A comparison of four algorithms for estimating 3-D rigid transformations
BMVC '95 Proceedings of the 1995 British conference on Machine vision (Vol. 1)
| Hi-index | 0.00 |
Spatially registering SPECT with CT makes it possible to anatomically localize SPECT tracers. In this study, an accurate method for the coregistration of ultra-high-resolution SPECT volumes and multiple cone-beam CT volumes is developed and validated, which does not require markers during animal scanning. Methods. Transferable animal beds were developed with an accurate mounting interface. Simple calibration phantoms make it possible to obtain both the spatial transformation matrix for stitching multiple CT scans of different parts of the animal and to register SPECT and CT. The spatial transformation for image coregistration is calculated once using Horn's matching algorithm. Animal images can then be coregistered without using markers. Results. For mouse-sized objects, average coregistration errors between SPECT and CT in X, Y, and Z directions are within 0.04 mm, 0.10 mm, and 0.19 mm, respectively. For rat-sized objects, these numbers are 0.22 mm, 0.14 mm, and 0.28 mm. Average 3D coregistration errors were within 0.24mm and 0.42mm for mouse and rat imaging, respectively. Conclusion. Extending the field-of-view of cone-beam CT by stitching is improved by prior registration of the CT volumes. The accuracy of registration between SPECT and CT is typically better than the image resolution of current ultra-high-resolution SPECT.