(Sunday, 7/12/2020) [Eastern Time (GMT-4)]
Room: AAPM ePoster Library
Purpose: quantify the basic physics measurements of TrueFidelity Deep Learning Image Reconstruction as applied to Gemstone Spectral Imaging (GSI).
Methods: GSI acquisition was used on a Revolution CT system (GE Healthcare, Waukesha WI). The system offers noise reduction via a hybrid iterative reconstruction method (ASIR-V). In this abstract we compare a Deep Learning Image Reconstruction method (TrueFidelity) applied to GSI data. The denoising occurs on the water and iodine basis data so that all material basis images and monochromatic images can be generated with a single denoising operation. For the purpose of this study ASiR-V 60% (high range of clinically prescribed settings) was compared with DLIR-H (High setting among three options).
Quantitative measurements were taken from the Gammex Multi-Energy CT Phantom (Sun Nuclear, Middleton WI), where the iodine rods were replaced with cylinders of liquid iodine at concentrations of 5, 10, 15 and 20ml. Standard noise and spatial resolution measurements were made as well using a 20 cm water phantom (at 78 keV) and a tungsten wire respectively.
Results: The average value in the material basis images reconstructed for water(iodine) within the water rod and the 5mg/ml rod were: (ASiR-V: 996.6, 997.3) and (DLIR: 996.7,998.4). The average value in the iodine(water) basis images within the 5mg/ml, 10mg/ml, 15mg/ml and 20mg/ml were: (ASiR-V: 4.93, 10.13, 14.90, 20.16) and (DLIR: 4.88, 10.07, 14.87, 20.09). The image noise values in HU were (ASiR-V: 2.8, 2.73) and (DLIR: 2.04, 1.99). The MTF 10% values (1/cm) were: (ASiR-V: 6.36) and (DLIR: 6.36).
Conclusion: Quantitative measurements in the basis images and image resolution varied by less than 5%, while the image noise was reduced by more than 25% for TrueFidelity (DLIR-H) compared with ASiR-V 60%.
Funding Support, Disclosures, and Conflict of Interest: Employees of GE Healthcare
Keywords
Not Applicable / None Entered.
Taxonomy
Not Applicable / None Entered.
Contact Email
