Osteoporosis risk prediction method based on relational network and GNN

Abstract

Existing methods for osteoporosis risk prediction are mainly implemented by machine learning models based on individual features, which have the problems of redundant features and poor performance. A risk prediction method for osteoporosis based on relational network and graph neural network (GNN) is proposed. Individuals are taken as the relational network nodes, and related medical history and eating habits are taken as the edges of the network to connect nodes, to construct a graph-structured relational network. Individual features are used as node features to construct a prediction model based on GNN while traditional machine learning models based on the feature set are constructed as baseline models for comparative analysis. The experimental results show that the GAT-based prediction model outperforms the baseline model, with accuracy, sensitivity, specificity and AUC of 89.5%, 83.4%, 94.0% and 0.93, respectively. the GCN-based prediction model performs close to the baseline models. The osteoporosis prediction method based on relational network and GNN is feasible. Compared with traditional machine learning methods, it has a simpler process and better prediction ability.