Options
2010
Conference Paper
Titel
Automated initialization and region of interest detection for successful head registration of truncated CT/MR head & neck images
Abstract
Maximization of a voxel based similarity metric like mutual information is the state of the art for the multimodal rigid registration of the head. To achieve satisfactory results the transform needs to be initialized properly and the region of interest (ROI) containing only rigid structures has to be defined in both images. In this paper we present and comprehensively evaluate an automated initialization and ROI detection method that enables the fully automated rigid registration of routinely gathered head CT and MR images. For ROI definition an automated head detection heuristic is presented that is robust against truncation and can be applied to images of different modalities. The registration transform is initialized based on a novel automated landmark detection method. Combined with standard mutual information based registration techniques, this result in a fully automated rigid registration framework for truncated CT/MR head & neck images. We evaluated the landmark detection in 81 routinely gathered truncated MR and CT images of the head & neck. Those images form 57 intra subject CT/MR pairs to which the overall approach was applied and evaluated by visual inspection. We show that the presented approach is robust against truncation and influence of non rigid structures. Not requiring any user interaction at all, it enables new applications in clinical practice.
Author(s)