A novel strategy for automatic localization of cephalometric landmarks

Cephalometric analysis has an important role in diagnosis and treatment planning for malocclusions. Most analysis steps are computable and underlying structure may be generated provided landmarks are correctly localized. Due to the complexity of human anatomy sensed in a cephalometric x-ray, the landmarks are localized by human experts. In the last few years, efforts have been made to automate this process. This work contributes into a novel approach by first making a generic template search to make a coarse localization of the landmarks. A finer search further combines the best suitable edge or region based final localization. In both the steps of searching, small search area is allocated for each landmark. Films were digitized at three hundred dpi(dots per inch) and the experiments were conducted by recording fourteen landmarks on a number of patients by three dentists. Ground truth of the coordinates was obtained by averaging and the results of the algorithm were compared with these reference values. The result indicates that most of the landmarks lie in the specified limits. In this manner, both speed and accuracy is achieved with performance comparable to humans.