Automatic Recognition of Anatomic Features on Cephalograms

Objectives:The purpose of this study was to develop a computer-assisted system for automatic recognition of cephalometric landmarks. Methods: A total of 465 lateral cephalogram images of Japanese orthodontic patients (mean age: 22Y6M) was employed. Twenty cephalometric landmarks were located on each image visually by an expert orthodontist. An image area including each landmark and its adjacent anatomic structure was transformed to a PPED vector (Yagi et al. (2002)) for 400 cephalogram images. Fifteen template vectors were generated from the 400 PPED vectors for each landmark using vector quantization (VQ) method. To evaluate the system recognition performance, the position of each landmark identified by the system on the remaining 65 cephalogram images employing the template matching operations was compared with the norm provided in a form of confidence ellipse (Tanikawa et al.(2003)). When the system-identified point was located within a confidence limit of 99%, the result of the landmark identification was assigned to be successful. Results: The rate of success for each landmark was calculated. The results showed mean success rate of 88% (77-100%). Conclusions: we concluded that the automatic cephalometric landmark recognition system was found to be reliable.