In the appropriate clinical setting, growth of a lung nodule is suggestive of malignancy. On the other hand, stability over more than 2 years is generally considered sufficient evidence to exclude malignancy. It has been suggested that the use of volumetric measurements, using automatic segmentation to define nodule margins on thin-section CT images, is the optimal method to assess for nodule growth [1].

Frequently, lung nodules are initially detected on screening CT examinations performed without thin-section technique, and follow-up CT studies may use varying techniques. For example, a nodule may be discovered on survey CT of the chest, abdomen, and pelvis using 5-mm-thick nonoverlapping sections, and the patient may return several months later for dedicated chest CT using 2.5-mm-thick overlapping sections. A number of effects of variations in scanning methods on volume measurements have been described in the literature [2-5]. The purpose of our study was to determine how slight differences in scanning collimation affect nodule volume measurements. Furthermore, we aimed to determine whether any differences in volume measurements were affected by nodule size, shape, or margination. In addition, we aimed to determine whether the use of two different commercially available software systems gave substantially different volume measurements.

It has been suggested that 3D volumetric measurement of pulmonary nodules is superior to 2D area measurement when assessing for small changes in lung nodule size [1]. However, the precision of volumetric analysis is imperfect. For example, using an automatic volumetric analysis tool, Wormanns and colleagues [7] reported that immediate repeat volume measurements obtained on metastatic lung nodules showed a variation of approximately ± 20%. Potential sources of volumetric variability and error include motion artifact [8], phase of cardiac cycle [3], phase of respiration (Novak at al., presented at the 2003 annual meeting of the Radiological Society of North America), and nodule location [9]. For example, juxtapleural and juxtavascular nodules have been shown to exhibit greater volume measurement variability than well-circumscribed intraparenchymal nodules [9]. In addition, volume calculations may vary substantially if nodule margin delineation is performed manually rather than using an automatic computer algorithm.

Our study concentrated on investigating the effect of CT slice thickness, in conjunction with nodule diameter, shape, and margination, on volume measurements. We also compared two different commercially available software systems for volumetric analysis. The study design, using the same scan data reconstructed with three different thicknesses, eliminated many of the potentially confounding factors described, such as differences in nodule location, degree of motion artifact, phase of cardiac cycle, or degree of respiration. Nodule segmentation and volumetric calculations were performed in an automated fashion, eliminating human bias in defining nodule margins. No juxtapleural nodules were included, thus eliminating the difficulty with automatic segmentation of such lesions. Whereas Revel and colleagues [9] found up to 6% variability in repeated volume measurements obtained from the same data set in 33% of their nodules, we found no intraobserver variability in any of the 25 nodules that we tested; this discrepancy may have been due to the large proportion of juxtapleural nodules in the data set used by Revel et al. compared with the total lack of juxtapleural nodules in our study.

In previously published studies, authors have reported greater variability and inaccuracy in volume determination of small nodules compared with large nodules for a given section thickness [2, 5, 8]. In one such study, tiny nodules (2-5 mm) varied 18.5% in volume, whereas large nodules (8-10 mm) varied 7.5% in volume [8]. In addition, recent investigations reported significant changes in volume measurements in a sizable proportion of nodules when different section thicknesses were used [4, 5]. In our study, we found that the degree of variability for small nodules was greater than the degree of variability for large nodules when comparing volumes obtained with different section thicknesses. Presumably, this finding is due to increased partial volume averaging for small nodules when thick sections are used; the effect of partial volume averaging is not as severe with large nodules. Partial volume averaging could also account for the failure of segmentation of 18 nodules on 5-mm CT sections, with 14 of these 18 nodules being 5 mm or less in greatest dimension. The results of our study suggest that thick sections (e.g., 5 mm) may be adequate for analysis of large nodules (> 15 mm in diameter), whereas thin sections (e.g., 1.25 mm) appear to be necessary for nodules < 10 mm.

In an investigation, Winer-Muram and colleagues [2] found that an increase in section thickness usually led to an increase in the estimation of nodule volume. However, that study was limited by manual, rather than automatic, nodule margin definition; the use of data sets obtained on two different dates, rather than using the same intrinsic scan data; and the use of thick sections (8 or 10 mm) in all but one patient. Zhao et al. [4] recently reported that increased section thickness led to increased mean calculated nodule volumes in lung metastases, although results for individual nodules were not reported. Our data also showed increased mean nodule volume with increased section thickness; however, analysis of individual volumes revealed that volume increased substantially in some nodules and decreased substantially in others. By reporting only mean volumetric data, it is possible that the findings reported by Zhao et al. may not reflect the full variation in volumetric results that was observed. Whereas our study showed lack of a statistically significant difference between volume measurements for several nodule size groups using different section thicknesses, many individual nodules showed large volume differences.

Neither the size breakdown nor the margination of nodules is specified in the study of Zhao et al. [4], so there may be differences in nodule populations between that study and ours. Moreover, those authors used a custommade volume system, whereas we used commercially available systems. In addition, section thicknesses used in our study (5, 2.5, and 1.25 mm) were thinner than those used by Zhao et al. [4] (7.5, 5, and 3.75 mm). Currently, CT for lung cancer screening and follow-up of small nodules that might represent bronchogenic carcinoma is generally performed using section thicknesses of no more than 2.5 mm, whereas screening and followup for lung metastases may be performed using thicker sections; thus, the methodology used to design these two studies differed because researchers had different patient populations in mind.

Winer-Muram and colleagues [2] found increased volume variability for irregular nodules, and our study showed similar results for spiculated versus smooth nodules, at least using one of the software packages. This is particularly important in lung cancer screening programs because spiculated nodules have traditionally been considered more suspicious for malignancy than smooth nodules.

Not only did our study use commercially available nodule segmentation software, but it also used two different software systems. The results indicate that although the effect of slice thickness is similar using both of these systems, there can be large differences in individual nodule volume measurements when shifting from one software system to another.

Our study had several limitations, including selection bias, with regard to nodules that were analyzed. Because juxtapleural, juxtavascular, and ground-glass nodules were excluded, our results cannot be applied to those types of commonly encountered nodules. In addition, our examinations lacked a small, focused field of view, possibly limiting the spatial resolution; a full patient-appropriate field of view was used because these were clinical examinations obtained to examine the lungs in their entirety. We had no proof of true nodule volumes; however, the purpose of our study was to evaluate the differences in calculated volumes using different section thickness, not to determine the accuracy of the volume technique. Because of the study design, many of the variables inherent in physically scanning a patient twice were eliminated; thus, our results reflect a best-case scenario with regard to measurement variability, and it would be expected that actual follow-up scanning would lead to larger apparent changes in nodule volumes.

In conclusion, although few statistically significant differences were found between volume measurements obtained using different slice thicknesses, our results indicate that a substantial proportion of individual nodules showed a large change with different section thicknesses on both software packages. This false change in nodule volume was most likely to occur for tiny and small (≤ 10 mm) nodules. When different software packages were used, significantly different volume measurements were generated at a 2.5-mm CT section thickness for all sizes of nodules. Nodule shape did not appear to affect the results using either software package, whereas margination affected volume measurements when using one of the software packages. The effect of section thickness was also found to be different in the two software systems. We therefore suggest than when following a nodule over time, it is best to use the same section thickness and overlap and the same segmentation software for all studies.

Address correspondence to M. Petrou ([email protected]).