Room: AAPM ePoster Library
Purpose: To quantitatively compare image noise as a function of virtual monochromatic energy (keV) of computed tomography (CT) images reconstructed using filtered back-projection (FBP), hybrid iterative reconstruction (ASiR-V), and deep learning image reconstruction (TrueFidelity) for Gemstone Spectrum Imaging (GSI).
Methods: To assess the image noise across the image reconstruction algorithms, we calculated the standard deviation in CT images of a uniform phantom scanned with dual energy (GSI). The images were reconstructed using FBP, iterative reconstruction (ASiR-V, 60%) and deep learning (TrueFidelity, DLIR-H). These algorithms are based upon the GSI image chain [Pal, SSA19-03] which includes correlated denoising such that image noise decreases monotonically as a function of keV.
A 30cm water phantom was scanned on Revolution CT (GE Healthcare, Waukesha, WI), and 0.625mm thick slices from 40 keV to 140 keV in 1 keV increments were reconstructed using the three reconstruction algorithms above. Noise was computed as the standard deviation of a 2D ROI placed at the center of the uniform water image. The image noise was averaged across multiple slices and reported for each algorithm.
Results: Image noise decreased monotonically as a function of virtual monochromatic energy across the three reconstruction algorithms. ASiR-V (42.8% ± 0.2%) and TrueFidelity (56.6% ± 0.9%) had similar noise reduction performance relative to FBP.
Conclusion: TrueFidelity image reconstruction allows for additional noise reduction in GSI dual energy imaging while preserving the noise vs. monochromatic energy behavior.
Funding Support, Disclosures, and Conflict of Interest: Employees of GE Healthcare
Not Applicable / None Entered.