Barbara Flach, Abstract Co-Author: Nothing to Disclose

Marcus Brehm, Abstract Co-Author: Nothing to Disclose

Jan Kuntz, Abstract Co-Author: Nothing to Disclose

Rolf Kueres, Abstract Co-Author: Nothing to Disclose

Soenke Heinrich Bartling MD, Abstract Co-Author: Research support, Siemens AG

Marc Kachelriess PhD, Presenter: Nothing to Disclose

To provide a continuously adaptive prior at high temporal resolution for motion correction in low dose tomographic fluoroscopy.

To guide minimally-invasive interventions a continuous data acquisition is necessary. In low dose tomographic fluoroscopy (3D+time) volumes are reconstructed from only 15 cone-beam projections per 180°. This keeps the patient dose level as low as in today’s C-arm-based projective fluoroscopy (2D+time). The combination of this highly sparse information with a high quality prior allows to continuously provide high quality update volumes for intervention guidance. To account for patient motion a continuous adaptation of the prior during the intervention is required. We developed a deformable volume-to-rawdata (3D-2D) registration that uses not more than the latest 15 projections. Thus, the running prior is continuously updated with a high temporal resolution. The registration calculates the forces for adaptation on the basis of the sum of squared differences in rawdata domain and regularizes the vector field by convolution with Gaussian kernels. Our approach was implemented in an experimental prototype flat detector CT scanner. To validate the new technique we used the head scan of a pig in vivo. During the intervention the position of the pig’s head was moved manually to mimic patient motion. Reconstruction was done with 3D-3D and with the new 3D-2D prior using only 15 projections for registration. The results were assessed by visual inspection. For quantitative evaluation the sum of squared differences (SSD) was calculated in image and rawdata domain.

The resulting running prior images obtained by 3D-2D registration show a higher matching to the ground truth compared to the 3D-3D prior. The SSD of 3D-2D prior in image domain is 88% of the 3D-3D prior. In rawdata domain the matching was improved by 35% compared to the 3D-3D prior. The time frames show less inconsistency which can be attributed to the higher temporal resolution of the 3D-2D prior.

Low dose tomographic fluoroscopy should use the 3D-2D running prior to ensure maximal image quality.

Improved visualization of the correct position of interventional material with respect to the surrounding patient tissue. No additional dose needed in case of patient motion during the intervention.

Flach, B, Brehm, M, Kuntz, J, Kueres, R, Bartling, S, Kachelriess, M, Deformable 3D-2D Registration-based Running Prior for Low Dose Tomographic X-Ray Fluoroscopy.  Radiological Society of North America 2014 Scientific Assembly and Annual Meeting, - ,Chicago IL. http://archive.rsna.org/2014/14004820.html