Automatic localization of landmarks for cephalometry using BP neural network

Objectives:To design a BP neural network model to analyze and localize landmarks in the cephalogram automatically. Methods:50 normal occlusion cephalogram images were selected, which were divided into testing group and experimental group. The testing group included 30 images, which were used to estimate and test a BP center network model.In experimental group, other 20 images were analyzed and experimented by the model.The testing group images were enhanced by morphologic method, and then recognized and segmented into several sub-images, which were applied to test the BP center network model, then the 20 experimental images were analyzed and landmarks were localized automatically by BP center network model. At the same time, the experimental images were manually recognized and 16 landmarks were localized by 3 orthodontist. Difference of all the data was compared. Results:The BP center network model accurately identified the 16 landmarks of the experimental images. The data of the automatic and the manual were not statistically different. Conclusions:The BP center network model could automatically analyze and identify the cephalogram quickly and accurately.