CN-120913863-B - Space-histology-based intestinal cancer metastasis prediction method, device, medium and equipment

CN120913863BCN 120913863 BCN120913863 BCN 120913863BCN-120913863-B

Abstract

The invention discloses a space histology-based intestinal cancer metastasis prediction method, a device, a medium and equipment, wherein the method comprises the steps of collecting original multicell data, performing modal alignment and quality control treatment to obtain preprocessed multicell data comprising second space transcriptome data, second single cell RNA sequencing data and second pathology image data; the method comprises the steps of performing cross-modal semantic embedding on second space transcriptome data based on second single-cell RNA sequencing data to generate a space enhancement expression spectrum, performing multi-scale image construction on the second space transcriptome data and second pathological image data to extract space heterogeneity characteristics, inputting the space enhancement expression spectrum and the space heterogeneity characteristics into a pre-trained transfer risk prediction model, outputting a liver transfer probability space heat map and a key driving characteristic list, and finally generating a clinical prediction report containing high-risk region positioning. According to the invention, the sensitivity of early transfer detection is remarkably improved by dynamically optimizing the collaborative modeling of the spatial resolution and the multi-scale features.

Inventors

HUANG BIN
WU HAN
AN HONGLIN
CHEN LIMING
LIU RUNPING
LI XIAOJIAOYANG
Qiu Yiman
CHEN WUJIN
LIN JIUMAO
LI MENGYUAN

Assignees

福建中医药大学

Dates

Publication Date: 20260508
Application Date: 20251013

Claims (5)

1. A space-histology-based bowel cancer metastasis prediction device, the device comprising: the data acquisition module is used for acquiring original multi-group data of an original intestinal cancer tissue sample, wherein the original multi-group data comprises first space transcriptome data, first single cell RNA sequencing data and first pathological image data; the data processing module is used for carrying out modal alignment and quality control processing on the original multi-group data to obtain preprocessed multi-group data, wherein the preprocessed multi-group data comprises second space transcriptome data, second single-cell RNA sequencing data and second pathological image data; The logic operation module is used for performing cross-modal semantic embedding on the second space transcriptome data according to the second single-cell RNA sequencing data to generate a space enhancement expression spectrum, the cross-modal semantic embedding comprises single-cell-space feature alignment and probabilistic sparse signal complementation based on contrast learning, and the space enhancement expression spectrum comprises a cell type deconvolution result and a confidence score of a transfer related gene; performing multi-scale map construction on the second space transcriptome data and the second pathological image data, extracting space heterogeneity characteristics, wherein the multi-scale map construction comprises hierarchical modeling of microscopic cell interaction maps, mesoscopic function region maps and macroscopic tissue structures, and the space heterogeneity characteristics comprise tumor-immune adjacent strength, vascular invasion front space gradient and high-risk subregion topological relation; The report generation module is used for generating a clinical prediction report of the intestinal cancer transfer risk according to the space heat map and the key driving characteristic list, wherein the report comprises high-risk area positioning, transfer probability scoring and treatment response prediction; Performing cross-modal semantic embedding of the second spatial transcriptome data based on the second single-cell RNA sequencing data to generate a spatially enhanced expression profile comprising: extracting a cell type signature matrix from the second single cell RNA sequencing data, the cell type signature matrix comprising a marker gene expression profile for each cell type; performing contrast learning alignment on each spatial site of the second spatial transcriptome data and the cell type feature matrix, and calculating similarity scores of the spatial sites and the cell types; Performing probabilistic cell type deconvolution on the space locus based on the similarity score to generate a cell type proportion matrix, wherein the deconvolution adopts non-negative matrix factorization and Markov random field joint optimization; performing sparse signal complementation on the low-expression genes of the second space transcriptome data, wherein the complement signals are derived from the prior expression distribution of the cells of the same type in single-cell data; fusing the cell type proportion matrix with the complemented gene expression data to generate a space enhanced expression profile, wherein the space enhanced expression profile comprises cell composition information and gene expression confidence evaluation of each space site; performing multi-scale map construction on the second spatial transcriptome data and the second pathology image data, extracting spatial heterogeneity features, including: constructing a microscopic cell interaction network map based on the single cell resolution expression profile of the second spatial transcriptome data, the microscopic cell interaction network map generated by ligand-receptor pair co-expression analysis and cell adjacency calculation; Performing tissue region segmentation on the second pathological image data, identifying a tumor core region, an invasion front region and a interstitial region, and constructing a mesoscopic function region map by combining second space transcriptome data expression characteristics of the corresponding region; Integrating the tissue structure characteristics of the second pathological image data at the full slice level with the global expression mode of the space transcriptome data, and establishing a macroscopic tissue transfer trend prediction graph; Extracting tumor-immune cell interaction intensity characteristics from the microscopic cell interaction network map, including spatial co-localization frequency of immune checkpoint molecule pairs; calculating the spatial gradient characteristics of the vascular invasion front from the mesoscopic functional region map, wherein the spatial gradient characteristics comprise attenuation coefficients of endothelial marker expression changing along with the distance; extracting topological features of high-risk subregions from the macroscopic tissue transfer trend prediction graph, wherein the topological features comprise spatial autocorrelation indexes of the microenvironment before transfer; and fusing the tumor-immune cell interaction intensity characteristic, the vascular invasion front space gradient characteristic and the high-risk subregion topological characteristic into a space heterogeneity characteristic.
2. The spatial histology-based bowel cancer metastasis prediction device of claim 1, wherein performing modal alignment and quality control processing on the raw multiple-study data to obtain preprocessed multiple-study data comprises: Screening out space loci meeting a preset quality threshold from the first space transcriptome data as second space transcriptome data, wherein the preset quality threshold comprises a gene detection number threshold and a mitochondrial gene duty ratio upper limit; Extracting cell sequencing data matched with an anatomical region of the second space transcriptome data in the first single-cell RNA sequencing data as second single-cell RNA sequencing data, wherein the anatomical region matching is realized through tissue slice coordinate mapping; Selecting an image area corresponding to the acquisition position of the second space transcriptome data from the first pathology image data as second pathology image data, wherein the spatial deviation of the acquisition position in matching does not exceed a preset alignment tolerance; establishing a spatial association relationship among the second spatial transcriptome data, the second single-cell RNA sequencing data and the second pathological image data, so that the three have a uniform coordinate reference system; performing batch effect correction on the second spatial transcriptome data to eliminate technical variation among different samples; Performing low quality cell filtration on the second single cell RNA sequencing data, retaining transcriptome data meeting cellular integrity criteria; generating preprocessed multi-group chemical data, and storing a plurality of the preprocessed multi-group chemical data into a quality control database.
3. The space-histology-based intestinal cancer metastasis prediction device of claim 1, wherein the probability cell type deconvolution of the spatial loci based on the similarity score generates a cell type proportion matrix, the deconvolution employing a non-negative matrix factorization in combination with markov random fields, comprising: establishing a neighborhood relation diagram of a Markov random field based on the space coordinate information of the second space transcriptome data by taking the similarity score as a non-negative matrix factorization objective function of the initial weight construction space locus-cell type, wherein the neighborhood relation diagram comprises topological connection relations among the space loci; Introducing a space smoothness constraint term into a non-negative matrix factorization objective function, wherein the space smoothness constraint term regulates the cell type proportion difference between adjacent sites through a Markov random field; iteratively updating the cell type proportion matrix by adopting an alternative optimization algorithm, wherein each iteration comprises the following steps: a fixed space constraint term, optimizing the cell type proportion of non-negative matrix factorization through multiplication updating rules, and optimizing the space consistency through graph Laplace regularization; terminating the optimization when the Frobenius norm variation of the cell type proportion matrix of the adjacent iteration is smaller than a preset convergence threshold; normalizing the optimized cell type proportion matrix to ensure that the sum of the cell type proportions of each space locus is 1; outputting the cell type proportion matrix, wherein the rows of the cell type proportion matrix correspond to space sites, the columns of the cell type proportion matrix correspond to cell types, and the element values are probability proportion weights.
4. The space histology-based intestinal cancer metastasis prediction device according to claim 1, wherein integrating the tissue structural features of the second pathology image data at the full slice level with the global expression pattern of the space transcriptome data creates a macroscopic tissue metastasis tendency prediction map, comprising: performing full-slice scanning on the second pathological image data, and extracting tissue structure characteristics, wherein the tissue structure characteristics comprise tumor gland arrangement directivity, interstitial fibrosis distribution density and necrosis area space ratio; performing spatial autocorrelation analysis on the global expression pattern of the second spatial transcriptome data, and calculating spatial clustering indexes of each gene expression hot spot region, wherein the global expression pattern comprises the spatial distribution of expression amounts of an epithelial-mesenchymal transition marker, an angiogenesis factor and an immunosuppression related molecule; Constructing a multi-mode fusion model based on a graph convolution neural network, carrying out feature alignment on the tissue structure features and a global expression mode, realizing weight distribution of pathological image features and transcriptome expression features through a cross-mode attention mechanism in the alignment process, and outputting a fused multi-mode feature matrix; establishing a macroscopic tissue transfer trend prediction graph in the multi-mode feature matrix, wherein the macroscopic tissue transfer trend prediction graph simulates potential transfer paths of tumor cells along tissue structural features and expression gradient features through a random walk algorithm; performing topological optimization on the macroscopic tissue transfer trend prediction graph, and identifying a high-risk transfer sub-region, wherein the topological optimization comprises the steps of calculating the space density and transfer direction consistency coefficient of transfer path intersection points; And outputting a final macroscopic tissue transfer trend prediction graph, wherein the macroscopic tissue transfer trend prediction graph comprises a transfer probability thermodynamic diagram, a high-risk subregion boundary label and a main transfer path vector field.
5. The spatial histology-based intestinal cancer metastasis prediction device according to claim 1, wherein inputting the spatial enhancement expression profile and spatial heterogeneity feature into a pre-trained metastasis risk prediction model, outputting a spatial heat map comprising liver metastasis probabilities and a list of key driving features, comprises: performing characteristic standardization treatment on the space enhancement expression profile to enable the expression quantity of different genes to be comparable, wherein the characteristic standardization treatment comprises Z-score normalization and batch effect correction; Performing dimension compression on the space heterogeneity characteristics, and reserving principal component characteristics with contribution rate exceeding a preset threshold by adopting a principal component analysis method; characteristic splicing is carried out on the standardized spatial enhancement expression spectrum and the compressed spatial heterogeneity characteristic, and a fusion characteristic matrix is generated; inputting the fusion feature matrix into a pre-trained graphic neural network model, wherein the graphic neural network model aggregates multi-scale space information based on an attention mechanism; calculating the liver transition probability of each spatial site through the map neural network model, and generating a spatial heat map based on probability values, wherein the spatial heat map and an original tissue slice keep a spatial corresponding relation; Extracting key driving features from the graph neural network model by adopting a gradient back propagation method, wherein the key driving features comprise gene expression features and spatial topological features with highest contribution to a prediction result; Sorting the key driving features according to the feature contribution degree to generate a key driving feature list containing feature names, contribution degree scores and biological explanation; And storing the space heat map and the key driving characteristic list in a correlated way, and establishing a mapping relation with clinical pathological parameters.

Description

Space-histology-based intestinal cancer metastasis prediction method, device, medium and equipment Technical Field The invention relates to the technical field of medical treatment, in particular to a space histology-based intestinal cancer metastasis prediction method, a device, a medium and equipment. Background Early prediction of intestinal cancer metastasis is critical for clinical treatment, traditional pathological examination relies on morphological observations, and although part of the high risk features can be identified, it is difficult to quantify the spatial heterogeneity of tumor microenvironment. Molecular detection techniques such as gene expression profiling can provide tumor molecular characteristics, but lose critical spatial information and cannot reflect regional differences inside tumors. Single cell sequencing improves resolution but still fails to preserve spatial distribution information of cells in situ tissue. In recent years, spatial histology offers the possibility of simultaneously acquiring gene expression and its spatial distribution, but existing spatial transcriptomes have limited detection resolution, and single data points may contain mixed cell signals, affecting identification of micrometastases. In addition, how to effectively integrate multi-scale data such as molecular expression, cell interaction, tissue structure and the like and establish a high-precision prediction model is still a technical problem to be solved. The existing prediction method depends on single data or simple feature superposition, and the complementarity of multi-source data cannot be fully utilized, so that the prediction accuracy of early stage transfer risk is insufficient. Disclosure of Invention In view of the above problems, the invention provides a space-histology-based intestinal cancer metastasis prediction method, a space-histology-based intestinal cancer metastasis prediction device, a space-resolution-based intestinal cancer metastasis prediction medium and space-resolution-based intestinal cancer metastasis prediction equipment, which realize high-sensitivity metastasis risk detection through dynamic optimization of space resolution and collaborative modeling of multi-scale features, and solve the problem of high early metastasis missed judgment rate. To achieve the above object, in a first aspect, the present application provides a method for predicting metastasis of intestinal cancer based on space histology, comprising: Collecting original multiple sets of chemical data of an original intestinal cancer tissue sample, wherein the original multiple sets of chemical data comprise first space transcriptome data, first single cell RNA sequencing data and first pathological image data; Performing modal alignment and quality control processing on the original multi-group data to obtain preprocessed multi-group data, wherein the preprocessed multi-group data comprises second space transcriptome data, second single-cell RNA sequencing data and second pathological image data; Performing cross-modal semantic embedding on the second spatial transcriptome data based on the second single-cell RNA sequencing data to generate a spatial enhancement expression profile, wherein the cross-modal semantic embedding comprises single-cell-spatial feature alignment and probabilistic sparse signal complementation based on contrast learning, and the spatial enhancement expression profile comprises a cell type deconvolution result and a confidence score of a transfer related gene; Carrying out multi-scale map construction on the second space transcriptome data and the second pathological image data, extracting space heterogeneity characteristics, wherein the multi-scale map construction comprises hierarchical modeling of microscopic cell interaction maps, mesoscopic function region maps and macroscopic tissue structures, and the space heterogeneity characteristics comprise tumor-immune adjacent intensity, vascular invasion front space gradient and high-risk subregion topological relation; inputting the spatial enhancement expression profile and the spatial heterogeneity characteristic into a pre-trained transfer risk prediction model, and outputting a spatial heat map containing liver transfer probability and a key driving characteristic list; Based on the space heat map and the key driving characteristic list, a clinical prediction report of the intestinal cancer transfer risk is generated, wherein the report comprises high-risk area positioning, transfer probability scoring and treatment response prediction. Further, performing modal alignment and quality control processing on the original multi-group data to obtain preprocessed multi-group data, including: screening out space loci meeting a preset quality threshold from the first space transcriptome data as second space transcriptome data, wherein the preset quality threshold comprises a gene detection number threshold and a mitochondrial gene duty rati