Enabling Consistent COPD Quantification across Energy-Integrating and Photon-Counting CT Acquisitions for Multi-Center Imaging Trials
Abstract
Purpose
To enable consistent COPD quantification in multi-center studies across energy-integrating (EICT) and photon-counting (PCCT) CT technologies, using a physics-based image harmonization technique.
Methods
Under an IRB-approved COPDGene Phase IV protocol, research subjects were imaged during a single visit on both EICT and PCCT scanners (SOMATOM Force and NAEOTOM Alpha; Siemens Healthineers) under matched radiation exposure conditions (14 – 47 mAs) at two hospitals. The acquired CT projection data were reconstructed using varied, clinically relevant algorithms (ADMIRE for EICT and PNR 70 keV-VMI for PCCT), kernels (Qr36-Qr44), in-plane pixel sizes (0.39 and 0.78 mm), and slice thicknesses (0.75/0.80 and 1.50 mm). Reconstructed images were transformed to a low-noise reference condition on PCCT (Qr36, 0.78x0.78x1.50 mm3) consistent with established quantitative chest-CT protocols utilizing a physics-based approach that harmonizes image noise and spatial resolution (via global noise index (GNI) and modular transfer function (MTF) measurements) and lung inflation volume. Furthermore, to address inter-scanner variability in HU-based lung density measurements, a bias correction based on conservation of lung mass was applied. The consistency of emphysema quantifications - LAA-950 (percentage of lung voxels below -950 HU) and Perc15 (15th percentile of the lung density histogram) - was evaluated pre- and post-harmonization across reconstruction conditions, patients, and scanner technologies.
Results
For LAA-950, the mean absolute deviation (MAD) and coefficient of variation (CoV) relative to the reference decreased from 4.0% and 1.8 (pre-harmonization) to 0.8% and 0.6 (post-harmonization). Similarly, for Perc15, MAD decreased by 64.6% from 10.8 HU to 3.8 HU, with a 45.6% reduction in CoV. For both biomarkers, reductions in deviation from the reference were statistically significant (signed-rank test, p << 0.001), indicating improved consistency of lung density measurements post-harmonization across acquisition conditions.
Conclusion
Physics-based image harmonization enabled consistent COPD quantification by mitigating variability introduced by CT technology (EICT vs. PCCT) and reconstruction techniques.