Comparison of different metrics for appearance-model-based 2D/3D-registration with X-ray images

Abstract

The general idea of the presented work is to overcome known problems with segmentation and analysis of 2D radiographs by registering a 3D appearance-model. Therefore this paper introduces a novel method to register 2D x-rays with 3D appearance-models by optimizing the appearance and pose of the model until a virtual radiograph of the generated model-instance optimally fits the investigated x-ray. The approach was tested on a sample set of 15 human femur specimen using different metrics and optimization techniques to investigate the impact on the resulting implicit 2D-segmentation. The first promising results are presented and discussed in detail.