Patrik Rogalla MD, Presenter: Nothing to Disclose

Bernice E. Hoppel PhD, Abstract Co-Author: Employee, Toshiba Corporation

Mini Vithal Pakkal MBBS, Abstract Co-Author: Nothing to Disclose

Christin Farrell, Abstract Co-Author: Employee, Toshiba Corporation

Sonja Kandel MD, Abstract Co-Author: Nothing to Disclose

To evaluate model-based iterative reconstruction (IR) using 512 and 1024 image matrix against hybrid iterative reconstruction (AIDR).

Raw-data from 20 randomly selected chest CTs (Toshiba Aquilion1) were reconstructed by using AIDR with 6 different kernels and filters that were optimised for lung imaging. 3 radiologists (18, 6 and 3 years of clinical experience, blinded to the reconstruction method) ranked the images separately according to their overall personal preference (forced ranking, no quality criteria given). The reconstruction technique with the highest median ranking was defined as the optimized reference standard for this study. All datasets were then reconstructed using model-based IR at 4 different regularization parameters with a 512 image matrix and 4 corresponding parameters with a 1024 matrix. All nine images (IR and the radiologist’s AIDR reference standard) were displayed on one screen and the same 3 radiologists (blinded to the reconstruction method) were asked to rank all images according to their preference for lung imaging. Image noise was measured on all reconstructions within air.

The preferred reference hybrid reconstruction techniques (AIDR) was based on FC 81 (high frequency kernel); the median/mode and mean SD of image noise in sequential order for IR at 512 matrix with b=300,400,500,700, for AIDR, and for IR at 1024 matrix with b=1600,2400,3200,4000 were 7/7 and 17.7, 5/6 and 15.7, 4/4 and 13.4, 6/5 and 11.3, 9/9 and 14.5, 2/2 and 17.4, 1/1 and 15.6, 3/3 and 13.5, 8/8 and 11.1, respectively. The difference between median rank 1 and 5, and 5 and 9 were statistically significant (both p<0.0001). With the exception of one case for one reader, AIDR always ranked the worst, and with the exception of b=4000, 1024 matrix was always preferred over 512. SD was the highest on 512 IR b=300 and lowest on 1024 IR b=4000 (p<0.0001).

1024 matrix model-based IR improved image quality compared to both 512 model based IR and AIDR. After further optimization of reconstruction parameters tailored to the specifics of lung imaging, model-based IR with 1024 image matrix may become the reconstruction method of choice.

Image quality in lung imaging can be improved by 1024 matrix model-based iterative reconstruction without modification of CT scanning parameters.

Rogalla, P, Hoppel, B, Pakkal, M, Farrell, C, Kandel, S, 1024 Matrix Model-based Iterative Reconstruction Improves Clinical Image Quality in Lung Imaging.  Radiological Society of North America 2014 Scientific Assembly and Annual Meeting, - ,Chicago IL. http://archive.rsna.org/2014/14016725.html