Radiogenomics Of Breast Cancer Based On Cell And Image Phenotype Network Analysis

Breast cancer is a malignant tumor that seriously threatens women’s health,and its high heterogeneity is the fundamental reason that limits the development of individualized diagnosis and treatment.Achieving accurate and effective differentiation of patients is of great significance for the precise treatment of breast cancer.It is known that the tumor microenvironment is the survival place of tumor cells,which complements the development of tumor cells and is a key factor affecting the occurrence and development of breast cancer.Therefore,the overall heterogeneous expression of various types of cells in the microenvironment and its impact on the prognosis of breast cancer need to be studied.In addition,molecular typing is an important clinical indicator for judging the prognosis of breast cancer and making treatment decisions.Therefore,based on the data of cell subsets in the tumor microenvironment,this paper will start from the analysis of the prognostic differences of molecular typing,and find effective cell molecular markers,which will provide reference value for clinical treatment decisions.At present,breast cancer-related research based on cell expression in the microenvironment mainly focuses on a certain type of cell or similar cells to explore its impact on the development and prognosis of breast cancer.The impact of multiple cell subsets that interact in the environment on the prognosis of breast cancer,and the multiple cell subsets that constitute the tumor microenvironment and the relationship between them can be modeled and described by complex network methods.On this basis,this study adopts a radiogenomics method that combines the interpretability of genomics and the non-invasive advantages of radiomics,and innovatively builds a prognosis prediction model by establishing the mapping relationship between the imaging phenotype network and the cellular network.Microenvironmental characteristics and imaging markers were correlated,and the prognostic value of the markers was verified by survival analysis.The main research contents of this paper are as follows:(1)Cell data decomposition and network construction: The genetic data is decomposed into 64 cell type data using the x Cell algorithm that describes cell heterogeneity,and the image and cell data sets are feature dimensionality reduction through consistent clustering and clustering centers.Finally,survival analysis was used to compare the prognostic differences between different molecular types,and the Luminal A and non-Luminal A cells and imaging phenotype networks were constructed.(2)Network alignment analysis of cells and image phenotypes between different molecular types and their association with prognosis: First,based on the community discovery algorithm,we mine cell or image feature groups with potentially similar functions,and divide the closely connected nodes in the network into communities.In this paper,multiple image subgroups and cell subgroups are divided into the same community.Then,the Graph Alignment algorithm is used to mine mappable feature groups according to the interaction mechanism between similar features(nodes)in different networks.Topology,that is,the matching between nodes.Secondly,by combining the calculation results of community discovery and graph alignment,the matching alignment between communities was obtained,and by analyzing the alignment difference between Luminal A and non-Luminal A networks,the cells with significantly changed internal node mapping and potential association with prognosis were obtained.Image Society.Finally,a Cox proportional hazards regression model was constructed to analyze the association between the found communities and prognosis.(3)Radiogenomics method based on deep learning of cell graph structure and analysis of prognostic imaging markers: First,a network is constructed based on the characteristics of cell subsets that independently affect survival.The transformation of network features,using the graph convolutional neural network algorithm to extract features from the network.Then,the patient signatures with significant survival differences were obtained by spectral clustering,and based on this,a radiogenomics signature was constructed to realize the association between image features and cellular features.Secondly,a prognostic prediction model was constructed according to the image features in the image community to verify the predictive value of the image genomics signature.The validation was carried out on two independent image validation sets and it was found that there was a significant survival difference between the predicted patient categories(P=0.021,P= 0.037).Finally,on another independent image validation set,the prognostic prediction performance of image features was evaluated based on different models,and it was found that the above image features had good prediction performance on both models(AUC=0.749,AUC=0.796).In this paper,the imaging genomics research of breast cancer is carried out through image phenotype network(network directly connected with image features)and cell network analysis,and the images and cell subsets that are associated with the prognosis of breast cancer are found,and the cell subsets are used as prognostic molecular markers By constructing the correlation between the characteristics of cell subsets and the characteristics of imaging subsets,the imaging characteristics with prognostic predictive value were discovered.The experimental results show that these cellular molecular and imaging feature markers have high prognostic value in breast cancer,and will provide valuable reference information for clinical individualized treatment decisions.