Segmentation of cervical and lumbar vertebrae in x-ray images using active appearance models and extensions

This thesis presents a hierarchical segmentation algorithm tailored to the segmentation of cervical and lumbar vertebrae in digitized X-ray images. The algorithm employs the Generalized Hough Transform (GHT) to obtain a suitable initialization for two segmentation stages that utilize Active Appearance Models (AAMs) that were proposed by Cootes et al. The advantage of using AAMs in medical image segmentation applications is that rather than creating models that are purely data driven, AAMs gain a priori knowledge through a thorough observation of the shape and texture variation across a training set. This thesis presents a detailed summary of the theory behind AAM along with proposed extensions and customizations of AAM. The proposed extensions (1) address the shortcomings of using the basic texture alignment procedures when using Neighborhood AAMs, (2) automate the selection of training parameters, and (3) modify the AAM search criterion to encourage the location of the edges of the vertebrae. In addition, AAM is utilized to rank the quality of multiple initializations provided by GHT. The proposed segmentation algorithm was tested on 273 cervical X-ray images and 262 lumbar images. If a successful segmentation is defined as a case in which the point-to-corresponding-point error is less than ten pixels for cervical images and twenty-five pixels for lumbar images, results from the proposed segmentation algorithm indicate a 65% success rate for segmentation of cervical vertebrae and a 68% success rate for lumbar vertebrae.