An AI-Driven Hybrid Framework for Cephalometric Landmark Detection via Metaheuristic-Optimized Heatmap CNN and Stationary Wavelet Transform
Year: 2026
Type: conference-paper
Abstract: High-precision detection of cephalometric skeletal landmarks (CSLs) enables accurate analysis of skeletal, dental, and soft-tissue relationships, which is essential for orthodontic diagnosis and treatment planning. In this study, regional inference using YOLOv11 was employed to automatically detect 17 CSLs on lateral cephalograms, and a heatmap-based ViTPose model was developed for each landmark. To enhance localization performance, stationary wavelet transform and hyperparameter optimization of the ViTPose model were used. The results show that the proposed approach improves CSL localization performance, reducing the mean radial error by approximately 19.75% and increasing the success detection rate by about 5% across all error thresholds. This highlights the suitability of the proposed model for cephalometric analysis.
Authors: Yusuf Bahri Özçelik, Aytaç Altan, Orhan Çiçek
Institution: Zonguldak Bülent Ecevit University
Language: English
Subfield: Oral Surgery
Field: Dentistry
Domain: Health Sciences
Open Access status: hybrid
OpenAlex ID: W7202201582