Segmentation of scoliotic spine silhouettes from enhanced biplanar X-rays using a prior knowledge Bayesian framework

In this paper, we propose a novel segmentation method which takes into account the variable appearance and geometry of a scoliotic spine (rotation, wedging) from X-ray images of poor quality in order to automatically isolate and extract the silhouettes of the anterior spinal body. An adaptive non-linear enhancement filter is first presented to enhance bone structures and reduce image noise. By incorporating prior anatomical information through a Bayesian formulation of the morphological distribution, a multiscale spine segmentation framework is then proposed for scoliotic patients. The likelihood of the model is computed based on an automatic learning process derived from labeled training data, while the Hessian image matrix is exploited to create an image-response map by attributing at each pixel the likeliness presence of a structure of interest. A qualitative evaluation of the vertebral contour segmentations obtained from the proposed method gave promising results while the quantitative comparison to manual identification yields an accuracy of 1.5 plusmn 0.6 mm based on the localization of the spine boundaries by a radiology expert.