Graph Neural Networks for Anomaly Detection in Internet of Medical Things

Recent advancements in the realm of Graph Neural Networks (GNN) have unveiled compelling new prospects for classifying and detecting anomalies within medical datasets. Within this study, we delve into the efficacy of GNNs by leveraging diverse graphical representations, such as tree graphs and bipartite graphs, to capture and exploit intricate relationships inherent in medical data. Through this graph conversion process, we have observed that GNNs gain the ability to analyze and train on a constrained set of configurations, thus enhancing both efficiency and classification performance. Moreover, by utilizing complete graphs and multiparametric temporal graphs, we are able to discern anomalies within health values through the utilization of temporal and multiparametric correlation factors. This sequential approach of reapplying GNNs to graphs devoid of erroneous values enables more precise classification, achieving an exceptional accuracy rate. These findings underscore the vital importance of graphs in medical data analysis while showcasing the efficacy and robustness of GNNs in this particular domain.