; Taking the central pixel of each candidate neural area as the center, extracting a corresponding candidate neural area mask, and cutting the candidate neural area into a binary image of a fixed-size pixel for storage, wherein the position with the gray value equal to 255 in the binary image is the neural area; And cutting out pixel images corresponding to the candidate nerve regions from the digital pathological panoramic slice images according to the mapping relation.
4. The method for quantifying tumor nerve infiltration based on digital pathology panorama according to claim 1, wherein extracting features of the candidate nerve region and the pixel image corresponding to the candidate nerve region in the mask file comprises extracting morphological features of the candidate nerve region and texture features of the pixel image corresponding to the candidate nerve region; The morphological characteristics comprise the area, roundness, eccentricity, axial length ratio and boundary Fourier frequency components of the nerve region, and are used for representing the geometrical structure and contour characteristics of the nerve region; The texture features comprise gray level non-uniformity, short run emphasis, long run emphasis, joint entropy and differential entropy, and are used for representing the internal color distribution characteristics of the nerve region and the dependency relationship among pixels.
5. The method for quantifying tumor nerve infiltration based on digital pathology panorama slicing according to claim 4, wherein the machine learning model employs a support vector machine model; The classification process is as follows: manually checking the pixel image to obtain a reliable segmentation sample and an unreliable segmentation sample which are manually checked; In the training stage of the support vector machine model, based on the morphological characteristics, the texture characteristics and the manual inspection result, the mapping relation between the neural area characteristics and the class labels is learned; and in the support vector machine model prediction stage, judging whether the neural area belongs to a segmentation reliable sample or not by using the trained support vector machine model.
6. The method for quantifying tumor nerve infiltration based on digital pathology panorama according to claim 1, wherein the calculating the epithelial wrapping index of the single nerve comprises: mapping the segmented reliable samples into a mask file and a digital pathological panoramic slice image through a mapping relation to obtain a mask image, wherein the mask image comprises a nerve region mask image and an epithelial tissue mask image; Constructing an indication function in the mask image When the gray value is 255, Otherwise ; Pixel coordinates based on the indication function Weighted average is carried out to obtain the barycenter coordinates of the mask image area : ; ; Taking the centroid coordinates of each neural area mask image as the center, and according to a preset angle within the range of 0-360 DEG Constructing radial rays, angular step ; Minimal circumscribed rectangular extension to each nerve region As a search range, pixel coordinates on a ray are expressed as: ; in the formula, Is the length of the current ray; recording the pixel point of each ray just penetrating out of the nerve region as the outermost boundary point of the nerve region , As boundary points The axis of the rotation is set to be at the same position, As boundary points An axis coordinate; taking the outermost boundary point as a starting position, continuing searching along the direction of the corresponding ray, detecting whether the pixel points belonging to the epithelial tissue mask area exist on the ray, and when the epithelial tissue pixel points are detected Recording the ray as an effective contact ray; As boundary points The axis of the rotation is set to be at the same position, As boundary points An axis coordinate; Based on the number of active contact rays Calculating to obtain the epithelial wrapping index of single nerve , 。
7. The method for quantifying tumor nerve infiltration based on digital pathology panorama according to claim 6, wherein the dividing the epithelial wrapping degree of the individual nerve comprises: by calculating boundary points on each nerve region And epithelial tissue pixel points Minimum Euclidean distance between Obtaining a minimum distance measure between the neural tissue and the epithelial tissue; introducing a pixel distance threshold, when the minimum Euclidean distance is When the pixel distance is smaller than the pixel distance threshold value, the nerve is a tumor epithelial paranerve, and the minimum Euclidean distance is obtained Multiplying the spatial resolution of the mask file to obtain the true distance ; When the true distance is In this case, according to the epithelial wrapping index of the individual nerve The epithelial wrapping degree of a single nerve is precisely divided into five grades: Class 1: Judging that no epithelial tissue is wrapped; Class 2: Judging that the epithelial tissue is slightly wrapped; Grade 3: Judging that the medium epithelial tissue is wrapped; grade 4: judging that the epithelial tissue is severely wrapped; grade 5: the extremely severe epithelial tissue wrap was determined.
8. The method for quantifying tumor nerve infiltration based on digital pathology panorama according to claim 7, further comprising setting an overall score index and a local score index, and evaluating a prognosis of the patient according to the indexes; For each patient, a nerve with an epithelial wrapping index of not 0 of the individual nerve was selected as an effective wrapped nerve unit, and the number of effective wrapped nerve units was calculated According to the total wrapping proportion of the effective wrapping nerve units Calculating to obtain the average wrapping proportion of the whole nerve of the patient ; For each patient, selecting a nerve region with the smallest distance between nerve tissue and epithelial tissue as a candidate target, extracting the epithelial wrapping index of a single nerve from the candidate target, arranging the objects in the candidate target in a descending order according to the epithelial wrapping index, selecting the objects in the top N positions in order, and calculating the average value of the epithelial wrapping ratio of the objects in the top N positions as the local nerve wrapping characteristic of the patient When the number of objects in the candidate target is less than N, averaging the epithelial wrapping proportion of all objects in the candidate target; Based on the average wrapping ratio of the patient's global nerves Patients are divided into two groups according to severity, and the patients are respectively endowed with integral grading index scores When (1) When giving an overall score index When (1) When giving an overall score index ; Based on local nerve wrapping characteristics of patients Patients are divided into two groups according to severity, and local scoring indexes are respectively assigned to the patients When (1) In this case, a local score index is given When (1) In this case, a local score index is given ; Scoring the overall score index Score with local score indicator Adding to obtain the nerve infiltration degree score of the patient ; And carrying out risk stratification on the patient according to the nerve infiltration degree score of the patient, and evaluating survival differences among different risk stratification groups by combining with the life cycle data of the patient so as to judge whether the distinguishing capability of the nerve infiltration quantitative result on the prognosis of the patient is obvious or not.
9. Tumor nerve infiltration quantization device based on digit pathology panorama section, characterized by comprising: The semantic segmentation module is used for acquiring a tumor digital pathological panoramic slice image and carrying out multi-tissue type semantic segmentation to obtain a full-slice semantic segmentation result; The nerve infiltration quantization module is used for cutting out a pixel image corresponding to a candidate nerve region in the mask file from the digital pathological panoramic slice image according to the mapping relation between the pixel coordinates of the mask image of the mask file and the pixel coordinates of the digital pathological panoramic slice image; Extracting features of the candidate neural areas in the mask file and pixel images corresponding to the candidate neural areas, and classifying by using a machine learning model to obtain a segmentation reliable sample; Mapping the segmented reliable samples into a mask file and a digital pathological panoramic slice image through a mapping relation, extracting an epithelial tissue mask from the mask file, and calculating an epithelial wrapping index of a single nerve by combining the nerve region mask; And according to the epithelial wrapping index of the single nerve, grading the epithelial wrapping degree of the single nerve, and obtaining the infiltration grade of the single nerve as a nerve infiltration quantification result.
10. A computer readable storage medium having stored thereon a computer program/instructions, which when executed by a processor, performs the steps of the digital pathological panoramic slice-based tumor nerve infiltration quantification method of any one of claims 1 to 8.

Description

Tumor nerve infiltration quantification method, device and medium based on digital pathological panoramic section Technical Field The invention relates to a tumor nerve infiltration quantification method, device and medium based on a digital pathology panoramic slice, and belongs to the technical field of medical image analysis and pathology calculation. Background Pancreatic ductal adenocarcinoma is one of the most malignant tumors of the digestive system, often accompanied by higher morbidity and mortality. Neuro-infiltration (PNI) is a pathological feature of pancreatic ductal adenocarcinoma, with incidence rates of up to 70% or more. The existing research shows that the neuro infiltration phenomenon in tumor tissues is closely related to the invasiveness of tumors, biological behaviors and prognosis of patients. However, in the pathological analysis of pancreatic ductal adenocarcinoma, the relevant diagnosis usually requires a pathologist to observe pathological sections for a long time and to complete interpretation of tumor types and grades based on expertise and experience. The method is limited by factors such as strong subjectivity, relatively poor repeatability and the like of a manual identification mode, objective and repeatable quantitative evaluation of the spatial relationship between the nerve structure and the tumor tissue is difficult, most of the existing pathological reports only give qualitative conclusions about whether nerve infiltration exists or not, and quantitative description of the nerve infiltration degree is lacking. With the development of digital pathology technology, a semantic segmentation model based on a computer vision technology is gradually applied to the automatic analysis of tumor sections, and a foundation is laid for the quantitative evaluation of tumor microenvironments. In the patent number CN119992552B, a seamless semantic segmentation method on a digital pathological section is disclosed, and the method can be used for automatically segmenting key components of tumor microenvironments such as epithelium, nerve bundles and the like, but a PNI quantitative analysis method is not established yet. More importantly, the nerve bundles and part of the tumor stroma components are very similar in appearance, resulting in a large number of false positives in the segmentation results, failing to form an accurate PNI score. Disclosure of Invention The invention aims to provide a tumor nerve infiltration quantification method, device and medium based on a digital pathological panoramic section, which construct quantification indexes through spatial morphological characteristics of nerve tissues and epithelial tissues in the digital pathological section, automatically identify the affected degree of a nerve region in the pancreatic cancer pathological section, provide standardized data support for pathological diagnosis, clinical curative effect prediction and scientific research analysis, and improve the interpretation and reliability of model prediction. In order to achieve the above purpose, the invention is realized by adopting the following technical scheme. In one aspect, the invention provides a tumor nerve infiltration quantification method based on digital pathological panoramic sections, comprising the following steps: Obtaining a tumor digital pathological panoramic slice image and carrying out multi-tissue category semantic segmentation to obtain a full-slice semantic segmentation result; Cutting out a pixel image corresponding to a candidate nerve region in the mask file from the digital pathological panoramic slice image according to the mapping relation between the pixel coordinates of the mask image of the mask file and the pixel coordinates of the digital pathological panoramic slice image; Extracting features of the candidate neural areas in the mask file and pixel images corresponding to the candidate neural areas, and classifying by using a machine learning model to obtain a segmentation reliable sample; Mapping the segmented reliable samples into a mask file and a digital pathological panoramic slice image through a mapping relation, extracting an epithelial tissue mask from the mask file, and calculating an epithelial wrapping index of a single nerve by combining the nerve region mask; And according to the epithelial wrapping index of the single nerve, grading the epithelial wrapping degree of the single nerve, and obtaining the infiltration grade of the single nerve as a nerve infiltration quantification result. Optionally, the multi-tissue class semantic segmentation is implemented based on an existing multi-tissue class semantic segmentation model including a neural class and an epithelial tissue class; The full-slice semantic segmentation result comprises a neural region segmentation result and an epithelial tissue segmentation result. Optionally, the process of clipping the pixel image corresponding to the candidate neural area in the mask file fr