Carlos Raul Castro-Pareja PhD, Abstract Co-Author: Nothing to Disclose

Raj Shekhar PhD, Presenter: Nothing to Disclose

Barry David Daly, Abstract Co-Author: Nothing to Disclose

To reduce the diagnosis time in virtual colonoscopy by simplifying localization of corresponding features in prone and supine images through automatic co-registration using an elastic transformation.

The algorithm was tested on virtual colonoscopy datasets with intra-slice resolutions of 0.67 mm and 1.0-1.6 mm slice thickness. The local deformation field was modeled using B-splines controlled by a rectangular grid of control points. The algorithm estimates the local deformation field by optimizing the values of the deformation field at each control point within the grid. The algorithm used a multi-resolution approach. Rigid registration, which determines the global transformation that best matches the images, was performed prior to the estimation of local deformations. The results were evaluated by comparing three different measures: the average root mean square errors per voxel between the prone images and the supine images before and after registration, the percentage of misregistered bone volume and the percentage of misregistered air volume. The latter two were calculated by using thresholding to segment the regions corresponding to bone and air in the images and counting the number of coinciding voxels between the prone and the transformed supine images.

The average root mean square error per voxel decreased on average by 13% after rigid registration and by 34% after elastic registration. The proportion of registered bone tissue to misregistered bone tissue improved by 173% after rigid registration and by 190% after elastic. For air the numbers were 150% and 302%, respectively. The latter numbers are especially significant because they represent the better correlation between the colon in the prone and the transformed supine image.

Using automated elastic registration improves the accuracy of cross-localization of features in the prone and supine images, which can potentially reduce the time required to diagnose cases, thereby helping to bring virtual colonoscopy into the mainstream as a non-invasive colorectal cancer diagnosis tool.

Castro-Pareja, C, Shekhar, R, Daly, B, Elastic Registration of Prone and Supine CT Datasets for Virtual Colonoscopy.  Radiological Society of North America 2005 Scientific Assembly and Annual Meeting, November 27 - December 2, 2005 ,Chicago IL. http://archive.rsna.org/2005/4418709.html