Yi Zhang, Presenter: Nothing to Disclose

Shuai Leng PhD, Abstract Co-Author: Nothing to Disclose

Lifeng Yu PhD, Abstract Co-Author: Nothing to Disclose

Cynthia H. McCollough PhD, Abstract Co-Author: Research Grant, Siemens AG

The performance correlation between human observers and a channelized Hotelling observer (CHO) has been previously demonstrated in low contrast lesion detection and lesion shape discrimination tasks using CT images. The purpose of this study was to use CHO to quantitatively assess the performance of an iterative reconstruction (IR) method and a traditional filtered backprojection (FBP) method in these two tasks.

11 lesion-mimicking rods were placed in a 35 × 26 cm torso-shaped water phantom and scanned on a 128-slice CT scanner (Definition Flash, Siemens) with 120 kV at 4 different mAs settings: 120, 240, 360 and 480 quality reference mAs (CAREDose4D), each repeated 100 times. CT images were reconstructed using FBP and IR (SAFIRE) methods available from the scanner. 3 rods had -15 HU contrast to mimic low contrast lesions. The other 8 rods had a relative higher contrast (2 shapes × 2 contrasts × 2 sizes), which were used to construct shape discrimination tasks, in which hexagon and circle rods were displayed side-by-side in a randomized order. A total of 5600 2-alternative forced-choice trials were analyzed using CHO with Gabor filters (6 passbands, 5 orientations and 2 phases) and percent correct (PC) was calculated for each task. Internal noise was added to test variables to estimate the variation of PC.

For the low-contrast lesion detection task, CHO didn’t predict improved performance with SAFIRE for most dose and size settings. The only improvements were observed for 3 mm lesion at 360 mAs, from 96±2% to 98±1% (p<0.05, two-tailed unpaired t-test), and for 5 mm lesion at 120 mAs, from 92±3% to 94±2% (p<0.05). In contrast, SAFIRE significantly improved CHO performance for lesion shape discrimination task for almost all dose, size and contrast settings (p<0.05), except for 480 mAs, and for 13 mm lesion with 90HU-contrast, likely due to the fact that PCs in these settings were close to saturation.

Image quality and performance improvement of IR depends on diagnostic tasks. A previously validated CHO predicted that there is a significant improvement of performance with IR for lesion shape discrimination task but not for low-contrast lesion detection task.

IR could be used to improve performance or reduce dose, with the amount of dose reduction dependent on lesion size and contrast, as well as the diagnostic task.

Zhang, Y, Leng, S, Yu, L, McCollough, C, Performance Evaluation of a CT Iterative Reconstruction Algorithm in Detection and Discrimination Tasks Using Model Observer.  Radiological Society of North America 2013 Scientific Assembly and Annual Meeting, December 1 - December 6, 2013 ,Chicago IL. http://archive.rsna.org/2013/13013496.html