A physics-based image harmonization pipeline improves reproducibility of lung density measurements from chest CT scans compared with the traditional MF-VALD approach, enabling more robust quantification across varying imaging conditions.

Sequential modules demonstrate progressively larger gains: aligning spatial resolution reduces Perc15 to 18.6 HU, adding noise matching further lowers it to 16.4 HU, and incorporating lung volume adjustment brings it down to 7.4 HU.

The study, led by Duke University’s Saman Sotoudeh-Paima, analyzes 1,159 COPDGene participants who had same-day full-dose and reduced-dose chest CTs between late 2014 and mid-2017.

Across all imaging conditions, harmonization reduces the Perc15 reproducibility coefficient from 35.6 HU to 7.4 HU, a 4.8-fold improvement.

The work was published in Radiology: Cardiothoracic Imaging on March 12, 2026, with full article access via the DOI.

The authors contend that more precise, harmonized lung-density quantifications could enable better assessment of disease severity and support longitudinal monitoring using local, harmonized metrics.

The combined noise harmonizer component outperforming volume-adjusted density alone achieved 7.7 HU versus 30.1 HU when used independently.

The harmonization pipeline targets three main sources of variability in CT lung density measurements—spatial resolution, image noise, and lung volume—via a sequential, physics-based processing framework.