CephaloNet: A Deep Learning based automatic landmark detection system for cephalometric X-ray images

Abstract

Abstract Landmark detection plays a vital role in describing the anatomy of the particular patient and provides a quantitative estimation of pathologies in the skull base and jaw regions. A common method used for landmarking is manually pointing the landmarks by experts, which proved to be a laborious and time consuming undertaking. Cephalometric analysis has key importance in clinical diagnosis, surgery and analysis. The advent of artificial intelligence and machine learning has made it possible to automate this process thus saving both time and labour in addition to improving accuracy and precision. In this paper, a novel artificially intelligent cephalometric landmark identification system for automatic cephalometric analysis has been proposed. The proposed model named, CephaloNet, employs a novel Convolution Neural Network (CNN) framework for detecting 19 anatomical landmarks. The model was trained and tested on the benchmark database of 400 lateral cephalograms. For effective training of the model, various preprocessing techniques were employed in order to increase the variability and number of images in the dataset. The experimental results depict the fair performance of the proposed method with 97.89% landmarks identified within 5 mm precision, 93.84% in 4 mm precision, 82.36% in 3 mm precision and 56.99% in 2 mm precision.