We showed that dynamic single-slice dual-energy CT images closely approximate whole-lung scans in pigs with and without a lung injury model over a variety of different PEEP levels and tidal volumes, with a small effect of inspiration upon this agreement. During inspiration, approximately one third of the slice moved out of the CT scanner field-of-view; however, it was replaced by the adjacent slice, which was of similar composition. The intra-tidal variation in bias between dCT and whole-lung imaging was less than 10% of the range of the studied parameter.
Comparison of dCT images with those taken during breath-holds
Mean dCT variables closely correlated with those from whole-lung scans with r2 ≥ 0.75. There was a bias toward underestimating CT lung density, soft tissue density, iodinated blood density, atelectasis and areas of poor aeration on the dCT slices; however, the mean bias was 31.2 HU in terms of CT density (equivalent to 3% of the HU range from − 1000 to 0) and at most 7.7% of the volume fractions measured. The correlation coefficients between the 4 continuous distributions of voxels tested were all ≥ 0.89, consistent with excellent agreement. This latter result is also important because the mean voxel value seen in the dCT slice does not adequately describe the range of voxel densities with that slice—typically, a bimodal distribution was seen (Fig. 3). Taken together, these results provide strong evidence that the make-up of the single dCT slice chosen approximated the whole lung.
Previous experiments in uninjured pigs have demonstrated that a subcarinal slice overestimates both overdistension and atelectasis and underestimates ventilated regions [9], in contrast with our results. Interestingly, the overdistended regions represented a significantly smaller fraction of the slice/lung volume in our study compared with those previously described (0.6% in dCT vs 1.2% in the whole lung as compared with 6.4% in dCT vs 3.1% in the whole lung in published results [9]). A possible reason for this discrepancy is that our study used 5 mm instead of 1-mm-thick CT slices. When thicker slices are used, as in our study, overdistended regions may comprise less of the whole lung volume, presumably due to the heterogeneity within a larger voxel tending to be averaged toward an intermediate CT density [22, 23].
A separate study in an oleic-acid lung injury model in pigs has demonstrated no significant difference between either a juxtadiaphragmatic or apical dCT slice and that of the whole lung in terms of CT density distributions [10]. Examination of the density distributions in these results demonstrates a trend for dCT to underestimate atelectatic fraction in expiration. Despite the differences in the lung injury model, the volume fractions of atelectasis, poor aeration, normal aeration and overdistension in this study were similar to those we demonstrated.
Effects of inspiration upon the imaged slice
Dynamic CT studies examine the effect of the ventilatory cycle upon the lung. As such, a bias between dCT and the whole-lung imaging may not be a major limitation: as long as this bias remains constant throughout the respiratory cycle, the delta-change values seen in dCT can be used to predict changes at the whole-lung level. We did, however, demonstrate a variable bias, which decreased in inspiration (Fig. 2, Table 3). This result suggests that the usage of dCT to estimate whole-lung CT parameters introduces an artefact, which could be erroneously identified as intra-tidal changes occurring within the lung. The mean variation in bias throughout the respiratory cycle was 39 HU in terms of CT density (equivalent to 3.9% of the typically investigated HU range between − 1000 and 0) and ranged between 0.1% and 6.9% for volume fractions of the various materials. We also report the 95% limits of these values such that researchers who demonstrate an effect size outside this range can be confident that the results seen are due to changes in the underlying pathophysiology, rather than as an artefact due to the use of dCT technology.
The direction of this variable bias is also important. In the three variables that demonstrated the largest artefactual effect of inspiration (CT density, poorly and normally aerated volume fractions), the variation in bias during inspiration is in the opposite direction to what would be expected. For example, normally aerated volume fraction typically increases with inspiration [7]; however, we demonstrated an artefactual reduction in bias associated with inspiration with 95% limits between − 12% and − 1.7% (Table 3), thus demonstration of an inspiration-related increase in normally aerated volume fraction represents an actual physiological change rather than artefact. Similarly, our results suggest that any inspiration-related decrease in CT density, blood volume or atelectatic volume fraction, and inspiration-related increases in blood volume above 4.7%, or overdistended volume fraction of more than 0.7% are not due to artefact.
A partial explanation for the improved matching between dCT and whole-lung variables in inspiration is evidenced by examining the effects inspiration has upon the slice. We demonstrated that the part of the lung that comprises the slice in expiration moves caudally with inspiration. Approximately 32% of the slice moved more than 5 mm in inspiration with the majority of this movement occurring within the middle third of the lung following injury, or within the middle and dependent thirds in uninjured animals (Fig. 5). The amount that moved more than 10 mm was minimal. It can be expected, therefore, that the part of the lung imaged in dCT during inspiration comprises two thirds of the slice imaged during expiration and one third of the next 5-mm slice above this.
Our CT scanner was able to simultaneously acquire three adjacent 5 mm slices during dCT. Thus, we were able to determine the effects this inspiratory movement would have had upon the make-up of the imaged slice. We demonstrated a small effect of cranio-caudal distance upon CT density, gas and normally aerated volume fractions, with the former increasing in the cranial direction (i. e. in the direction of the hila) and the latter two decreasing. This effect of cranio-caudal level upon CT density was not demonstrated in a previous study using 8-mm slice separation [10]; however, this other study assessed the difference based upon slices selected from whole-lung spiral CT during end-expiratory and end-inspiratory apnoeas. These slices may have differing compositions from dCT slices as end-inspiratory apnoeas are typically longer than the 4 s required for lung recruitment in the saline-lavage model [4], whereas inspiratory time in our dCT imaging was only 2 s. The movement of denser slices located closer to the hila into the field-of-view of the CT scanner during inspiration would cause an increase in density of the dCT slice imaged (prior to considering the effects of an increased gas volume during inspiration). This is in keeping with the underestimation of whole-lung CT density by dCT in expiration being reduced in inspiration. Similar effects are seen with the other variables.
Limitations of this study
The main limitation of this study is that the use of whole-lung imaging during apnoeas as the gold-standard may not be appropriate and that four-dimensional imaging of the entire lung during tidal ventilation, as previously described using modification of the ECG-gating procedure used for cardiac CT [24], may be a better comparator, but such technology was unavailable for the DECT imaging in our scanner. Of course, if such technology were routinely available, it would obviate the need for using the single-slice technique to study experimental lung injury.
The finding that the single-slice technique is representative of the whole lung in uninjured and lung-lavaged pigs is not necessarily directly applicable to other ventilatory techniques (e.g. airway pressure release ventilation [25] or proning), lung injury models or human studies of ARDS. No animal model is capable of reproducing all of the key characteristics of ARDS in humans and any animal model is relevant for only limited aspects of ARDS pathophysiology in humans. In particular, in the case of our study, the anatomy of the pig thorax is such that a large volume of lung parenchyma exists between the inferior border of the heart and the diaphragm, which is suitable for use with the single-slice technique. This region contains representation of four out of the six pig lung lobes as previously characterised [26]. In the human, the heart sits upon the diaphragm and no such juxtadiaphragmatic slice exists that would be a suitable candidate to accurately represent the entire lung. In the case of normal human lungs and ARDS patients, a technique using 10 separate levels and extrapolating them to the entire lung may be more appropriate [27].
Finally, the choice of level of the slice used in dCT imaging is important, as well as the segmentation used to exclude regions outside the lung parenchyma. We pragmatically chose a region of the lung that contains the largest amount of lung within the 5-mm slice and demonstrated this was representative of the whole lung. Other studies [10, 27, 28] have assessed slices taken at different levels along the cranio-caudal axis, and thus, direct comparisons may not be appropriate. The hilar vessels and bronchi contain a large proportion of very high- and very low-density voxels, respectively, and inclusion of a small amount of these regions can dramatically affect results. Previously, it has been demonstrated that such regions comprise about 6% of lung volume segmented for dCT vs whole-lung studies [9], and thus, again, any small changes in what is chosen as comprising the segmented area or not would lead to large discrepancies between reported volumes in different studies.