In one of the first pulmonary nodule validation studies using an anthropomorphic thoracic phantom and current low-dose CT technology, we have shown that a clinically used lung cancer screening protocol with low-dose CT has 100 % sensitivity of detection for spherical pulmonary nodules sized 5 mm in diameter and larger. In addition, we have shown that low-dose CT yields more accurate nodule volume measurements when using a semi-automated method than in the case of the manual method, with negligible underestimation of actual size, especially for small nodules.
We found a sensitivity of 100 % for nodules with a diameter equal to or larger than 5 mm for all three nodule densities in this anthropomorphic thoracic phantom, and 60–80 % and 0–20 % for solid and non-solid nodules with a diameter of 3 mm, respectively. This anthropomorphic thoracic phantom was also used in another study in which sensitivity of 95 % for solid nodules and 74–81 % for non-solid nodules were found for nodule diameter equal to or larger than 5 mm . Unlike the low-dose acquisition protocol for lung cancer screening in this study, the authors did not use a low-dose protocol, which limits comparability. In addition, the authors used 5-mm reconstructed slice thickness, compared with 1 mm in this study. It is well known that a larger slice width results in lower sensitivity of pulmonary nodules . This anthropomorphic phantom was also used for an image database of nodules of diameter larger than 5 mm of several shapes, but results for nodule detectability and comparing between manual and semi-automated measurements were not reported . In some nodule detectability studies, solid nodules with a diameter of 2 to 3 mm were detected in all cases [8, 13, 14]. In these studies, the nodules were placed in known order and examined in a thoracic phantom without lung vessels. On the other hand, in our study the artificial nodules were randomly positioned inside the lungs of an anthropomorphic thoracic phantom, thus limiting detection bias and strengthening the findings. As adjacent tissue can interfere with the nodule image reconstruction and reading, especially for non-solid nodules, and because this interference increases with decreasing radiation dose, we expect that this interference explains why we could not detect some of the 3-mm nodules.
No false-positive nodules were found compared with the control examinations. That is to say, all nodules detected on low-dose CT were actual nodular lesions. Nevertheless, in clinical practice, pulmonary parenchyma can be distorted and may contain scars and variations, which can erroneously be interpreted as a pulmonary nodule; thus, the specificity in a clinical setting is usually decreased.
We found an increasing underestimation of the nodule volume at smaller nodule diameters, which was also found in some previous studies [15, 16]. However, some other studies reported an increasing overestimation of the nodule volume at smaller nodule diameters [17–22]. For small pulmonary nodules, the transit zone between nodule and pulmonary background caused by partial volume effect was found to be important for accurate volumetry . Thus, measurement errors in small nodules when measured manually should be considered.
In this study, we found that the measured nodule density was significantly different from the expected density. In lung cancer screening by unenhanced CT examination, accurate CT density is important mainly to differentiate between solid and non-solid nodules, and to evaluate increases in density over time in the case of non-solid nodules. However, as we only had one type of solid nodule, and two types of non-solid nodule with a relatively large difference in CT density compared with the solid nodules, no reliable conclusion can be drawn about the potential impact of CT density variation on lung cancer screening results. Future studies with more variation in the density of solid nodules, with CT densities within the clinically relevant range, have been planned to investigate the impact of CT density on nodule volumetry in more detail.
No difference was found between low-dose 16- and 64-row multi-detector CT from the same vendor regarding manual and semi-automated volumetry. However, a previous study found different nodule volumetry outcomes for four 16-row CT systems from different vendors . As the follow-up of screened participants or clinical patients can last for an extensive period, different CT systems from different vendors can be used. A direct comparison of nodule volumes obtained from different CT systems from the same vendor seems valid, at least for the vendor investigated in this study. However, whether similar nodule volumes would have been found for other vendors is unknown.
Some lung cancer screening projects were mainly based on nodule diameter , whereas other lung cancer screening projects were mainly based on nodule volume measurements . In the National Lung Screening Trial (NLST) study, a positive result indicating suspected lung cancer on low-dose CT was defined as the presence of a nodule with a largest transverse axis of at least 4 mm . In the Dutch-Belgian Randomized Lung Cancer Screening Trial (NELSON) study, a positive result was defined as either a fast-growing nodule with a volume of at least 50 mm3, i.e., nearly 5 mm in diameter, or a nodule with a volume of at least 500 mm3 . The results of this study validate these screening protocols, as all solid nodules and non-solid nodules with a diameter of 5 mm could be detected by observers on low-dose CT against an anthropomorphic pulmonary background.
Pulmonary nodule volumetry is used to guide the diagnostic strategy in the follow-up of lung cancer screening . Repeated CT-derived volumetry of pulmonary nodules can be used to determine the risk of lung cancer and can be used to monitor tumour response in the case of non-surgical therapy. The accuracy and precision of pulmonary nodule volumetry depend on a number of factors, including image acquisition and reconstruction parameters, nodule characteristics, and the performance of algorithms for nodule segmentation and volume estimation . Size measurement needs to be as accurate as possible in order to enable the assessment of nodule growth. A commonly used criterion for pulmonary nodule growth is given by the Response Evaluation Criteria in Solid Tumors (RECIST), which states that nodules in the stable disease category should not be larger than 20 % or smaller than 30 % in diameter on subsequent CT examinations . However, a 20 % error in diameter measurement could result in an error in volume for a spherical nodule of up to 73 %, which could result in inaccurate growth rate evaluation. To improve accuracy in growth rate evaluation, semi-automated volumetry is favoured over manual volumetry.
There are limitations to this study. Firstly, only spherical nodules were used with five discrete sizes. Additional data on the sensitivity of nodules with sizes between 3 and 5 mm in diameter are needed in order to determine the sensitivity of current low-dose CT and to optimise diagnostic screening strategies for small nodules. A further extension to our study is the assessment of the sensitivity of low-dose CT for non-spherical and irregularly shaped (lobulated and/or spiculated) nodules. Secondly, we simulated healthy pulmonary tissues. The sensitivity of nodule detection is dependent on pulmonary structures, and surrounding pathological lesions such as fibrosis, emphysema or consolidation could influence nodule detectability, which can make pulmonary nodules undetectable or be erroneously interpreted as pulmonary nodules. We therefore expect that the sensitivity in an in vivo setting may be lower, and that the false-positive rate may be higher, compared with the findings in this study. Thirdly, we used only one clinical low-dose CT screening protocol. Although the sensitivity of pulmonary nodules is also dependent on CT protocol, the protocol we used is the most common in current lung cancer screening projects using thin-slice, low-dose CT. Therefore, we expect that this protocol is the most relevant for the sensitivity of pulmonary nodule detection in lung cancer screening. Finally, semi-automated volumetry was not performed for non-solid nodules, because the present commercially available volumetry software was only for solid nodules. In case of the considerable volumetry deviation from the actual volume in non-solid nodules by manual measurements, a special software package for semi-automated volume measurement of non-solid nodules is needed to assess these non-solid nodules.
In conclusion, this anthropomorphic phantom study shows that a lung cancer screening protocol with low-dose CT is highly reliable for the detection of spherical pulmonary nodules of 5 mm in diameter and larger. Low-dose CT yields more accurate nodule volumetry when using a semi-automated software tool than manual measurements, with negligible underestimation of actual size, especially for small nodules. For early lung cancer detection, in which mostly small nodules are found, accurate measurement is especially necessary to enable assessment of volume doubling time. Thus, a semi-automated volume measurement should be used in the setting of lung cancer screening. No difference in the accuracy of volumetry was found between 16- and 64-row multi-detector CT.