Jonathan Hero Chung MD, Presenter: Nothing to Disclose

Kalpana M. Kanal PhD, Abstract Co-Author: Nothing to Disclose

Jin Wang PhD, Abstract Co-Author: Nothing to Disclose

Puneet Bhargava MBBS, Abstract Co-Author: Nothing to Disclose

Martin Lee David Gunn MBChB, Abstract Co-Author: Nothing to Disclose

William Phelps Shuman MD, Abstract Co-Author: Research grant, General Electric Company, Milwaukee, WI

Bill H. Warren MD, Abstract Co-Author: Nothing to Disclose

Jennifer Ruth Kohr MD, Abstract Co-Author: Nothing to Disclose

Brent K. Stewart PhD, Abstract Co-Author: Nothing to Disclose

00030490-DMT et al, Abstract Co-Author: Nothing to Disclose

For low-contrast liver lesions in phantom and patient 64-channel MDCT, we wished to investigate the relationship between noise index/radiation dose and diagnostic accuracy.

A helical CT phantom containing hypodense lesions (2.4mm-10mm) in a liver equivalent background was scanned with a GE 64-channel CT scanner at sequential noise indices (5 to 49). Circular areas of 3 cm diameter were extracted from the phantom scans; each image either contained a single hypodense lesion or no lesion. The extracted images were scored on a PACS workstation by 3 radiologists using a 4-point lesion detectability scale ranging from no lesion to definite lesion. Subsequently, 50 patient images containing liver lesions and 50 patient images containing no liver lesions were retrospectively collected; scans had been performed with a noise index (NI) of 15. Three additional 100-image sets were created by adding quantum noise with a vendor-validated noise addition tool to simulate 25% less radiation (NI 17.4), 50% less radiation (NI 21.2), and 75% less radiation (NI 29.7). These 400 images were scored on PACS workstation by 3 radiologists using a 5-point Likert lesion detectability scale.

Inter-reader agreement showed good to excellent correlation, ranging from 0.71 to 0.85, during all phases of the study. Phantom: At a noise index of 11-15, the average AUC (area under the ROC curve) was 0.86 (p>0.05), indicating a good, but not definitive test. The average AUC dropped to 0.77 (p>0.05) at a noise index of 17-21. Further decreasing the noise index to 23-27 resulted in a drop in average AUC to 0.72 (p>0.05) indicating a fair test. Clinical images: Original images (NI 15) demonstrated excellent average AUC of 0.967 (p<0.001). Performance did not decrement for simulated images at 25% or 50% decreased radiation dose [average AUC of 0.973 (p<0.001) and 0.960 (p<0.001), respectively]. Performance decremented marginally for simulated images at 75% dose reduction (average AUC of 0.913 (p<0.001)).

Our study suggests that low-contrast liver lesions can be detected at substantially lower radiation dose than currently employed without loss of diagnostic accuracy.

Low-contrast liver lesions can be detected at substantially higher noise levels (lower radiation dose) than currently employed at our institution without loss of diagnostic accuracy.

Chung, J, Kanal, K, Wang, J, Bhargava, P, Gunn, M, Shuman, W, Warren, B, Kohr, J, Stewart, B, et al, 0, Optimal Noise Index in the Evaluation of Subtle Liver Lesions: Phantom and Clinical Evidence.  Radiological Society of North America 2009 Scientific Assembly and Annual Meeting, November 29 - December 4, 2009 ,Chicago IL. http://archive.rsna.org/2009/8002889.html