VGG-16 based Deep Learning Approach for Cephalometric Landmark Detection

Abstract

Aims: The aim of this research work is to compare the accuracy and precision of manual landmark identification Background: Cephalometric landmark detection is a critical task in orthodontics and maxillofacial surgery and accurate identification of landmarks is essential for treatment planning and precise diagnosis outcomes. It entails locating particular anatomical landmarks on lateral cephalometric radiographs of the skull that can be utilised to evaluate the relationships between the skeleton and the teeth as well as the soft tissue profiles. Many software tools and landmark identification approaches have been implemented over time to increase the precision and dependability of cephalometric analysis. Objective: The primary objective of this research is to evaluate the effectiveness of an automated deep learning-based VGG-16 algorithm for cephalometric landmark detection and to compare its performance against traditional manual identification methods in terms of accuracy and precision. Methods: The study employs a VGG16 transfer learning model on a dataset of skull X-ray images from the IEEE 2015 ISBI Challenge to automatically identify 19 cephalometric landmarks on lateral cephalometric radiographs. The model is fine-tuned to predict the precise XY coordinates of these landmarks enhancing the accuracy of cephalometric analysis by minimizing manual intervention and improving detection consistency. Results: The experimental findings indicate that the presented cephalometric landmark detection system has attained Successful Detection Rates (SDR) of 26.84%, 41.57%, 59.89% and 94.42% in the 2, 2.5, 3 and 4mm precision range respectively and a Mean Radial Error (MRE) of 2.67mm. Conclusion: This paper has presented an approach for cephalometric landmark detection using the VGG-16 model a widely used deep learning architecture in computer vision. Through the experiments it is shown that the VGG-16 model can achieve state-of-the-art performance on the task of cephalometric landmark detection. The results have demonstrated that the VGG-16 model can automatically extract relevant features from cephalometric images allowing it to accurately detect anatomical landmarks. It is also shown that fine-tuning the pre-trained VGG-16 model on cephalometric data can improve its performance on this task. The suggested technique may enhance the effectiveness and precision of cephalometric landmark detection and facilitate clinical decision-making in orthodontics and maxillofacial surgery.