A novel coarse-to-fine computational method for three-dimensional landmark detection to perform hard-tissue cephalometric analysis

Cephalometric analysis has an important and essential role to treat the patients with craniofacial and dentofacial deformities. Cephalometric analysis is a relationship of human geometry which can be quantified and derived from the linear and angular measurements. To treat any patient, such analysis is required to be performed on the Head X-ray image of the patient. The objective of the proposed work is to detect cephalometric landmarks automatically on CT (computational tomography) images. Twenty cephalometric landmarks were automatically localized on 100 CT scans using hybrid coarse-to-fine computational method. The mean error for landmark detection was computed as 2.88 mm and standard deviation of 1.85 mm. The highest detection rate for cephalometric landmarks was received as 100% for Nasion landmark under 4-mm error and the highest detection rate was received as 99% for Nasion landmark under 3-mm error. The less number of datasets were used for the training and higher number of datasets were used for the testing. Compared to the literature methods, our method used higher number of datasets to demonstrate the accuracy of the proposed method.