Accurate 3D data stitching in circular cone-beam micro-CT

In circular cone-beam microcomputed tomography (micro-CT), it is likely that the length of the Field of View (FOV) of a single acquisition is shorter than the total length of the object to be imaged, such as a rat in the case of preclinical application of micro-CT. This leads to multiple acquisitions using different bed positions with bed translations in between, which can be automated using a motorized bed stage. However, subtle mechanical inaccuracies can cause undesired effects when the reconstructed volumes of the different acquisitions are combined into one larger volume. In this paper, we develop an automated method for accurately stitching 3D computed tomography (CT) data using an image registration scheme, and validate this technique in a circular cone-beam micro-CT scanner.

The approach is based on precalculated spatial transformation matrices acquired by a calibration phantom with point markers at stitching positions. The spatial transformation between two adjacent subvolumes was calculated only once with a rigid-body matching algorithm. Once all transformation matrices are obtained, all subsequent reconstructed subvolumes imaged at these fixed positions can be stitched accurately, and efficiently using these precalculated matrices.

We applied this method to real object/animal imaging in circular cone-beam micro-CT and compared the result with that obtained by stitching method calculated only by translation distances and CT voxel size. Both stitching errors calculated using point markers and stitched volumes of rigid object (a syringe) and small animal (a rat) illustrated the success of our proposed approach.

Preliminary experimental results demonstrate that "3D data stitching" using an image registration scheme provides a good solution to the voxel mismatch caused by limited FOV length in circular cone-beam micro-CT. This method can be extended to other tomography techniques which need to acquire data at fixed scanning positions.