Author: Zijia Guo, Michael F. McNitt-Gray, Frederic Noo, Karl Stierstorfer 👨🔬
Affiliation: Siemens Healthineers, University of Utah, David Geffen School of Medicine at UCLA 🌍
Purpose: Accurately assessing lung parenchyma health is critically important in the management of patients with chronic obstructive pulmonary disease. CT attenuation values are valuable for this purpose. However, they lack absolute accuracy due to various physical effects. We aim to assess how beam hardening affects lung CT attenuation values in a controlled experiment where other physical effects are avoided.
Methods: Voxelized patient models were created with each voxel representing soft tissue or bone. Bone voxels were modeled as cortical bone of various density, chosen to match Hounsfield Unit values at 120kV. The lung parenchyma was modeled as low-density water with attenuation value of −925HU. One female and two male patient models were considered, and four axial slices evenly distributed throughout the chest were selected within each of these. Realistic polychromatic fan-beam data was generated for each slice and reconstructed using filtered back-projection. X-ray source spectrum, bowtie filter, and detector energy response were included. Pre-filtration with tin was investigated. Beam hardening correction was applied using channel-dependent polynomials that convert the data to water lengths.
Results: The images corrected for beam hardening exhibit subtle shade artifacts within the lung parenchyma that can be traced back to interactions between the vertebra and ribs. The orientation of these artifacts is accordingly both patient- and slice-dependent. The error varies between -12HU and +18HU. For healthy tissue that differs from diseased tissue by +8HU, the beam hardening error results in a drop of 0.06 in discrimination performance measured by the area under the ROC curve. The tin filter eliminates the drop.
Conclusion: We assessed the effect of beam hardening on the accuracy of CT attenuation values independently of other sources of bias. Subtle patient and slice-location dependent effects were highlighted with impact on discrimination between tissues for contrast differences below 15HU, unless tin filtration is used.