Research On Breast Cancer Survival Prediction Based On Graph Convolutional Network And Bilinear Pooling

In recent years,the incidence of cancer in the world has been increasing year by year.The incidence of female breast cancer has exceeded that of lung cancer.Breast cancer has become one of the main diseases affecting women’s health.Accurate survival prediction of breast cancer patients is a key issue in the prognosis of breast cancer.Because it can aid physicians make informed decisions and further guide the establishment of palliative treatment.At the same time,the development of gene sequencing technology and digital imaging technology has brought massive multimodal data,such as cancer omics and pathological images,providing a solid data basis for the survival prediction of breast cancer patients.Previous studies have shown that both omics and pathological image data contain information related to breast cancer survival prediction,and there is complementarity and correlation between different types of data.Therefore,effective use of the above multimodal data is of great significance to improve the prediction accuracy of breast cancer survival.However,most existing cancer survival prediction methods directly splicing multimodal data,which cannot fully use multimodal data to train the survival prediction model,resulting in poor prediction performance.In view of the above problems,from the perspective of multimodal data fusion of breast cancer,this paper utilizes graph convolution network and bilinear pooling to study the survival prediction of breast cancer.The main work is as follows:(1)Aiming at the problem that existing algorithms are difficult to fully mine topological structure information among different omics data,this paper proposes a multi-omics data fusion method based on graph convolution network for breast cancer survival prediction(OAFN).Firstly,OAFN constructed an affinity network based on multiple omics data to mine the topology information between and within different omics data.Secondly,with affinity network as the guide,graph convolution network is used to generate fusion features with data correlation.In addition,OAFN constructs multiple deep neural networks for different omics data,and extracts specific features based on single omics data to learn the differences between different data types.The experimental results show that OAFN can obtain better expression of higher-order fusion features and achieve better prediction accuracy.(2)Aiming at the problem of low signal-to-noise ratio(SNR)in multi-omics data,this paper further proposed a breast cancer survival prediction method based on attention mechanism and graph convolution network(MAFN).On the basis of OAFN,MAFN solved the weight of correlation between multiple omics features and task objectives by constructing stacked shallow attention network,and continuously updated the weight according to the optimization objectives during training,so as to enhance the features related to task objectives adaptively.The experimental results show that the weighted feature formed by introducing attention mechanism has stronger discrimination,and can further improve the survival prediction accuracy of breast cancer.(3)In view of the failure of multiple omics data to fully represent the survival factors of cancer patients,a multimodal data fusion method(DNBP)based on bilinear pooling was proposed by combining multiple omics data and histopathological image data to predict the survival of breast cancer patients.First,DNBP extracted image features from histopathological image data by Cell Profiler tools.Secondly,the factorized bilinear pooling was used to effectively fuse various omics features with pathological image features to learn the interaction between different modal data adaptively.The experimental results show that this method can effectively fuse multi-modal data,and improve the survival prediction performance of breast cancer.