Assessing the impact of nodule features and software algorithm on pulmonary nodule measurement uncertainty for nodules sized 20 mm or less
Artit Jirapatnakul,Rowena Yip,Kyle J Myers,Siyang Cai,Claudia I Henschke,David Yankelevitz
DOI: https://doi.org/10.21037/qims-23-1501
2024-07-01
Abstract:Background: Measurements are not exact, so that if a measurement is repeated, one would get a different value each time. The spread of these values is the measurement uncertainty. Understanding measurement uncertainty of pulmonary nodules is important for proper interpretation of size and growth measurements. Larger amounts of measurement uncertainty may require longer follow-up intervals to be confident that any observed growth is due to actual growth rather than measurement uncertainty. We examined the influence of nodule features and software algorithm on measurement uncertainty of small, solid pulmonary nodules. Methods: Volumes of 107 nodules were measured on 4-6 repeated computed tomography (CT) scans (Siemens Somatom AS, 100 kVp, 120 mA, 1.0 mm slice thickness reconstruction) prospectively obtained during CT-guided fine needle aspiration biopsy between 2015-2021 at Department of Diagnostic, Molecular, and Interventional Radiology in Icahn School of Medicine at Mount Sinai, using two different automated volumetric algorithms. For each, the coefficient of variation (standard deviation divided by the mean) of nodule volume measurements was determined. The following features were considered: diameter, location, vessel and pleural attachments, nodule surface area, and extent of the nodule in the three acquisition dimensions of the scanner. Results: Median volume of 107 nodules was 515.23 and 535.53 mm3 for algorithm #1 and #2, respectively with excellent agreement (intraclass correlation coefficient =0.98). Median coefficient of variation of nodule volume was low for the two algorithms, but significantly different (4.6% vs. 8.7%, P<0.001). Both algorithms had a trend of decreasing coefficient of variation of nodule volume with increasing nodule diameter, though only significant for algorithm #2. Coefficient of variation of nodule volume was significantly associated with nodule volume (P=0.02), attachment to blood vessels (P=0.02), and nodule surface area (P=0.001) for algorithm #2 using a multiple linear regression model. Correlation between the coefficient of variation (CoV) of nodule volume and the CoV of the x, y, z measurements for algorithm #1 were 0.29 (P=0.0021), 0.25 (P=0.009), and 0.80 (P<0.001) respectively, and for algorithm #2, 0.46 (P<0.001), 0.52 (P<0.001), and 0.58 (P<0.001), respectively. Conclusions: Even in the best-case scenario represented in this study, using the same measurement algorithm, scanner, and scanning protocol, considerable measurement uncertainty exists in nodule volume measurement for nodules sized 20 mm or less. We found that measurement uncertainty was affected by interactions between nodule volume, algorithm, and shape complexity.