Program Information
BEST IN PHYSICS (IMAGING): A Channelized Hotelling Observer Model-Based Image Quality Survey of Routine Abdomen Protocols On a Diverse Fleet of CT Scanners
C Favazza*, A Ferrero , S Dirks , J Weaver , L Yu , S Leng , C McCollough , Mayo Clinic, Rochester, MN
Presentations
TH-EF-601-1 (Thursday, August 3, 2017) 1:00 PM - 3:00 PM Room: 601
Purpose: To perform a survey on low contrast detectability in routine abdominal CT for different scanner models using a channelized Hotelling observer (CHO) model and to assess different evaluation metrics for meaningful comparisons of scanner performance.
Methods: A custom 3D-printed phantom insert containing 24 different low contrast rods (6 contrast levels and 4 diameters) was placed inside three different sized anthropomorphic abdominal phantoms (30, 35, and 40 cm lateral dimensions) . The phantoms were scanned using 10 different CT scanner models, representing three major vendors. Each phantom was scanned 50 times using the scanner’s clinical routine abdomen protocol, yielding 50 independent images for each object and 50 corresponding background images to train and test the CHO model. The area under the ROC curve (AUC) and detectability index (d’) for the detection task were evaluated for each scanner, phantom size, and object. CTDIvol values were recorded for each scanner and phantom size. Performance and dose data from each scanner-protocol combination were compiled, analyzed, and compared with all other scanner-protocol combinations.
Results: CHO performance varied over the scanners investigated. Measurements yielded maximum differences in AUC of 0.19 for the lowest contrast object (averaged over all object sizes) and 0.17 for the smallest object (averaged over all object contrasts). Normalizing d’ for all objects by a Rose model-based scale factor, [object contrast*object diameter]⁻¹, enabled the computation of a single, mean detectability score (mean d’Rose) and corresponding dose efficiency value (mean d’Rose/CTDIvol) for each scanner and phantom size. Across all scanners, mean d’Rose ranged from 2.3-3.9, 1.6-2.9 and 1.2-2.5 [cm-10HU]⁻¹, whereas dose efficiency values ranged from 3.1-5.3, 1.4-2.6 and 0.6-1.5 [cm-10HU-10mGy]⁻¹ for small medium and large phantoms, respectively.
Conclusion: Task-based image quality assessment of different scanners and protocols demonstrated varied performance, providing objective data to guide dose optimization and image quality standardization.
Funding Support, Disclosures, and Conflict of Interest: CHM received research support from Siemens Healthcare.
Contact Email: