RSNA 2012 

Abstract Archives of the RSNA, 2012


SSK15-04

Quantitative Image Quality Evaluation of Low Dose Iterative Reconstruction

Scientific Formal (Paper) Presentations

Presented on November 28, 2012
Presented as part of SSK15: Physics (CT Imaging/Phantoms)

Participants

David L. Wilson PhD, Presenter: Owner, BioInVision Inc Research Grant, Koninklijke Philips Electronics NV
Jun L. Miao MENG, Abstract Co-Author: Nothing to Disclose
Nilgoun Raihani, Abstract Co-Author: Nothing to Disclose
Kevin M. Brown MS, Abstract Co-Author: Employee, Koninklijke Philips Electronics NV

CONCLUSION

Detection of low contrast disks predicts improvements with IMR as compared to FBP. The CHO model reasonably predicts results over this range of variables. It remains to be seen if this will hold over a broader range of independent variables. It will be important to determine if these laboratory studies correlate well with clinical image quality evaluations.

BACKGROUND

There is a need to develop quantitative image quality assessments of emerging, iterative image reconstruction techniques, enabling one to optimize and compare algorithms at low x-ray dose. In this study, we compare a model-based, iterative reconstruction technique, IMR, to filtered back projection (FBP) as a function of dose. We use detection of low contrast targets as our assessment tool.

DISCUSSION

With paired exposures, IMR was better than FBP in every subject trial. With IMR processing, detection was often nearly perfect, leading to uncertain d’. As a function of exposure, better detection was obtained with IMR at 20% exposure than with FBP at 100% exposure. CHO with internal noise gave results similar to humans, even though noise characteristics of IMR and FBP differed.

EVALUATION

4 alternative forced choice (4-AFC) experiments were used. Test sub-images of 69x69 pixels were obtained from a digital Catphan phantom. 100 contained a low contrast (0.5% contrast) 6 mm pin target and 300 contained no target. After 100 4-AFC trials, the probability correct, Pc, and d’ were determined. A channelized Hotelling observer (CHO) model with internal noise was used to model measurements. Subjects were radiologists and biomedical engineers. All images were examined using a fixed gray-scale window in a darkened room.Comparing IMR and FBP at nominal exposure (98mAs) with 8 radiologists and 4 engineers, we found that IMR gave improved detection for each subject. For FBP and IMR, average Pc values were 0.64+/-0.15 and 0.97+/-0.03, respectively, giving d’ values of 1.31+/-0.52 and large, uncertain d’, due to non-linear behavior of d’. In a second experiment, we compared IMR and FBP as a function of exposure. IMR always gave much higher Pc values and uncertain d’ values.

Cite This Abstract

Wilson, D, Miao, J, Raihani, N, Brown, K, Quantitative Image Quality Evaluation of Low Dose Iterative Reconstruction.  Radiological Society of North America 2012 Scientific Assembly and Annual Meeting, November 25 - November 30, 2012 ,Chicago IL. http://archive.rsna.org/2012/12028977.html