Automatic Estimation of Cephalometric Landmarks for Lateral Skull X-Rays using CNN

Calderon-Sastre Enrique, Lopez-Ramirez Jose Luis, Beltran-Hernandez Jose Guadalupe, Ibarra-Manzano Mario Alberto, Ruiz-Pinales Jose

TL;DR: A CNN implementation to estimate 19 Cephalometric Landmarks in X-ray images

Keywords: Landmarks, CNN, Lateral Skull X-rays

Abstract:

Automatic detection of landmarks in lateral skull cephalograms is considered as a Computer-Aided Diagnostics (CAD) tool. It helps physicians to detect pathologies such as craniofacial growth, orthodontic diagnosis, and oral-maxillofacial treatment planning.

This study aims to find 19 landmarks out of 300 images (previously labeled by experts), which form the public dataset proposed by the 2014 IEEE International Symposium on Biomedical Imaging (ISBI-2014). Deep Convolutional Neural Networks (CNN) were used as a direct regression framework for the probabilistic estimation of the 19 landmarks.The obtained outcomes show up to 1.2 mm of average accuracy per image. The results were calculated between the 19 landmarks proposed by the expert and the estimated points.