Automatic identification and localization of craniofacial landmarks using multi layer neural network

Abstract

Cephalometric evaluation of lateral x-rays of the skull, used mainly by orthodontists, is usually carried-out manually to locate certain craniofacial landmarks. This process is time consuming, which is both tedious and subject to human error. In this paper we propose a novel algorithm based on the use of the Multi-layer Perceptron (MLP) to locate landmarks on the digitized x-ray of the skull. The main feature of this proposed algorithm is that its performance is independent of the quality of radiographs. Preprocessing techniques are used to enhance the quality of the image and to extract the outer edges of the skull. Four points are selected to form the basis for additional features representing the size, rotation and offset of the skull. The extracted features are then used as inputs to the MLP. The corresponding outputs represent the horizontal and vertical coordinates of the selected landmark. MLP's are efficient function approximators and in this work are trained to locate landmarks by using a number of manually labeled data as a training set. After training, the MLP is used to locate landmarks on target digitized images of radiographs. The MLP is trained using 55 manually labeled images and tested on a separate set consisting of 134 images, which are not used for training. Results obtained show an improvement over template-matching and line-following techniques. This is apparently evident when the search encounters a lost tooth, cavity filling or when the image is of a low quality. © Springer-Verlag Berlin Heidelberg 2003.