Modeling of measurement error range in lung nodule volumetry: Prospective study with same day repeat computed tomography for metastatic lung nodules

Abstract

Introduction: Semi-automated lung nodule volumetry is a promising tool for early detection of volume change, which might led to early diagnosis of malignancy and treatment failure. Measurement variability in volumetry is known to be affected by various factors. The purpose of our study was to model the range of variability in lung nodule volumetry, in patient with metastatic lung nodules by same day repeat computed tomography (CT) scans Methods: The present prospective study included 50 patients with known pulmonary metastatic nodules between November 2013 and April 2014, with written informed consents. Two consecutive noncontrast chest CT scans were performed within 10 minutes of time interval. Non-calcified nodules with diameter between 4mm and 15mm were segmented using in-house software for each CT scans. After calculation of mean segmented nodule volume (Vm), surface voxel proportion (SVP) was defined as the proportion of surface voxels in total segmented voxels, while attachment proportion (APN) was defined as the proportion of voxels with greater attenuation than N in total voxels just outside of surface of nodule. Absolute percentage error (APE) and relative percentage error (RPE) were calculated from segmented nodule volume on each CT scans. Univariate and multivariate quantile regression analyses were performed for estimation of 95% upper limit of APE. Results: The 95% limits of variability of RPE was from -20.81% to 20.62%. In univariate quantile analyses, Vm, Sphericity, SVP, AP-700 and AP-600 were significant variables for estimation of APE. In multivariate analysis, SVP and AP-700 were proven to be independently significant variable and final model for 95% limit of APE was as follows: APE = 46.01• SVP + 36.32 • AP-700 -12.94. Conclusions: In conclusion, SVP and AP were independent factors for variability in lung nodule volumetry. With those two parameters, 95% limit of absolute percentage error in lung nodule volumetry could be estimated with linear model.