Gender Estimation from Panoramic Dental X-ray Images using Deep Convolutional Networks

Abstract

Current techniques for gender estimation from X-ray images, except being time-consuming, require a highly experienced expert to perform the process. Deep convolutional neural networks have shown to be a very successful technique in many computer vision tasks, mainly because of high accuracy, stability, and processing speed. In this paper, we propose a new method to the gender estimation from panoramic dental X-ray images based on analysis of images using deep convolutional neural networks trained to perform a binary classification. Detailed insight is provided into architecture, hyperparameters and training procedure of our best performing model obtaining an accuracy of 94.3% on a test set. Further experiments have been performed to get a better understanding of anatomical structures which carry the most important information for gender estimation. The presented method requires no special knowledge or equipment to be used, and besides high accuracy, it is also extremely fast with only 18ms of processing time per image on a dedicated GPU.