Xiang Li, Presenter: Research grant, General Electric Company

Ehsan Samei PhD, Abstract Co-Author: Advisory Board, Ion Beam Applications, SA Consultant, Siemens AG

Huiman Barnhart PhD, Abstract Co-Author: Nothing to Disclose

Ana Maria Gaca MD, Abstract Co-Author: Nothing to Disclose

Caroline Laurens Hollingsworth MD, Abstract Co-Author: Nothing to Disclose

Charles M. Maxfield MD, Abstract Co-Author: Nothing to Disclose

Caroline W. T. Carrico MD, Abstract Co-Author: Nothing to Disclose

James G. Colsher PhD, Abstract Co-Author: Employee, General Electric Company

Donald P. Frush MD, Abstract Co-Author: Research Consultant, General Electric Company Research Consultant, Siemens AG

To determine the quantitative relationship between image quality (nodule detectability and noise) and the diagnostic accuracy of radiologists in detecting small lung nodules in pediatric CT.

The study included clinical chest CT images of 30 pediatric patients (0-16 years old) acquired at tube currents of 55-180 mA. Calibrated noise addition software was used to simulate cases as they would have been acquired at three nominal tube current settings: 70 mA (lowest original mA for all but one patient), 35 mA (50% reduction), and 17.5 mA (75% reduction), resulting in a range of quantum noise values between 7 and 32 Hounsfield unit (HU). Using a validated nodule simulation technique, small lung nodules with diameters of 3-5 mm and peak contrasts of 200-500 HU were inserted into the cases, which were then randomized and rated independently by four experienced pediatric radiologists for nodule presence on a continuous scale from 0 (definitely absent) to 100 (definitely present). The receiver operating characteristic (ROC) analysis was used to quantify the relationship between diagnostic accuracy (the area under the ROC curve, Az) and nodule detectability (the product of nodule peak contrast and displayed diameter to noise ratio, CDNRdisplay). Diagnostic accuracy as a function of noise was then determined for representative nodule diameter and contrast values.

Az increased rapidly from 0.63 ± 0.06 to 0.88 ± 0.03 (p < 0.001) when CDNRdisplay increased from 44 to 99 mm, followed by a slow increase to 0.92 ± 0.03 (p = 0.52) when CDNRdisplay further increased to 265 mm. For a given nodule size and contrast, Az increased rapidly with decreasing noise, but reached a plateau beyond a threshold noise. The threshold noise was larger for higher nodule size and contrast.

Diagnostic accuracy increases with nodule detectability, but reaches a plateau beyond a threshold detectability level. For a given nodule size and contrast, diagnostic accuracy increases with decreasing noise, but reaches a plateau at a threshold noise, the value of which increases with increasing nodule size and contrast.

The protocol- and scanner-independent relationship between image quality and diagnostic accuracy can guide CT protocol design to achieve the desired diagnostic accuracy at the minimum radiation dose.

Li, X, Samei, E, Barnhart, H, Gaca, A, Hollingsworth, C, Maxfield, C, Carrico, C, Colsher, J, Frush, D, Lung Nodule Detection in Pediatric CT: Quantitative Relationship between Image Quality and Radiologist Performance.  Radiological Society of North America 2010 Scientific Assembly and Annual Meeting, November 28 - December 3, 2010 ,Chicago IL. http://archive.rsna.org/2010/9005637.html