The registration of preoperative CT to intra-operative reality systems is a crucial step in Computer Assisted Orthopedic Surgery (CAOS). The intra-operative sensors include 3D digitizers, fiducials, X-rays and Ultrasound (US). FAPBED is designed to process CT volumes for registration to tracked US data. Tracked US is advantageous because it is real time, noninvasive, and non-ionizing, but it is also known to have inherent inaccuracies which create the need to develop a framework that is robust to various uncertainties, and can be useful in US-CT registration. Furthermore, conventional registration methods depend on accurate and absolute segmentation. Our proposed probabilistic framework addresses the segmentation-registration duality, wherein exact segmentation is not a prerequisite to achieve accurate registration. In this paper, we develop a method for fast and automatic probabilistic bone surface (edge) detection in CT images. Various features that influence the likelihood of the surface at each spatial coordinate are combined using a simple probabilistic framework, which strikes a fair balance between a high-level understanding of features in an image and the low-level number crunching of standard image processing techniques. The algorithm evaluates different features for detecting the probability of a bone surface at each voxel, and compounds the results of these methods to yield a final, low-noise, probability map of bone surfaces in the volume. Such a probability map can then be used in conjunction with a similar map from tracked intra-operative US to achieve accurate registration. Eight sample pelvic CT scans were used to extract feature parameters and validate the final probability maps. An un-optimized fully automatic Matlab code runs in five minutes per CT volume on average, and was validated by comparison against hand-segmented gold standards. The mean probability assigned to nonzero surface points was 0.8, while nonzero non-surface points had a mean value of 0.38 indicating clear identification of surface points on average. The segmentation was also sufficiently crisp, with a full width at half maximum (FWHM) value of 1.51 voxels.
[1]
R. Martin,et al.
Three-Dimensional Spatial Compounding of Ultrasound Scans with Weighting by Incidence Angle
,
2000,
Ultrasonic imaging.
[2]
Jacques Demongeot,et al.
Automated Registration of Ultrasound with CT Images: Application to Computer Assisted Prostate Radiotherapy and Orthopedics
,
1999,
MICCAI.
[3]
S. Winter,et al.
Registration of 3D CT and ultrasound datasets of the spine using bone structures.
,
2002,
Computer aided surgery : official journal of the International Society for Computer Aided Surgery.
[4]
Paul J. Besl,et al.
A Method for Registration of 3-D Shapes
,
1992,
IEEE Trans. Pattern Anal. Mach. Intell..
[5]
Stefan Wesarg,et al.
Validation of 3D ultrasound: CT registration of prostate images
,
2003,
SPIE Medical Imaging.
[6]
Dinesh K. Pai,et al.
Direct surface extraction from 3D freehand ultrasound images
,
2002,
IEEE Visualization, 2002. VIS 2002..
[7]
Russell H. Taylor,et al.
Understanding bone responses in B-mode ultrasound images and automatic bone surface extraction using a Bayesian probabilistic framework
,
2004,
SPIE Medical Imaging.
[8]
Martin Styner,et al.
Fast Automatic Bone Contour Extraction in Ultrasound Images
,
2002
.