Abstract
Objective: The objective of this study is to develop a convolutional neural network (CNN) for the automatic detection of soft and hard tissue landmarks and the classification of lip thickness on lateral cephalometric radiographs.
Methods: A dataset of 1019 pre-orthodontic lateral cephalograms from patients with diverse malocclusions was utilized. A CNN-based model was trained to automatically detect 22 cephalometric landmarks. Upper and lower lip thicknesses were measured using some of these landmarks, and a pre-trained decision tree model was employed to classify lip thickness into the thin, normal, and thick categories.
Results: The mean radial error (MRE) for detecting 22 landmarks was 0.97 ± 0.52 mm. Successful detection rates (SDRs) at threshold distances of 1.00, 1.50, 2.00, 2.50, 3.00, and 4.00 mm were 72.26%, 89.59%, 95.41%, 97.66%, 98.98%, and 99.47%, respectively. For nine soft tissue landmarks, the MRE was 1.08 ± 0.87 mm. Lip thickness classification accuracy was 0.91 ± 0.04 (upper lip) and 0.90 ± 0.04 (lower lip) in females and 0.92 ± 0.03 (upper lip) and 0.88 ± 0.05 (lower lip) in males. The area under the curve (AUC) values for lip thickness were ≥0.97 for all gender-lip combinations.
Conclusions: The CNN-based landmark detection model demonstrated high precision, enabling reliable automatic classification of lip thickness using cephalometric radiographs.