Ke Li PhD, Presenter: Nothing to Disclose

John W. Garrett MS, Abstract Co-Author: Nothing to Disclose

Yongshuai Ge, Abstract Co-Author: Nothing to Disclose

Guang-Hong Chen PhD, Abstract Co-Author: Research funded, General Electric Company Research funded, Siemens AG Research funded, Varian Medical Systems, Inc Research funded, Hologic, Inc

The purpose of this study was to assess the task- and dose-dependence of the spatial resolution of statistical model based iterative reconstruction (MBIR) along both axial and z directions and to examine the impact of spatial resolution on the model observer performance. 

The thoracic section of an anthropomorphic phantom was scanned using a 64-slice clinical CT system (CT750 HD, GE Healthcare) at four dose levels (CTDIvol=4, 8, 12, 16 mGy). Both FBP and MBIR (Veo, GE Healthcare) were used for image reconstruction. Nine objects embedded in the phantom with contrast values ranging from 13 to 1710 HU were used to assess spatial resolution. The axial and z resolutions were quantified locally in the image domain by point spread functions (PSF) and slice sensitivity profiles (SSP) respectively. All the scans were repeated by 100 times and averaged, which enabled spatial resolution of ultra-low contrast objects to be accurately measured. The repeated scans also enabled assessment of local noise standard deviation and the detectabilities index (d') of a channelized Hoteling (CHO) model observer.

The axial resolution of MBIR improved monotonically with increasing dose and contrast level (FWHM_PSF=2.0 mm at 13 HU/25% dose and was 0.8 mm 1710 HU/100% dose). In comparison, axial resolution of FBP is independent of dose and contrast (FWHM_PSF=1.2 mm). The z resolution of MBIR demonstrated similar contrast dependence but only negligible dose dependence. The spatial resolution of MBIR and FBP became equivalent at some transitional contrast levels (280 HU for 25% dose and 90 HU for 100% dose), above which MBIR demonstrated superior resolution than FBP (and vice versa). Spatial resolution and noise assessed at the same location demonstrated a strong tradeoff in MBIR, and CHO detectability index d' was improved by [48%, 115%] with MBIR depending on the contrast and dose level.

MBIR produces images with unique spatial resolution characteristics and introduces new challenges to its clinical use and evaluation. One potential solution, as suggested by this work, is to perform rigorous spatial resolution and noise measurements at different dose levels for each specific task.

The results provide new guidance for the optimal prescription of scanning parameters when a MBIR method is used in clinical routine CT exams for dose reduction and/or diagnostic performance improvement.

Li, K, Garrett, J, Ge, Y, Chen, G, Experimental Assessment of Z- and Axial Spatial Resolution of Statistical Model Based Iterative Reconstruction and Its Correlation with Image Noise.  Radiological Society of North America 2014 Scientific Assembly and Annual Meeting, - ,Chicago IL. http://archive.rsna.org/2014/14019521.html