구강악안면영역에서 해부학적 요소들의 상관구조에 대한 네트워크 분석

Abstract

The main aim of this study was to find the correlation structure of lateral cephalometric measurements in Korean adults. It helped to group (cluster) related variables in the oromaxillofacial region and quantify how significant each variable was. In particular, it was possible to deal with the details of craniofacial characteristics by considering 65 cephalometric variables based on a well-known cephalometric analysis (eg, Downs et al.), which is about three times the number of variables considered in previous studies. In addition, a simplification process was used that could reveal more complex correlation structures caused by many variables. The subjects were 1,708 Korean adults aged 18-25 years. 753 of them had been previously diagnosed with normal occlusion and others were malocclusion patients. The correlation matrix among 65 cephalometric variables was analyzed using network analysis methodology. Each variable was represented as a node, and the weight of edge was calculated by the absolute value of Pearson correlation coefficient between two variables. First, we adopted the minimum spanning tree (MST) and threshold-based network visualization to simplify the complex network structure. Second, community detection algorithm was used to group highly related variables. In this study, the correlation network provided detailed information of the dependence structure among cephalometric variables. The modular structures of the network refer to the existence of groups with highly correlated features (ex. ANB, A-B plane angle, APDI et al.). We also uncovered 10-cluster structures which describe anatomical features in the oromaxillofacial region. Based on finding central nodes for each group, we found 10 significant variables in cephalometric analysis. Lastly, the comparison of correlation structure between normal occlusion and Class I, II and III groups helped to understand malocclusion. The visualized network structure could allow us to intuitively and effortlessly describe the relationship among cephalometric variables and cluster groups with highly correlated features. These results would also be meaningful for the data-based analysis on normal occlusion and malocclusion.