Graph Convolutional Networks Based Multi-modal Data Integration for Breast Cancer Survival Prediction

Recently, multi-modal breast cancer survival prediction (MBCSP) has been widely researched and made huge progress. However, most existing MBCSP methods usually overlook the structural information among patients. While certain studies may address structural information, they often ignore the abundant semantic information within multi-modal data, despite its significant impact on the efficacy of cancer survival prediction. Herein, we propose a novel method for breast cancer survival prediction, termed graph convolutional networks based multi-modal data integration for breast cancer survival prediction (GMBS). In essence, GMBS firstly defines a series multi-modal fusion module to integrate diverse patient data modalities, yielding robust initial embeddings. Subsequently, GMBS introduces a patient-patient graph construction module, aiming to delineate inter-patient relationships effectively. Lastly, GMBS incorporates a Graph Convolutional Network framework to harness the intricate structural information encoded within the constructed graph. Extensive experiments on two well-known MBCSP datasets demonstrate the superior performance of GMBS method compared to representative baseline methods.