SIFT-based motion registration for sequence images of instant CT

In order to achieve in vivo high resolution imaging, our group tried to develop a new CT system named instant CT which uses more than a dozen couples of x-ray sources and detectors. It scans the target region-of-interest (ROI) dozens of times. At each position, all of the x-ray sources are fast exposed synchronously. Then, the gantry rotates a little angle to prepare for the next exposure until the rotation covers the angle of two X-ray sources. Since each exposure is very fast (about 0.5-5ms) the organ motions can be greatly reduced. Instant CT image reconstruction includes three steps: 1) few-view CT reconstruction from one simultaneous exposure data of all the x-ray sources; 2) motion registration among these sequence images from different exposures; 3) motion correction based image reconstruction with all the data to obtain the final 4-D images. This paper focused on the above second problem. We proposed a SIFT-based (scale invariant feature transforms) motion registration algorithm for these sequence instant CT images at different time phases. With the global rigid motion model, the motion parameters were calculated from the registration information. Then, the motions inside the CT images at different phases were reduced, and the final instant CT image was obtained by fusing all of the different exposure images which were removed motions. Numerical experiments validated the efficiency of our algorithm which can greatly mitigate the blurs and deformation caused by the object motion during the CT scanning.