Room: Davidson Ballroom B
Purpose: It is well understood that quantitative emphysema scoring metrics such as RA-950 and PERC15 are sensitive to CT acquisition and reconstruction parameters. As a result, studies utilizing these metrics as end-points (e.g., SPIROMICS, COPDGene) require strict protocol standardization. In this study, simple image domain denoising with an adaptive bilateral filter was investigated as a means to improve the robustness of quantitative emphysema scoring to changes in dose, reconstruction kernel, and slice thickness.
Methods: Scans from 142 lung screening subjects were reconstructed using weighted filtered backprojection (FreeCT_wFBP) and different combinations of slice thickness (0.6, 1.0, 2.0mm), reconstruction kernel (smooth, medium, sharp), and acquisition dose (100%, 50%, 25%, 10% of clinical lung screening dose) for a total of 36 reconstructions per subject. Change in emphysema score relative to a reference condition (100% dose, 1.0mm slices, smooth kernel) was computed for each reconstruction. Bilateral filtering, adapted based on slice thickness and dose, was applied and changes relative to reference were recomputed and compared. A parameter configuration was deemed "acceptable" if it resulted in an average change of <5% RA-950 score, or <10HU PERC15 score (i.e. unlikely to change a subjects diagnosis).
Results: Without adaptive denoising, only a small subset of parameter configurations investigated produced "acceptable" results. However, with adaptive bilateral filtering all reconstruction conditions investigated were acceptable, with a maximum average change of approximately 3%. Subjects with high emphysema scores at reference were sensitive to the very low-dose configurations (10%, 25% configurations) and sharp reconstruction kernel. Results were similar for PERC15 measurements.
Conclusion: Adaptive bilateral filtering was an effective means to improve the reliability of quantitative emphysema scoring, and presents possible pathways to substantially extend the clinical use of of these methods in the absence of strict protocol standardization.
Funding Support, Disclosures, and Conflict of Interest: Funding support for this research was provided in part by the University of California Office of the President Tobacco-Related Disease Research Program (UCOP-TRDRP grant #22RT-0131) and the National Cancer Institutes Quantitative Imaging Network (QIN grant U01-CA181156).