Descriptive image feature for object detection in medical images

Abstract

Algorithms based on the local description of interest regions are well adapted to the task of detecting and matching equivalent points between two images. Classical descriptors such as SIFT or SURF are efficient when applied to regular images with rich information. When it comes to medical images, these algorithms are not longer applicable without adaptation. For this reason, we propose in this paper a feature-based framework for the detection of objects in medical images with poor information (e.g. X-Ray images). Our approach is based on a modified version of SURF. In order to illustrate our purpose, we apply our framework to the cervical vertebra detection on X-Ray images. The results show that this modified descriptor is an efficient solution in the medical domain. It allows to properly process the vertebra detection in better computing times than other classical descriptors. © 2012 Springer-Verlag.