Author: Frank F. Dong, Megan C. Jacobsen, Ke Li, Xinming Liu, Humberto Monsivais, John Rong 👨🔬
Affiliation: Purdue University, UT MD Anderson Cancer Center 🌍
Purpose: To investigate the noise characteristics of a clinical photon counting detector CT (PCD-CT) system along axial and through-plane (Z) dimensions using 3D NPS measurements.
Methods: This study examined four NAEOTOM Alpha PCD-CT reconstruction kernels and five reconstruction types: T3D, 70keV virtual monoenergetic, 70 keV with Quantum Iterative Reconstruction strength 3 (70keV-QIR3), total-energy bin (T1), and high-energy bin (T2) reconstructions (obtained using Expert Mode). Phantom scans were repeated and subtracted to generate noise-only image volumes of each reconstruction type for calculating 3D NPS. 3D NPS was integrated along the kz axis to produce the axial 2D NPSxy, while the 3D NPS was integrated along both kx and ky to generate the 1D NPSz. Noise magnitude was also evaluated for slice thicknesses between 0.4 and 10 mm.
Results: For a given kernel, T1 and T2 images exhibit the highest noise magnitude but the least noise correlation along Z, with flat NPSz along the kz axis. Noise correlation along the Z dimension was found in the reconstructed T3D, 70keV, and 70keV-QIR3 images, with 70keV-QIR3 showing the strongest Z correlation. Due to this correlation, image variance of T3D and VME was no longer inversely proportional to the reconstructed slice thickness; instead, it was inversely proportional to the 0.9th power of thickness. At thicknesses>3 mm, the noise magnitude of T3D was equivalent to that of T1. In the axial plane, all image types except 70keV-QIR displayed similar NPSxy shapes for a given kernel. The shape of NPSz does not show significant dependence on the kernel.
Conclusion: The 3D NPS results suggest that T3D and monoenergetic reconstructions introduce noise correlation along both the detector column and row directions. QIR enforces stronger denoising in both directions. Comparisons of PCD-CT with EID-CT should account for potentially different Z correlation conditions to ensure a fair and accurate evaluation.