Multi-omics fusion prognostic assessment method and system for IDH wild-type glioblastoma

Abstract

The present application relates to the field of image analysis technology, and in particular to a multi-omics fusion IDH wild-type glioblastoma prognosis assessment method and system, the method comprising: obtaining a data set, obtaining pathological images, gene data and transcriptional data respectively; and extracting omics features from pathological images, gene data and transcriptional data respectively; based on the omics features, constructing a single omics prognosis model corresponding to the omics features respectively; screening the omics features of the single omics prognosis model to obtain multi-omics fusion features, and constructing a multi-omics fusion model based on the multi-omics fusion features, and then stratifying the patients for risk; performing differential analysis on high-risk groups and low-risk groups, obtaining cross-pathways, and determining the key biological pathways and biological behaviors that affect the prognosis of the multi-omics fusion model. The present application integrates pathological omics, proteomics, and genomic features to construct a multi-omics fusion model, which increases the prognostic prediction value of glioblastoma.

Description

Technical Field

[0001] The present application relates to the field of image analysis technology, and in particular to a multi-omics fusion IDH wild-type glioblastoma prognosis assessment method and system. Background Art

[0002] The incidence of primary malignant brain tumors is about 7 in 100,000, among which glioblastoma is the most common malignant brain tumor in adults, and its incidence increases after the age of 40 and peaks at 75 to 84 years old. The fifth edition of the Classification of Tumors of the Central Nervous System, released by the World Health Organization (WHO) in 2021, divides adult diffuse gliomas into three categories, among which glioblastoma is defined as isocitrate dehydrogenase (IDH) wild-type glioblastoma. Compared with other malignant brain tumors, IDH wild-type glioblastoma is more malignant and has a shorter life expectancy after diagnosis. Studies have shown that the 5-year survival rate after diagnosis of malignant brain tumors is about 36%, the median survival time of patients with IDH wild-type glioblastoma is 12 to 21 months, and the 5-year survival rate is about 7%. The current treatment plan for glioblastoma is mainly maximal safe surgical resection combined with postoperative radiotherapy and chemotherapy, which has been proven to be associated with improved overall survival of patients. Other studies have shown that chemotherapy combined with tumor electric field therapy can improve patient survival. However, no matter what treatment is received, recurrence of glioblastoma is inevitable, and the median progression-free survival is about 7 months.

[0003] At present, significant progress has been made in the understanding of glioblastoma in terms of molecular pathology and related cell signaling pathways. Multi-omics fusion research is also increasingly valued in the field of precision oncology. The results of a multi-omics cohort study of breast cancer showed that multi-omics research can better guide precision treatment decisions than single-omics. Another multi-omics machine learning model based on transcriptomics, proteomics, etc. showed good performance in the prognosis prediction of hepatocellular carcinoma. The development of multi-omics data analysis has made more complex tumor classification methods possible. A study of high-grade serous ovarian cancer used multi-omics data fusion to explore and discover new subtypes of the disease. Therefore, it is very necessary to develop a multimodal model to predict the prognosis of glioblastoma.

[0004] Multi-omics data fusion mainly refers to the use of computers to perform comprehensive processing of multi-omics data, which can extract and integrate information from different omics in order to make better decisions. At present, the fusion of multimodal data is mostly completed based on machine learning model methods. Medical data is complex and diverse. Through multi-omics fusion algorithms, complementary information from different sources can be used to reflect the consensus information of the inherent nature. Multi-omics data fusion is widely used in the medical field and can be used for the diagnosis and treatment of cancer and prognosis, and to establish a prognosis prediction model.

[0005] However, most of the existing machine learning-based glioma prediction models are based on a single factor, and the relationship between different omics needs to be further explored, and the prediction performance and stability of the model need to be improved. In terms of data sources, a larger sample size of international multi-center, multi-ethnic datasets is needed to train and validate glioblastoma prediction models. Summary of the invention

[0006] The present application provides a multi-omics fusion IDH wild-type glioblastoma prognosis assessment method and system, which constructs a comprehensive prognostic model for IDH wild-type glioblastoma patients by integrating pathological omics, proteomics, and genomic features, thereby increasing the prognostic prediction value of glioblastoma.

[0007] In order to solve the above technical problems, in the first aspect, the embodiment of the present application provides a multi-omics fusion IDH wild-type glioblastoma prognosis assessment method, comprising the following steps: first, obtaining a data set; the data set includes an IDH wild-type glioblastoma patient data set; then, based on the data set, obtaining pathological images, gene data and transcriptional data respectively; and extracting omics features from the pathological images, gene data and transcriptional data respectively; the omics features include: pathological omics features, mutation omics features, copy number variation omics features and transcriptomics features; next, based on the omics features, constructing Single-omics prognostic models corresponding to omics features; single-omics prognostic models include: pathological omics prognostic model, mutation omics prognostic model, copy number variation omics prognostic model and transcriptomics prognostic model; then, the omics features of the single-omics prognostic model were screened to obtain multi-omics fusion features, and based on the multi-omics fusion features, a multi-omics fusion model based on machine learning was constructed; based on the multi-omics fusion model, patients were risk stratified and divided into high-risk group and low-risk group; finally, the high-risk group and the low-risk group were differentially analyzed, and the cross-pathways were obtained by weighted gene co-expression network analysis (WGCNA) and gene set enrichment analysis (GSEA) to determine the key biological pathways and biological behaviors that affect prognosis in the multi-omics fusion model.

[0008] In some exemplary embodiments, a pathological image is obtained based on a data set, and pathological omics features are extracted from the pathological image, including: using a pathological slice scanner to perform tissue slice scanning on the patient's pathological slices to obtain a whole-field digital pathological image (Whole Slide Image, WSI); preprocessing the whole-field digital pathological image and screening the whole-field digital pathological image; segmenting and feature extracting the preprocessed and screened pathological image, and extracting pathological omics features.

[0009] In some exemplary embodiments, the full-field digital pathology image is preprocessed and the full-field digital pathology image is screened; the preprocessed and screened pathology image is segmented and feature extracted, and pathology omics features are extracted, including: using an image search engine to fully automatically preprocess the full-field digital pathology image, segmenting the pathology image into multiple 1024×1024 plaques; classifying the plaques into tissue plaques and non-tissue plaques; wherein plaques with a tissue content of less than 85% are regarded as non-tissue plaques; performing K-means clustering analysis on the tissue plaques according to the imaging mode of the tissue plaques to obtain several plaque clusters, and selecting the optimal plaque from each plaque cluster; performing feature extraction on the optimal plaque selected from each plaque cluster to obtain pathology omics features.

[0010] In some exemplary embodiments, feature extraction is performed on the optimal plaque selected from each plaque cluster to obtain pathological omics features, including: using an unmixing color module to unmix a representative image at 1024×1024 pixels to identify the foreground and background of the target tissue; automatically identifying the cell nucleus, cell body and cytoplasm through three modules with adaptive thresholds; the three modules are respectively an identification primary module, an identification secondary module, and an identification tertiary module; using functional modules to perform in-depth analysis on the image to measure the object features in the image; the functional modules include: an object size and shape measurement module, a texture measurement module, a granularity measurement module, a relationship measurement module between objects, an image occupied area measurement module, an object intensity measurement module and an object intensity distribution measurement module; the object features include the size, shape, texture, density, distribution and intensity features of the cell nucleus and cytoplasm; statistical analysis is performed on the measured object features to obtain different statistical indicators to comprehensively and objectively describe the features and properties of the objects in the image; the statistical indicators include mean, median and standard deviation.

[0011] In some exemplary embodiments, after extracting the omics features and before constructing the single omics prognostic models corresponding to the omics features based on the omics features, it also includes: preprocessing the omics features and screening the preprocessed omics features.

[0012] In some exemplary embodiments, the omics features are preprocessed and the preprocessed omics features are screened, including: logarithmic transformation of pathological omics features, followed by z-score standardization, so that different features have a unified dimension and scale, thereby eliminating the impact of the dimension, so as to screen out key features for building the model; the mutation omics features are recorded as binary variables, with mutations recorded as 1 and non-mutations recorded as -1, and genes with at least 3 or more mutations in the enrolled patients are screened out as key features for subsequent analysis; the copy number variation omics features and transcriptomics features are determined using the absolute median difference method. The 50% quantile is determined to identify and eliminate outliers to improve the robustness of the model; and then the z-score method is used for standardization to screen out key features for building the model.

[0013] In some exemplary embodiments, based on the omics features, single omics prognostic models corresponding to the omics features are respectively constructed, including: key features obtained through feature screening based on pathological omics features, mutation omics features, copy number variation omics features and transcriptomics features, each of which is combined into a single omics prognostic model corresponding to the omics features; the key features are weighted using the coefficients of ridge regression to form a linear combination model; based on the linear combination model, the risk scores of each omics are calculated, which are Path Score, Mut Score, CNV Score and FPKM Score, respectively.

[0014] In some exemplary embodiments, the omics features of the single-omics prognostic model are screened to obtain multi-omics fusion features, and based on the multi-omics fusion features, a multi-omics fusion model based on machine learning is constructed, including: determining the cutoff value of the risk score of each omics in the training set of the data set, and dividing the patients into a low-risk group and a high-risk group; evaluating the relationship between each omics feature and the survival time by survival analysis curve analysis, and using the log-rank test method to evaluate the survival difference between the two groups; applying the cutoff value to the internal validation data set, calculating the cutoff value of the external validation set based on the statistical characteristics of the external validation set, and dividing them into a high-risk group and a low-risk group, and then drawing a survival analysis curve and a log-rank test; performing univariate Cox analysis on clinical pathological risk factors and the risk scores of each omics to obtain factors that are significant to the patient's prognosis; and comparing P All factors with a p<0.05 were included in the multivariate Cox analysis to obtain independent risk factors, and the risk scores of each omics were evaluated as independent risk factors. Patients with four-omics feature data were screened out to build a multi-omics fusion model. The risk scores of each omics after verification were used as new features, and the coefficients of each omics risk score were calculated using ridge regression. After weighting, the fused linear combination model was obtained, and the fused risk score was calculated, recorded as Multscore.

[0015] In some exemplary embodiments, weighted gene co-expression network analysis and gene set enrichment analysis methods are used to obtain cross-pathways, and key biological pathways and biological behaviors that affect prognosis in the multi-omics fusion model are determined, including: using weighted gene co-expression network analysis and gene set enrichment analysis to obtain respective enrichment pathway results; using weighted gene co-expression network analysis and gene set enrichment analysis to obtain cross-biological pathways, calculate the Gene Set Variation Analysis (GSVA) ​​score corresponding to each pathway, perform Pearson correlation analysis with the risk score of the multi-omics fusion model, and study the relationship between each characteristic value in the risk score of the multi-omics fusion model and the cross-biological pathway, so as to compare the expression differences in key biological pathways between the high-risk group and the low-risk group divided by the multi-omics fusion model, reveal the biological mechanism behind the multi-omics fusion characteristics, and provide potential biomarkers for clinical treatment.

[0016] In the second aspect, the embodiment of the present application also provides a multi-omics fusion IDH wild-type glioblastoma prognosis evaluation system, including: a dataset module, a multi-omics feature module, a single-omics prognosis model module, a multi-omics fusion model module and a data analysis module connected in sequence; the dataset module is used to obtain a dataset; the dataset includes an IDH wild-type glioblastoma patient dataset; the multi-omics feature module is used to obtain pathological images, gene data and transcriptional data according to the dataset; and extract omics features from the pathological images, gene data and transcriptional data respectively; the omics features include: pathological omics features, mutation omics features, copy number variation omics features and transcriptomics features; the single-omics prognosis model module is used to, according to the omics features, respectively. A single-omics prognostic model corresponding to the omics features is constructed; the single-omics prognostic model includes: a pathological omics prognostic model, a mutation omics prognostic model, a copy number variation omics prognostic model and a transcriptomics prognostic model; the multi-omics fusion model module is used to screen the omics features of the single-omics prognostic model to obtain multi-omics fusion features, and based on the multi-omics fusion features, a multi-omics fusion model based on machine learning is constructed; based on the multi-omics fusion model, patients are risk stratified and divided into high-risk group and low-risk group; the data analysis module is used to perform differential analysis on the high-risk group and the low-risk group, and weighted gene co-expression network analysis and gene set enrichment analysis methods are used to obtain cross-pathways, and determine the key biological pathways and biological behaviors that affect prognosis of the multi-omics fusion model.

[0017] The technical solution provided by the embodiments of the present application has at least the following advantages:

[0018] The present application embodiment provides a multi-omics fusion IDH wild-type glioblastoma prognosis assessment method and system, the method comprising the following steps: first, obtaining a data set; the data set comprises an IDH wild-type glioblastoma patient data set; then, based on the data set, obtaining pathological images, gene data and transcriptional data respectively; and extracting omics features from the pathological images, gene data and transcriptional data respectively; the omics features comprise: pathological omics features, mutation omics features, copy number variation omics features and transcriptomics features; next, based on the omics features, constructing a single omics prognosis model corresponding to the omics features respectively; the single omics prognosis model comprises Including: pathological omics prognostic model, mutation omics prognostic model, copy number variation omics prognostic model and transcriptomics prognostic model; then, the omics features of the single-omics prognostic model were screened to obtain multi-omics fusion features, and based on the multi-omics fusion features, a multi-omics fusion model based on machine learning was constructed; based on the multi-omics fusion model, the patients were risk stratified and divided into high-risk group and low-risk group; finally, the high-risk group and the low-risk group were differentially analyzed, and the weighted gene co-expression network analysis and gene set enrichment analysis methods were used to obtain the cross-pathways, and the key biological pathways and biological behaviors that the multi-omics fusion model affects the prognosis were determined.

[0019] The multi-omics fusion IDH wild-type glioblastoma prognosis assessment method provided in this application, on the one hand, this application establishes a multi-omics model based on machine learning and verifies the risk stratification of glioblastoma. On the other hand, multi-omics analysis uses two methods, WGCNA and GSEA, to analyze disease-related biological pathways from different angles and cross-validate the analysis results. Then, the cross-pathways of the two are taken to reveal the biological basis behind the multi-omics model, enhancing the reliability and accuracy of the enrichment pathway. In addition, the multi-omics model provided in this application analyzes the pathological omics characteristics, genes and transcriptional information of glioblastoma patients. By further integrating this information, this application studies the biological mechanisms and related signaling pathways of glioblastoma, which helps this application to better understand the physiological mechanisms of the disease. BRIEF DESCRIPTION OF THE DRAWINGS

[0020] One or more embodiments are exemplarily described by pictures in the corresponding drawings, and these exemplifications do not constitute limitations on the embodiments. Unless otherwise stated, the pictures in the drawings do not constitute proportional limitations.

[0021] Figure 1 A schematic flow chart of a multi-omics fusion method for prognosis assessment of IDH wild-type glioblastoma provided in one embodiment of the present application.

[0022] Figure 2AAn embodiment of the present application provides a survival difference analysis chart for evaluating the high-risk and low-risk groups in the training set during the screening of the omics features of the single-omics prognostic model.

[0023] Figure 2B An embodiment of the present application provides a survival difference analysis chart for evaluating the high-risk and low-risk groups in the validation set during the screening of the omics features of the single-omics prognostic model.

[0024] Figure 3 This is a survival difference analysis chart for evaluating the high-risk and low-risk groups in a multi-omics fusion model provided in one embodiment of the present application.

[0025] Figure 4 A schematic diagram of the structure of a multi-omics fusion IDH wild-type glioblastoma prognosis assessment system provided in one embodiment of the present application. DETAILED DESCRIPTION

[0026] As can be seen from the background technology, most of the existing machine learning-based glioma prediction models are based on a single factor. The connection between different omics needs to be further explored, and the prediction performance and stability of the model need to be improved. In terms of data sources, a larger sample size of international multi-center, multi-ethnic data sets are needed to train and validate glioblastoma prediction models.

[0027] In order to solve the above technical problems, the embodiment of the present application provides a multi-omics fusion IDH wild-type glioblastoma prognosis assessment method and system, the method comprising the following steps: first, obtaining a data set; the data set includes an IDH wild-type glioblastoma patient data set; then, based on the data set, respectively obtaining pathological images, gene data and transcriptional data; and respectively extracting omics features from the pathological images, gene data and transcriptional data; the omics features include: pathological omics features, mutation omics features, copy number variation omics features and transcriptomics features; next, based on the omics features, respectively constructing a single omics prognosis model corresponding to the omics features; the single omics prognosis model The prognostic model includes: pathological omics prognostic model, mutation omics prognostic model, copy number variation omics prognostic model and transcriptomics prognostic model; then, the omics features of the single omics prognostic model are screened to obtain multi-omics fusion features, and based on the multi-omics fusion features, a multi-omics fusion model based on machine learning is constructed; based on the multi-omics fusion model, the patients are risk stratified and divided into high-risk group and low-risk group; finally, the high-risk group and the low-risk group are differentially analyzed, and the cross-pathways are obtained by weighted gene co-expression network analysis and gene set enrichment analysis methods to determine the key biological pathways and biological behaviors that affect the prognosis of the multi-omics fusion model. The present application provides a multi-omics fusion IDH wild-type glioblastoma prognosis evaluation method and system, which integrates pathological omics, proteomics, and genomic features to construct a comprehensive prognostic model for IDH wild-type glioblastoma patients, thereby increasing the prognostic prediction value of glioblastoma.

[0028] The following will describe the various embodiments of the present application in detail with reference to the accompanying drawings. However, it will be appreciated by those skilled in the art that in the various embodiments of the present application, many technical details are provided in order to enable the reader to better understand the present application. However, even without these technical details and various changes and modifications based on the following embodiments, the technical solution claimed in the present application can be implemented.

[0029] See also Figure 1 The present application embodiment provides a multi-omics fusion IDH wild-type glioblastoma prognosis assessment method, comprising the following steps:

[0030] Step S1, obtaining a data set; the data set includes an IDH wild-type glioblastoma patient data set.

[0031] Step S2: based on the data set, pathological images, gene data and transcriptional data are obtained respectively; and omics features are extracted from the pathological images, gene data and transcriptional data respectively; the omics features include: pathological omics features, mutation omics features, copy number variation omics features and transcriptomics features.

[0032] Step S3: Based on the omics features, constructing the single-omics prognostic models corresponding to the omics features respectively; the single-omics prognostic models include: pathology omics prognostic model, mutation omics prognostic model, copy number variation omics prognostic model and transcriptomics prognostic model.

[0033] Step S4: Screen the omics features of the single-omics prognostic model to obtain multi-omics fusion features, and build a multi-omics fusion model based on machine learning based on the multi-omics fusion features; based on the multi-omics fusion model, risk stratify the patients into high-risk group and low-risk group.

[0034] Step S5: Perform differential analysis on the high-risk group and the low-risk group, use weighted gene co-expression network analysis and gene set enrichment analysis to obtain cross-pathways, and determine the key biological pathways and biological behaviors that affect prognosis in the multi-omics fusion model.

[0035] The multi-omics fusion prognostic evaluation method for IDH wild-type glioblastoma provided in this application first performs feature preprocessing on pathological omics, genomics, and transcriptomics, and randomly divides the enrolled patients into a training group and a test group at a ratio of 6:4, and then performs univariate COX analysis and Light GBM (Light Gradient Boosting Machine) regression analysis on the training group of each omics to screen out the best pathological omics features, genomic features, and transcriptomics features. Subsequently, a prognostic model was established using the Ridge regression model (Ridge fitting), and the risk score (Riskscore) of each omics was calculated to predict the patient's survival prognosis. According to the risk score and survival data of each omics, an appropriate cutoff value was selected to divide the patients into high-risk and low-risk groups, and the Kaplan-Meier curve analysis and multivariate Cox regression analysis were used to verify the correlation between each omics feature and the patient's survival period. External validation was performed using glioblastoma data in the TCGA and CGGA public databases to determine the survival analysis curves (Kaplan-Meier, KM) of high and low risk groups divided according to the model. The risk scores of each group of internal and external validation were used as new features, and the ridge regression model was used to fuse multi-omics features and calculate the risk score (multscore) of the fusion model. Patients with complete multi-omics data were screened for model reliability verification, and Kaplan-Meier curves and multifactor COX analysis were performed. Based on independent risk factors, a nomogram was drawn for personalized prediction of patient survival. A DCA decision curve was developed to verify the clinical practicality of the nomogram, and a calibration curve was used to test the predictive efficacy of the model. Finally, a differential analysis was performed on the high- and low-risk groups after multi-omics fusion, and cross-pathways were obtained using WGCNA and GSEA analysis methods to explore the pathways and biological behaviors that the fusion model affects prognosis, and further biological interpretation was performed to enhance the biological significance of the model.

[0036] This application uses the "Cellprofiler" software to extract pathological omics features and integrates genomic and transcriptomic information to construct a comprehensive prognostic model for patients with IDH wild-type glioblastoma, and further explores the advantages of this prognostic model in terms of prediction increment, stability and reliability compared to traditional prognostic models based on clinical and pathological factors. In addition, this application also involves the use of paired pathological features, whole-exome sequencing (WES) and high-throughput transcriptome sequencing (RNA sequencing, RNA-seq) data to explore the biological mechanisms behind the multi-omics fusion model. This application uses integrated analysis of pathological characteristics and gene expression data. Compared with single-omics, multi-omics data fusion can better reflect disease characteristics, significantly improve the predictive performance and stability of the model, and better guide personalized clinical decision-making; by performing differential analysis on the high-risk group and low-risk group divided by the multi-omics fusion model, and using weighted gene co-expression network analysis and gene set enrichment analysis methods to obtain cross-pathways, it is possible to conduct a more in-depth study of the biological mechanisms of glioblastoma and its related signaling pathways, aiming to provide a more accurate reference for the formulation of treatment strategies and prognosis evaluation for IDH wild-type glioblastoma patients.

[0037] The data set in step S1 is mainly obtained from cases of patients who underwent brain tumor surgical resection in the neurosurgery department of the hospital and were diagnosed with glioblastoma according to CNS5.

[0038] The inclusion criteria for the patient cohort were: (1) cases diagnosed with glioblastoma by postoperative histopathology and molecular testing; (2) the completeness and availability of survival data from postoperative follow-up; (3) the completeness and availability of pathological HE-stained sections, whole exome sequencing (WES) data, or RNA sequencing data of surgical resection specimens; and (4) patients aged ≥18 years.

[0039] The exclusion criteria for the patient cohort were as follows: (1) patients who did not undergo surgical treatment or only underwent puncture biopsy at the time of diagnosis; (2) patients who did not undergo surgical treatment in this hospital; (3) patients with poor HE staining effect, artifacts, excessive bleeding, and other quality problems in the surgically removed tissue specimens; (4) patients who were lost to follow-up after surgery or died of non-glioblastoma-related causes; and (5) patients with a follow-up survival date of less than 2 months.

[0040] After the above rigorous screening process, patients who met the criteria were included in the study cohort. It is worth noting that some patients in the study cohort had paired whole exome sequencing (WES) data or RNA sequencing data. In addition, this application collected detailed clinical and pathological data of each patient through the clinical medical record system, including: gender, age, radiotherapy, chemotherapy, tumor grade, preoperative Karnofsky Performance Status (KPS), and degree of tumor resection (complete or incomplete). Follow-up data came from telephone follow-up and patient review.

[0041] In some embodiments, in step S2, based on the data set, gene data and transcription data are obtained by WES and RNA sequencing (RNA-seq), respectively. In order to deeply understand the molecular mechanism of glioblastoma and provide molecular data support for the construction of a prognostic evaluation model, this application further tested the IDH or 1p / 19q gene status for some patients who underwent surgical resection of tumor tissue on the basis of conventional pathological testing. For patients who have undergone IDH or 1p / 19q pathological testing, their test results are directly used. For patients who have not undergone these tests, their embedded tissue wax blocks are borrowed from the pathology department on the premise of obtaining informed consent from the patients. Sanger sequencing and fluorescence in situ hybridization (FISH) technology are used to detect the corresponding molecular information. In addition, some of the enrolled tumor specimens are selected for WES and RNA-seq. Through WES, mutation (Mutation) and copy number variation (CNV) data of glioblastoma patients are obtained. Through RNA-seq, transcriptomics (Fragments Per Kilobase of transcript per Million mapped reads, FPKM) data were obtained to reflect the expression level of genes. Mutation data and CNV data were used to analyze the genetic variation characteristics of tumors. FPKM data were used to evaluate gene expression patterns and further explore the molecular mechanisms associated with the occurrence and development of glioblastoma. Integrating WES and RNA-seq data provided molecular evidence for the construction of a multi-omics prognostic evaluation model.

[0042] In some embodiments, in step S2, a pathological image is obtained based on the data set, and pathological omics features are extracted from the pathological image, including:

[0043] Step S201: Scan the patient's pathological slices with a pathological slice scanner to obtain a full-field digital pathological image.

[0044] Step S202: preprocess the full-field digital pathology image and screen the full-field digital pathology image; segment and extract features from the preprocessed and screened pathology image, and extract pathology genomics features.

[0045] To ensure the quality and consistency of pathological image data, this application uses standardized equipment and parameters to collect full-field digital pathology images (WSI). All patients in the study cohort underwent routine HE staining after surgery, and the pathological sections of the patients borrowed from the pathology department were scanned at a magnification of 20 times using a high-resolution digital pathology section scanner (KF-PRO-120-HI) to obtain full-field digital pathology images. Glioblastoma WSI section data were downloaded from the TCGA public database to verify the consistency and reliability of pathological genomic features in different patient groups.

[0046] In some embodiments, in step S202, the full-field digital pathology image is preprocessed and the full-field digital pathology image is screened; the preprocessed and screened pathology image is segmented and feature extracted, and pathology genomics features are extracted, including: using an image search engine to fully automatically preprocess the full-field digital pathology image, segmenting the pathology image into multiple 1024×1024 plaques; classifying the plaques into tissue plaques and non-tissue plaques; wherein plaques with a tissue content of less than 85% are regarded as non-tissue plaques; performing K-means clustering analysis on the tissue plaques according to the imaging mode of the tissue plaques to obtain several plaque clusters, and selecting the optimal plaque from each plaque cluster; performing feature extraction on the optimal plaque selected from each plaque cluster to obtain pathology genomics features.

[0047] Specifically, this application adopts advanced image processing technology and uses an image search engine called "Yottixel" to fully automate the preprocessing of glioblastoma WSI. The processing process includes three main steps: first, automatic segmentation, paving WSI into many 1024×1024 patches; second, classification of tissue and non-tissue plaques, plaques with less than 85% tissue content are considered non-tissue plaques; finally, K-means clustering analysis of tissue plaques is performed according to the imaging mode of tissue plaques. In this way, each patient obtains three different plaque clusters (K0, K1, K2), and plaques in the same plaque cluster are considered to have similar imaging characteristics. Subsequently, experienced pathologists select the top 5 plaques with the best image quality from each plaque cluster for further analysis.

[0048] In some embodiments, feature extraction is performed on the optimal plaque selected from each plaque cluster to obtain pathological omics features, including: using an unmix colors module to unmix a representative image by 1024×1024 pixels, thereby identifying the foreground and background of the target tissue; automatically identifying the cell nucleus, cell body and cytoplasm through three modules with adaptive thresholds; the three modules are respectively an identification module (Identify Primary Objects), an identification module (Identify Secondary Objects), and an identification module (Identify TertiaryObjects); using functional modules to perform in-depth analysis on the image to measure the object features in the image; the functional modules include: an object size and shape measurement module, a texture measurement module, a granularity measurement module, a relationship measurement module between objects, an image occupied area measurement module, an object intensity measurement module and an object intensity distribution measurement module; the object features include the size, shape, texture, density, distribution and intensity features of the cell nucleus and cytoplasm; statistical analysis is performed on the measured object features to obtain different statistical indicators to comprehensively and objectively describe the features and properties of the objects in the image; the statistical indicators include mean, median and standard deviation.

[0049] Specifically, when extracting pathological omics features, the Cellprofiler software was used to segment and extract features from the pre-processed and screened pathological images, and a large number of quantitative pathological omics features were extracted. The process first unmixed the representative images by using the unmixed color module. Then, the recognition primary module, recognition secondary module, and recognition tertiary module with adaptive Otsu threshold were used to automatically identify the nucleus, cell body, and cytoplasm. Then, the recognition objects were measured by multiple modules such as "measuring object size and shape", "measuring texture", "measuring granularity", "measuring the relationship between objects", "measuring image occupied area", "measuring object intensity", and "measuring object intensity distribution". The measurement results of these modules were expressed by calculating the mean, median, and standard deviation to represent the final features. Finally, 2382 image features were obtained, covering the defined features of the size, shape, texture, pixel intensity distribution, and neighboring relationship of the nucleus and cytoplasm. The mean of each selected feature was extracted for the five pathological images of the three feature sequences of each sample. After eliminating useless features, a total of 6456 useful quantitative pathological omics image features were retained to provide data support for subsequent analysis.

[0050] In some embodiments, after extracting the omics features in step S2 and before constructing the single omics prognostic models corresponding to the omics features based on the omics features in step S3, it also includes: preprocessing the omics features and screening the preprocessed omics features.

[0051] It should be noted that when preprocessing multi-omics features, appropriate data preprocessing methods are selected for the extracted pathological features, mutation (Mut) features, copy number variation (CNV) features, and FPKM features according to the data itself.

[0052] In some embodiments, the omics features are preprocessed and the preprocessed omics features are screened, including: logarithmic transformation of pathological omics features to reduce the dynamic range of the data and reduce the impact of extreme values. Then z-score standardization is performed to make different features have a unified dimension and scale, thereby eliminating the impact of the dimension to screen out the key features for building the model; the mutation omics features are recorded as binary variables, with mutations recorded as 1 and non-mutations recorded as -1, and genes with at least 3 or more mutations in the enrolled patients are screened as key features for subsequent analysis; the copy number variation omics features and transcriptomics features use the Median Absolute Deviation (MAD) method to determine the 50% quantile to identify and eliminate outliers to improve the robustness of the model; and then the z-score method is used for standardization to screen out the key features for building the model.

[0053] The pre-treated omics features were screened. The feature screening strategy included: (1) The pre-treated features were analyzed using a univariate Cox proportional hazard model to evaluate the relationship between each feature and patient survival time, and to screen out features that were significantly correlated with prognosis (p<0.05). (2) The features significantly correlated with prognosis were further analyzed using the Light GBM model. LightGBM is an efficient machine learning algorithm based on a gradient boosting framework. By setting the objective function, evaluation index, number of leaves, maximum depth, learning rate, number of iterations, and regularization parameters, the generalization ability of the model is improved and the risk of overfitting is reduced. The 30 features with the greatest impact on prognosis survival time were screened out. (3) SHAP (SHapley Additive exPlanations) analysis was used to quantify the contribution of each feature to model prediction. The SHAP value can be used to evaluate the global importance of the feature to the model prediction, thereby improving the interpretability and reliability of the model.

[0054] In some embodiments, in step S3, based on the omics features, single omics prognostic models corresponding to the omics features are respectively constructed, including: key features obtained by feature screening based on pathological omics features, mutation omics features, copy number variation omics features and transcriptomics features are each combined into a single omics prognostic model corresponding to the omics features; the key features are weighted using the coefficients of ridge regression to form a linear combination model; based on the linear combination model, the risk scores (Risk Score) of each omics are calculated, which are Path Score, Mut Score, CNV Score and FPKM Score respectively.

[0055] In some embodiments, in step S4, the omics features of the single-omics prognostic model are screened to obtain multi-omics fusion features, and based on the multi-omics fusion features, a multi-omics fusion model based on machine learning is constructed, including: using the R package "survminer" to determine the cutoff value of the risk score of each omics in the training set of the data set, and dividing the patients into a low-risk group and a high-risk group; evaluating the relationship between each omics feature and survival time (OS) by survival analysis curve analysis, and using the log-rank test (Log-rank) to evaluate the survival difference between the two groups (such as Figure 2A The cutoff value was applied to the internal validation data set, and the cutoff value of the external validation set was calculated based on the statistical characteristics of the external validation set. After the groups were divided into high-risk group and low-risk group, the survival analysis curve and log-rank test were drawn (as shown in Figure 2B ); Univariate Cox analysis was performed on clinical pathological risk factors and risk scores of each omics to obtain factors that were significant for patient prognosis; all factors with P < 0.05 were included in the multivariate Cox analysis to obtain independent risk factors, and the risk scores of each omics were evaluated as independent risk factors; patients with four-omics feature data were screened out to construct a multi-omics fusion model; the risk scores of each omics after verification were used as new features, and the coefficients of each omics risk score were calculated using ridge regression. After weighting, the fused linear combination model was obtained, and the fused risk score was calculated, recorded as Multscore.

[0056] After obtaining the fused linear combination model, the fused risk score was calculated, and the R package “survminer” was used again to determine the risk score cutoff value of the fusion model. The fusion model was divided into high-risk and low-risk groups. The relationship between the fusion model characteristics and OS was evaluated by survival analysis curves and Log-rank tests, as well as the survival differences between the high-risk and low-risk groups of the fusion model (e.g. Figure 3). The clinicopathological risk factors and fusion multi-omics risk scores were further subjected to univariate Cox analysis to obtain factors that were significant for patient prognosis. Then, all factors with P < 0.05 were included in the multivariate Cox analysis to obtain independent risk factors, and the multi-omics feature risk score was evaluated as an independent risk factor. Based on the multivariate Cox analysis, the independent clinicopathological risk factors were incorporated into the clinicopathological nomogram, and the multi-omics fusion-clinical nomogram was the fused multi-omics risk score combined with the independent clinicopathological risk factors. The two nomograms were compared to evaluate the incremental value of the multi-omics fusion model for patient prognosis assessment. The R package rms was used to draw the calibration curve, compare the calibration and discrimination performance of the multi-omics fusion-clinical pathology nomogram and the clinicopathological nomogram, and compare the calibration performance between observed and predicted survival. The C-index was calculated using the R package survival as an indicator for evaluating the prognostic model. Net Reclassification Improvement (NRI) was used to evaluate the advantages and disadvantages of the multi-omics fusion model, and Akaike Information Criterion (AIC) was used to evaluate the overfitting risk of the model. Finally, the R package rmda was used to draw the decision curve to evaluate the practical application value of the model in clinical practice.

[0057] In some embodiments, step S5 uses weighted gene co-expression network analysis and gene set enrichment analysis methods to obtain cross-pathways and determine the key biological pathways and biological behaviors that affect prognosis in the multi-omics fusion model, including: using weighted gene co-expression network analysis (WGCNA) and gene set enrichment analysis (GSEA) to obtain respective enrichment pathway results; using weighted gene co-expression network analysis and gene set enrichment analysis to analyze the cross-biological pathways, calculate the GSVA score corresponding to each pathway, perform Pearson correlation analysis with the risk score of the multi-omics fusion model, and study the relationship between each characteristic value in the risk score of the multi-omics fusion model and the cross-biological pathway, thereby comparing the expression differences between the high-risk group and the low-risk group divided by the multi-omics fusion model in the key biological pathways, revealing the biological mechanism behind the multi-omics fusion characteristics, and providing potential biomarkers for clinical treatment.

[0058] WGCNA can be used to analyze gene expression data, identify co-expressed genes, and group them into modules or clusters based on their similar expression patterns. Co-expression network analysis of tumor tissue gene expression data can be implemented using the R package WGCNA. The following is a detailed description of the WGCNA process: (1) Data preprocessing: The data is normalized to eliminate any systematic bias and remove low-quality samples. (2) Gene co-expression network construction: The correlation matrix is ​​first converted into an adjacency matrix using the weighted value of the correlation between genes. The adjacency matrix is ​​then converted into a topological overlap matrix (TOM), which measures the degree of shared neighbors between two genes. TOM is used to construct a network of interconnected gene modules, where each module represents a group of highly co-expressed genes. (3) Module detection: Hierarchical clustering is used to identify gene modules in the co-expression network. First, the difference between the TOMs of each gene pair is calculated, and then a clustering algorithm is used to group genes with similar TOMs into modules. The resulting dendrogram can be cut at a specific height to create a specified number of gene modules. (4) Module characterization: Each gene module has a representative expression profile, called module eigengene (ME). It is the overall characteristic of gene expression in the module and can reflect the gene expression status of the module. ME is the first principal component of the gene expression matrix in the module. Then, based on the GSVA of the samples, the modules significantly associated with the multi-omics fusion score were determined, where the false discovery rate (FDR) corrected P < 0.05 indicated significance. In order to identify the enriched biological pathways and processes within the gene module, the R package "clusterProfiler" was used to obtain the functional enrichment pathways, with FDR < 0.05 indicating statistical significance.

[0059] GSEA is a method for analyzing gene expression data to identify differentially expressed gene sets. The following is a detailed description of the GSEA process: (1) Data preprocessing: The data is normalized to remove any systematic bias and to scale the expression values. The resulting expression matrix should contain only high-quality samples and genes. (2) Gene set selection: Gene sets representing specific biological pathways, processes, or functions are selected from publicly available databases (KEGG, Reactome, Gene Ontology, Hallmark, PID, BioCarta, WikiPathways). (3) Calculation of enrichment scores: The expression of genes in the gene set is compared with the expression of genes outside the gene set to determine whether the gene set is significantly enriched under a specific biological condition. The calculated enrichment score (ES) reflects the degree to which the genes in the gene set are overrepresented at the top or bottom of the gene ranking list based on the differential expression between the two conditions. (4) Statistical significance test: Randomly arrange sample labels and recalculate the enrichment score of each gene set, repeat multiple times to generate a null distribution of enrichment scores, which is used to calculate the normalized enrichment score (NES) of each gene set. FDR < 0.05 and |NES| > 1 are selected to indicate significance. Calculate the GSVA value corresponding to the pathway and evaluate its correlation with the multi-omics fusion score. FDR < 0.05 indicates statistical significance.

[0060] Two analysis methods, WGCNA and GSEA, were used to obtain the respective enrichment pathway results. In order to make the results more reliable, accurate and reproducible, and reduce abnormal data and errors, this application uses cross-pathways of the two analysis methods. Then, the GSVA score corresponding to each pathway was calculated, and a Pearson correlation analysis was performed with the prognostic multi-omics fusion score. The multi-omics fusion features were interpreted according to the relevant biological pathways to reveal the biological mechanisms behind the multi-omics fusion features. The differences between biological pathways in high- and low-risk groups divided by multi-omics fusion features were explored to provide potential biomarkers for clinical treatment.

[0061] Based on this, the multi-omics fusion prognostic assessment method for IDH wild-type glioblastoma provided in this application is mainly divided into four core steps: single-omics feature extraction and preprocessing, construction of single-omics prognostic model, multi-omics fusion model development and bioinformatics analysis of multi-omics fusion model. In the process of single-omics feature extraction and preprocessing, this application collected pathological, genetic and transcriptional data of the research cohort. Pathological omics feature extraction includes key steps such as tissue section scanning, WSI image segmentation and feature extraction (such as Figure 2A , Figure 2B These features provide the basis for the construction of subsequent prognostic models.

[0062] When constructing a single-omics prognostic model, the extracted single-omics features are preprocessed to eliminate noise and inconsistency in the data. Subsequently, a single-omics prognostic model is constructed based on these features, and the model performance is verified by rigorous statistical methods to ensure the accuracy and reliability of the model (e.g. Figure 3 In addition, this application also performed an interpretability analysis on the key features of the constructed model to reveal the contribution of individual features to the prognostic model.

[0063] When constructing a multi-omics fusion model, based on the single-omics analysis, this application further integrates the multi-omics data and constructs a multi-omics fusion prognostic model. The fusion model comprehensively considers the information from different data sources, thereby improving the accuracy of prognostic assessment and clinical application value. This application has comprehensively verified the fusion model (such as Figure 4 to ensure its generalizability across different patient populations.

[0064] Finally, the multi-omics fusion model was subjected to bioinformatics analysis. This application conducted an in-depth analysis of the differentially expressed genes (RNA sequencing data) in the high- and low-risk groups defined by the fusion model. Through GSEA and WGCNA analysis, this application identified biological pathways that were significantly associated with the multi-omics fusion score. This step not only revealed the key biological mechanisms behind the multi-omics fusion model, but also provided important information for potential therapeutic targets for glioblastoma.

[0065] See also Figure 4The embodiment of the present application also provides a multi-omics fusion IDH wild-type glioblastoma prognosis evaluation system, including: a data set module 101, a multi-omics feature module 102, a single-omics prognosis model module 103, a multi-omics fusion model module 104 and a data analysis module 105 connected in sequence; the data set module 101 is used to obtain a data set; the data set includes an IDH wild-type glioblastoma patient data set; the multi-omics feature module 102 is used to obtain pathological images, gene data and transcription data according to the data set; and extract omics features from the pathological images, gene data and transcription data respectively; the omics features include: pathological omics features, mutation omics features, copy number variation omics features and transcriptomics features; the single-omics prognosis model module 103 is used to obtain the pathological images, gene data and transcription data according to the data set; and extract omics features from the pathological images, gene data and transcription data respectively; the omics features include: pathological omics features, mutation omics features, copy number variation omics features and transcriptomics features; the single-omics prognosis model module 103 is used to obtain the pathological images, gene data and transcription data according to the data set. Omic features, respectively constructing single-omics prognostic models corresponding to the omics features; the single-omics prognostic models include: pathological omics prognostic model, mutation omics prognostic model, copy number variation omics prognostic model and transcriptomics prognostic model; the multi-omics fusion model module 104 is used to screen the omics features of the single-omics prognostic model, obtain the multi-omics fusion features, and construct a multi-omics fusion model based on machine learning based on the multi-omics fusion features; based on the multi-omics fusion model, the patients are risk stratified and divided into high-risk group and low-risk group; the data analysis module 105 is used to perform differential analysis on the high-risk group and the low-risk group, and use weighted gene co-expression network analysis and gene set enrichment analysis methods to obtain cross-pathways, and determine the key biological pathways and biological behaviors that affect the prognosis of the multi-omics fusion model.

[0066] The present application embodiment provides a multi-omics fusion IDH wild-type glioblastoma prognosis assessment method and system. First, according to the comprehensive diagnostic reclassification proposed by CNS5, more attention is paid to the role of molecular markers in the classification of nervous system tumors, and the molecular diagnosis of glioblastoma is determined to be IDH wild-type glioblastoma. Subsequently, a multi-omics model based on machine learning was established, and the risk stratification of glioblastoma was verified. Secondly, multi-omics analysis uses two methods, WGCNA and GSEA, to analyze disease-related biological pathways from different angles, and cross-validate the analysis results. Then, the cross-pathways of the two are taken to reveal the biological basis behind the multi-omics model, which enhances the reliability and accuracy of the enrichment pathway. Third, the multi-omics model analyzes the pathological omics characteristics, genes and transcriptional information of glioblastoma patients. By further integrating this information, the present application studies the biological mechanisms and related signaling pathways of glioblastoma, which helps the present application to better understand the physiological mechanisms of the disease.

[0067] The present application embodiment also provides a multi-omics fusion IDH wild-type glioblastoma prognosis assessment method and system, the advantage is that: previous studies have revealed the correlation between the MYC gene and the prognosis of glioblastoma treatment by training machine learning models through the genomic characteristics of glioblastoma. Another study combined the proteomics and genomics of glioma to reveal the laws of development and regulation of glioma. These research results show that multi-omics fusion analysis has broad application prospects and important significance in clinical and basic research. In this application, a multi-omics fusion model was established, combining pathological omics, proteomics, and genomic features to deeply study the biological mechanisms and prognostic evaluation of glioblastoma. Further analysis shows that the fusion of multi-omics information can increase the prognostic value and improve the accuracy and stability of the model.

[0068] Genomic studies of gliomas have shown that in glioblastoma, prognostic radiomic features are associated with hypoxia pathways in glioblastoma. These previous studies mainly focused on the analysis of a single group of omics. This application applies a multi-omics fusion analysis model to increase the prognostic prediction value of glioblastoma. WGCNA is a powerful method for analyzing radiogenomic data, identifying gene modules co-expressed with multi-omics fusion models, and providing a method for identifying potential molecular mechanisms of multi-omics fusion models. GSEA is a widely used bioinformatics tool for analyzing gene expression data in the context of predefined gene sets or pathways. Since the two methods have different focuses and analysis ideas on gene enrichment, this application combines the two analysis methods of WGCNA and GSEA to improve the repeatability and reliability of biological pathway identification in multi-omics fusion model analysis.

[0069] Based on the above technical scheme, the embodiment of the present application provides a multi-omics fusion IDH wild-type glioblastoma prognosis assessment method and system, the method comprising the following steps: first, obtaining a data set; the data set comprises an IDH wild-type glioblastoma patient data set; then, based on the data set, obtaining pathological images, gene data and transcriptional data respectively; and extracting omics features from the pathological images, gene data and transcriptional data respectively; the omics features comprise: pathological omics features, mutation omics features, copy number variation omics features and transcriptomics features; next, based on the omics features, constructing a single omics prognosis model corresponding to the omics features respectively; the single omics prognosis The post-models include: pathological omics prognostic model, mutation omics prognostic model, copy number variation omics prognostic model and transcriptomics prognostic model; then, the omics features of the single-omics prognostic model were screened to obtain multi-omics fusion features, and based on the multi-omics fusion features, a multi-omics fusion model based on machine learning was constructed; based on the multi-omics fusion model, the patients were risk stratified and divided into high-risk group and low-risk group; finally, the high-risk group and the low-risk group were differentially analyzed, and the weighted gene co-expression network analysis and gene set enrichment analysis methods were used to obtain the cross-pathways, and the key biological pathways and biological behaviors that affect the prognosis of the multi-omics fusion model were determined.

[0070] The multi-omics fusion IDH wild-type glioblastoma prognosis assessment method provided in this application, on the one hand, this application establishes a multi-omics model based on machine learning and verifies the risk stratification of glioblastoma. On the other hand, multi-omics analysis uses two methods, WGCNA and GSEA, to analyze disease-related biological pathways from different angles and cross-validate the analysis results. Then, the cross-pathways of the two are taken to reveal the biological basis behind the multi-omics model, enhancing the reliability and accuracy of the enrichment pathway. In addition, the multi-omics model provided in this application analyzes the pathological omics characteristics, genes and transcriptional information of glioblastoma patients. By further integrating this information, this application studies the biological mechanisms and related signaling pathways of glioblastoma, which helps this application to better understand the physiological mechanisms of the disease.

[0071] Those skilled in the art can understand that the above-mentioned embodiments are specific examples for implementing the present application, and in practical applications, various changes can be made to them in form and detail without departing from the spirit and scope of the present application. Any person skilled in the art can make their own changes and modifications without departing from the spirit and scope of the present application, so the scope of protection of the present application shall be based on the scope defined in the claims.

Claims

1. A multi-omics fusion method for the prognosis assessment of IDH wild-type glioblastoma, characterized in that: The following steps are involved: Acquire a data set; the data set includes an IDH wild-type glioblastoma patient data set; Based on the data set, pathological images, gene data and transcription data are obtained respectively; and extracting omics features from pathological images, gene data and transcriptional data respectively; the omics features include: pathological omics features, mutation omics features, copy number variation omics features and transcriptomics features; Based on the omics features, single omics prognostic models corresponding to the omics features are respectively constructed; the single omics prognostic models include: a pathology omics prognostic model, a mutation omics prognostic model, a copy number variation omics prognostic model and a transcriptomics prognostic model; The omics features of the single-omics prognostic model are screened to obtain multi-omics fusion features, and based on the multi-omics fusion features, a multi-omics fusion model based on machine learning is constructed; based on the multi-omics fusion model, the patients are risk-stratified into a high-risk group and a low-risk group; The high-risk group and the low-risk group were differentially analyzed, and cross-pathways were obtained using weighted gene co-expression network analysis and gene set enrichment analysis to determine the key biological pathways and biological behaviors that affect prognosis in the multi-omics fusion model; Based on the omics features, single-omics prognostic models corresponding to the omics features are constructed respectively, including: Based on the key features obtained through feature screening based on pathological omics features, mutation omics features, copy number variation omics features and transcriptomics features, each is combined into a single omics prognostic model corresponding to the omics features; The coefficients of ridge regression were used to weight the key features to form a linear combination model. Based on the linear combination model, the risk scores of each group were calculated, namely Path Score, Mut Score, CNV Score and FPKM Score. The omics features of the single-omics prognostic model were screened to obtain multi-omics fusion features, and based on the multi-omics fusion features, a multi-omics fusion model based on machine learning was constructed, including: The cutoff value of the risk score of each group was determined in the training set of the data set, and the patients were divided into a low-risk group and a high-risk group; the relationship between each group characteristic and survival time was evaluated by survival analysis curve analysis, and the log-rank test method was used to evaluate the survival difference between the two groups; The cutoff value is applied to the internal validation data set, and the cutoff value of the external validation set is calculated based on the statistical characteristics of the external validation set. After the two groups are divided into a high-risk group and a low-risk group, a survival analysis curve and a log-rank test are drawn; Univariate Cox analysis was performed on clinical pathological risk factors and risk scores of each group to obtain factors that were significant for patient prognosis; all factors with P < 0.05 were included in multivariate Cox analysis to obtain independent risk factors, and the risk scores of each group were evaluated as independent risk factors; Patients with four-omics feature data were screened out for constructing a multi-omics fusion model. The risk scores of each omics after verification were used as new features, and the coefficients of each omics risk score were calculated using ridge regression. After weighting, the fused linear combination model was obtained, and the fused risk score was calculated, recorded as Multscore.

2. The multi-omics fusion prognosis assessment method for IDH wild-type glioblastoma according to claim 1, characterized in that: Based on the data set, a pathological image is obtained, and pathological omics features are extracted from the pathological image, including: A pathology section scanner is used to scan the patient's pathology sections to obtain full-field digital pathology images; Preprocessing the full-field digital pathology image and screening the full-field digital pathology image; The pre-processed and screened pathological images are segmented and feature extracted, and pathological omics features are extracted.

3. The multi-omics fusion prognosis assessment method for IDH wild-type glioblastoma according to claim 2, characterized in that: Preprocessing the full-field digital pathology image and screening the full-field digital pathology image; Segment and feature extract the pre-processed and screened pathological images, and extract pathological omics features, including: Using an image search engine to perform fully automated preprocessing on the full-field digital pathology image, and segmenting the pathology image into a plurality of 1024×1024 plaques; Plaques were classified into tissue plaques and non-tissue plaques; plaques with tissue content less than 85% were considered non-tissue plaques; Perform K-means clustering analysis on tissue plaques according to their imaging modes to obtain several plaque clusters, and select the best plaque from each plaque cluster; Feature extraction is performed on the optimal plaque selected from each plaque cluster to obtain pathological omics features.

4. The multi-omics fusion prognosis assessment method for IDH wild-type glioblastoma according to claim 3, characterized in that: Feature extraction is performed on the optimal plaque selected from each plaque cluster to obtain pathological omics features, including: The representative images were unmixed at 1024 × 1024 pixels using the unmixed color module to identify the foreground and background of the target tissue; The cell nucleus, cell body and cytoplasm are automatically identified through three modules with adaptive thresholds; the three modules are identification of primary module, identification of secondary module and identification of tertiary module; The functional modules are used to deeply analyze the image and measure the features of the object in the image; the functional modules include: object size and shape measurement module, texture measurement module, granularity measurement module, object relationship measurement module, image occupied area measurement module, object intensity measurement module and object intensity distribution measurement module; the object features include the size, shape, texture, density, distribution and intensity features of the cell nucleus and cytoplasm; The measured object features are statistically analyzed to obtain different statistical indicators to comprehensively and objectively describe the features and properties of the objects in the image; the statistical indicators include mean, median and standard deviation.

5. The multi-omics fusion prognosis assessment method for IDH wild-type glioblastoma according to claim 1, characterized in that: After extracting the omics features and before constructing the single omics prognostic models corresponding to the omics features based on the omics features, the method further includes: The omics features are preprocessed and the preprocessed omics features are screened.

6. The multi-omics fusion prognosis assessment method for IDH wild-type glioblastoma according to claim 5, characterized in that: Preprocess the omics features and screen the preprocessed omics features, including: The pathological omics features were logarithmically transformed and then z-score standardized to make different features have a unified dimension and scale, thereby eliminating the impact of the dimension and screening out the key features for building the model; The mutagenesis features were recorded as binary variables, with mutations recorded as 1 and no mutations recorded as -1. Genes with at least 3 or more mutations in the enrolled patients were screened as key features for subsequent analysis; The copy number variation omics features and transcriptomics features used the absolute median difference method to determine the 50% quantile to identify and eliminate outliers to improve the robustness of the model; then the z-score method was used for standardization to screen out the key features for building the model.

7. The multi-omics fusion prognosis assessment method for IDH wild-type glioblastoma according to claim 1, characterized in that: Weighted gene co-expression network analysis and gene set enrichment analysis were used to obtain cross-pathways and determine the key biological pathways and biological behaviors that affect prognosis in the multi-omics fusion model, including: Two analysis methods, weighted gene co-expression network analysis and gene set enrichment analysis, were used to obtain the respective enrichment pathway results; The cross-biological pathways of weighted gene co-expression network analysis and gene set enrichment analysis were used to calculate the gene set variation analysis score corresponding to each pathway, and Pearson correlation analysis was performed with the risk score of the multi-omics fusion model to study the relationship between each characteristic value in the risk score of the multi-omics fusion model and the cross-biological pathway, thereby comparing the expression differences in key biological pathways between the high-risk group and the low-risk group divided by the multi-omics fusion model, revealing the biological mechanism behind the multi-omics fusion characteristics, and providing potential biomarkers for clinical treatment.

8. A multi-omics fusion IDH wild-type glioblastoma prognosis assessment system, characterized in that: include: The dataset module, multi-omics feature module, single-omics prognostic model module, multi-omics fusion model module and data analysis module are connected in sequence; The data set module is used to obtain a data set; the data set includes an IDH wild-type glioblastoma patient data set; The multi-omics feature module is used to obtain pathological images, gene data and transcription data respectively according to the data set; and extracting omics features from pathological images, gene data and transcriptional data respectively; the omics features include: pathological omics features, mutation omics features, copy number variation omics features and transcriptomics features; The single-omics prognostic model module is used to construct a single-omics prognostic model corresponding to the omics features according to the omics features; the single-omics prognostic model includes: a pathology omics prognostic model, a mutation omics prognostic model, a copy number variation omics prognostic model and a transcriptomics prognostic model; The multi-omics fusion model module is used to screen the omics features of the single-omics prognostic model to obtain multi-omics fusion features, and based on the multi-omics fusion features, construct a multi-omics fusion model based on machine learning; based on the multi-omics fusion model, the patients are risk-stratified and divided into a high-risk group and a low-risk group; The data analysis module is used to perform differential analysis between high-risk and low-risk groups, obtain cross-pathways using weighted gene co-expression network analysis and gene set enrichment analysis methods, and determine the key biological pathways and biological behaviors that affect prognosis in a multi-omics fusion model; Based on the omics features, single-omics prognostic models corresponding to the omics features are constructed respectively, including: Based on the key features obtained through feature screening based on pathological omics features, mutation omics features, copy number variation omics features and transcriptomics features, each is combined into a single omics prognostic model corresponding to the omics features; The coefficients of ridge regression were used to weight the key features to form a linear combination model. Based on the linear combination model, the risk scores of each group were calculated, namely Path Score, Mut Score, CNV Score and FPKM Score. The omics features of the single-omics prognostic model were screened to obtain multi-omics fusion features, and based on the multi-omics fusion features, a multi-omics fusion model based on machine learning was constructed, including: The cutoff value of the risk score of each group was determined in the training set of the data set, and the patients were divided into a low-risk group and a high-risk group; the relationship between each group characteristic and survival time was evaluated by survival analysis curve analysis, and the log-rank test method was used to evaluate the survival difference between the two groups; The cutoff value is applied to the internal validation data set, and the cutoff value of the external validation set is calculated based on the statistical characteristics of the external validation set. After the two groups are divided into a high-risk group and a low-risk group, a survival analysis curve and a log-rank test are drawn; Univariate Cox analysis was performed on clinical pathological risk factors and risk scores of each group to obtain factors that were significant for patient prognosis; all factors with P < 0.05 were included in multivariate Cox analysis to obtain independent risk factors, and the risk scores of each group were evaluated as independent risk factors; Patients with four-omics feature data were screened out for constructing a multi-omics fusion model. The risk scores of each omics after verification were used as new features, and the coefficients of each omics risk score were calculated using ridge regression. After weighting, the fused linear combination model was obtained, and the fused risk score was calculated, recorded as Multscore.

Images