CN-122023371-A - Immunohistochemical staining image quantitative analysis method and system thereof

CN122023371ACN 122023371 ACN122023371 ACN 122023371ACN-122023371-A

Abstract

The invention relates to the technical field of quantitative analysis of immunohistochemical staining images, in particular to a quantitative analysis method and a quantitative analysis system of the immunohistochemical staining images, wherein the quantitative analysis method comprises the following steps of S1, obtaining digital images of the immunohistochemical staining sections; and S2, performing color deconvolution processing on the digital image, and decomposing the digital image into at least one exclusive channel image and one background channel image corresponding to the target coloring agent. According to the invention, color deconvolution is realized through a non-negative matrix factorization algorithm, a target stain signal in a composite staining image, a background and other staining signals are accurately split, the problem that the existing system staining components are not thoroughly separated is solved, positive and negative cells in different staining modes can be accurately identified through a Mask R-CNN deep learning network and combining with multi-mode adaptation parameters, and a plurality of scenes such as nuclear staining, membrane staining, slurry staining and the like are adapted.

Inventors

CHEN YAN
ZHANG YUAN
TANG XIULIANG
Maoxuelian
YU SHAORONG
ZHU DI
XU XINYU
TAO LI

Assignees

江苏省肿瘤医院

Dates

Publication Date: 20260512
Application Date: 20260210

Claims (10)

1. The immunohistochemical staining image quantitative analysis method is characterized by comprising the following steps of: s1, obtaining a digital image of an immunohistochemical staining section; S2, performing color deconvolution on the digital image, and decomposing the digital image into at least one exclusive channel image and one background channel image corresponding to the target coloring agent; S3, based on the exclusive channel image, combining with preset adaptive parameters aiming at different dyeing modes, automatically identifying positive dyeing cells and negative dyeing cells in the image; s4, extracting the dyeing optical density value of each identified cell in the exclusive channel image, and classifying the cells into negative, weak positive, medium positive or strong positive grades according to a preset unified dyeing intensity classification standard; And S5, calculating the percentage of positive cells and the percentage of cells of each intensity level according to the dividing result of the cells of each level, and generating an analysis result comprising the dyeing intensity pseudo color mark, the positive rate statistical data and the dyeing space distribution information.
2. The quantitative analysis method of immunohistochemical staining images according to claim 1 wherein the color deconvolution processing in S2 adopts a non-negative matrix factorization algorithm to matrix the original image data Decomposition into a dye spectral base matrix And concentration coefficient matrix The mathematical model is expressed as: ; Wherein, the Representing the input RGB image data matrix, For the total number of pixels, The number of color channels; The representation comprises A spectral basis vector matrix of seed stains and background; representing the relative concentration matrix of each pixel under each stain channel; The decomposition process needs to meet And is also provided with Is not a negative constraint condition of (2).
3. The method for quantitative analysis of immunohistochemical staining images according to claim 2, wherein the implementation of S2 comprises the following specific steps: s21, initializing a spectral base matrix of the coloring agent Its initial value is based on standard spectral data of known colorants; S22, adopting multiplication updating rule to iterate and optimize matrix And Until a preset convergence condition is met; S23, from the concentration coefficient matrix Extracting a row corresponding to the target coloring agent, and reconstructing to generate the exclusive channel image.
4. The immunohistochemical staining image quantitative analysis method according to claim 1, wherein the different staining patterns in S3 include nuclear staining, membrane staining and slurry staining; the preset adaptation parameters comprise cell morphological characteristic thresholds, staining area positioning rules and segmentation mask generation strategies aiming at different staining modes.
5. The quantitative analysis method of immunohistochemical staining images according to claim 4 wherein the implementation of S3 comprises the following specific steps: s31, receiving the exclusive channel image and the background channel image; s32, according to the dyeing mode corresponding to the current image to be analyzed, corresponding adaptation parameters are called from a preset parameter library; S33, inputting the exclusive channel image, the background channel image and the adaptation parameters into a deep learning example segmentation network; s34, outputting an example segmentation mask of each cell by the deep learning example segmentation network, and judging whether the cell is a positive staining cell or a negative staining cell according to the signal intensity of the pixel in the mask in the exclusive channel.
6. The quantitative analysis method of immunohistochemical staining images according to claim 5, wherein the deep learning example segmentation network is Mask R-CNN network, the loss function of which By loss of classification Regression loss of bounding box And mask segmentation loss The weighting composition is expressed as: ; Wherein, the And The weight coefficient which can be adjusted through model training or manual configuration is used for balancing the contribution degree of each loss in the training process.
7. The method for quantitative analysis of immunohistochemical staining images according to claim 1, wherein the staining optical density value in S4 Calculated by the following formula: ; Wherein, the As the average pixel intensity value of the cell region in the dedicated channel image, The average pixel intensity value of a corresponding region in the background channel image or a preset maximum transmission light intensity threshold value.
8. The method for quantitative analysis of immunohistochemical staining images according to claim 7, wherein the preset unified staining intensity classification standard is defined as: When (when) When the result is negative, judging that the result is negative; When (when) When the test piece is judged to be weak positive; When (when) When the test piece is judged to be moderately positive; When (when) When the test piece is judged to be strong positive; Wherein, the 、、 Is a preset optical density threshold value and meets 。
9. The quantitative analysis method of immunohistochemical staining images according to claim 1 wherein the analysis result generated in S5 further comprises at least one commonly used immunohistochemical score index of H-score, allred score or Quick-score, the calculation of the score index being based on the ratio of cells of each intensity class and the corresponding assigned weight.
10. An immunohistochemical staining image quantitative analysis system for realizing the method for the quantitative analysis of an immunohistochemical staining image according to any one of claims 1 to 9, comprising: the image acquisition module is used for acquiring digital images of the immunohistochemical staining slices; the input end of the color deconvolution processing module is connected with the output end of the image acquisition module and is used for decomposing the digital image into at least one exclusive channel image and one background channel image corresponding to the target coloring agent; the multi-mode adaptation module is used for storing and providing preset adaptation parameters for different dyeing modes; The input end of the cell nucleus detection module is respectively connected with the output ends of the color deconvolution processing module and the multi-mode adaptation module and is used for automatically identifying positive staining cells and negative staining cells in the image; The input end of the color quantifying module is connected with the output end of the cell nucleus detecting module, and receives the adapting parameters from the multi-mode adapting module, and is used for classifying the dyeing intensity of each identified cell; And the input end of the result output module is connected with the output end of the color quantifying module and is used for generating and outputting comprehensive analysis results comprising positive cell percentages, cell percentages of all intensity levels, dyeing intensity pseudo-color labeling images, positive rate statistical data, dyeing space distribution analysis diagrams and common scoring indexes.

Description

Immunohistochemical staining image quantitative analysis method and system thereof Technical Field The invention relates to the technical field of quantitative analysis of immunohistochemical staining images, in particular to a quantitative analysis method and a quantitative analysis system of an immunohistochemical staining image. Background The immunohistochemical staining technology is a key means for evaluating the expression of tumor cell antigens in pathological diagnosis, a target antigen is developed through a specific staining agent, a doctor judges the illness state according to the number and the staining depth of the stained cells, the result directly influences the treatment scheme formulation and the prognosis evaluation, and the method is an important tie for connecting pathological morphology observation and clinical diagnosis. The Ki-67 immunohistochemical pathological image automatic quantitative analysis system proposed by the prior published patent CN101799926B realizes automatic image segmentation and parameter extraction through multi-module cooperation, and reduces manual intervention to a certain extent. However, the system does not design a special algorithm for component splitting of the composite dyeing image, does not establish a unified dyeing intensity grading standard, still depends on the traditional image splitting and feature extraction modes, and is difficult to adapt to analysis requirements of different dyeing modes. In the prior art, manual interpretation depends on the experience of doctors, and the difference exists between the judgment standards of different doctors on the staining intensity, so that the semi-quantitative scoring repeatability is poor, while the existing automatic analysis system lacks an accurate staining component separation technology, cannot thoroughly distinguish the chromogenic signals of different staining agents, does not form a standardized staining intensity grading system, cannot objectively quantify antigens with different expression levels, and finally causes the deviation between a quantitative result and the actual antigen expression state, thereby influencing the consistency and the accuracy of diagnosis. Disclosure of Invention Aiming at the defects of the prior art, the invention provides an immunohistochemical staining image quantitative analysis method and a system thereof, which solve the problems of poor manual interpretation repeatability, inaccurate separation of staining components and uneven grading standard in the prior art, which cause deviation of quantitative results. In order to achieve the aim, the invention is realized by the following technical scheme that the immunohistochemical staining image quantitative analysis method comprises the following steps: s1, obtaining a digital image of an immunohistochemical staining section; S2, performing color deconvolution on the digital image, and decomposing the digital image into at least one exclusive channel image and one background channel image corresponding to the target coloring agent; S3, based on the exclusive channel image, combining with preset adaptive parameters aiming at different dyeing modes, automatically identifying positive dyeing cells and negative dyeing cells in the image; s4, extracting the dyeing optical density value of each identified cell in the exclusive channel image, and classifying the cells into negative, weak positive, medium positive or strong positive grades according to a preset unified dyeing intensity classification standard; And S5, calculating the percentage of positive cells and the percentage of cells of each intensity level according to the dividing result of the cells of each level, and generating an analysis result comprising the dyeing intensity pseudo color mark, the positive rate statistical data and the dyeing space distribution information. Further, the color deconvolution in S2 uses a non-negative matrix factorization algorithm to matrix the original image dataDecomposition into a dye spectral base matrixAnd concentration coefficient matrixThe mathematical model is expressed as: ; Wherein, the Representing the input RGB image data matrix,For the total number of pixels,The number of color channels; The representation comprises A spectral basis vector matrix of seed stains and background; representing the relative concentration matrix of each pixel under each stain channel; The decomposition process needs to meet And is also provided withIs not a negative constraint condition of (2). Further, the implementation of S2 includes the following specific steps: s21, initializing a spectral base matrix of the coloring agent Its initial value is based on standard spectral data of known colorants; S22, adopting multiplication updating rule to iterate and optimize matrix AndUntil a preset convergence condition is met; S23, from the concentration coefficient matrix Extracting a row corresponding to the target coloring agent, and reconstruct