Introduction

Chest computed tomography (CT) has substantially changed the understanding and management of patients with acute respiratory distress syndrome (ARDS) [1–4] as it allows both a diagnostic evaluation (i.e., presence of alveolar consolidation, ground glass opacification, and reticular pattern) and a quantitative analysis (i.e., assessment of lung recruitability, opening and closing, and hyperinflation) of injured lungs [5, 6]. Of note, no other imaging technique currently applied provides similar information [2, 5, 7].

The serial use of CT in patients with ARDS is, however, not free of risks as it (1) requires the transportation of critically ill patients to the radiology department [8] and (2) exposes the patients to a certain amount of ionizing radiation that could be associated with an increased incidence of cancer [9–11]. The dose of radiation applied during CT image acquisition is directly related to image noise (the higher the dose, the lower the image noise) which influences image quality (the higher the noise, the lower the image quality). Despite its widespread use, no consensus for optimal CT acquisition is currently available and a great variability of CT parameters is reported in the literature, with ranges of peak tube voltage (kVp) and tube current–time products (mAs) of 120–140 kVp and 100–350 mAs, respectively.

The use of low-dose chest CT was first described in the 1990s by Naidich and colleagues [12]. In their paper, the authors describe the possibility to visualize lung parenchyma despite a reduction of mAs from 140 to 10. Importantly, the study was performed both in patients with healthy lungs, i.e., patients in which the intrinsic high contrast of the lung was preserved, and in patients with diffuse lung disease, i.e., patients in which the intrinsic high contrast of the lung was partially lost.

Thereafter, several studies performed on lung cancer screening showed an equivalent visualization of anatomical and pathological structures between CTs performed with standard and low doses [13–17].

As mentioned above, besides the visual diagnostic evaluation, chest CT also allows one to perform a quantitative analysis (qCT) [18]. Through the measurement of lung recruitability, alveolar collapse, lung hyperinflation, and the amount of pleural effusion, lung qCT provides useful information to optimize mechanical ventilation in patients with ARDS, thus potentially reducing its harmful effects [5, 19–24].

Lung qCT performed on low-dose CT was found to be accurate for the measurement of hyperinflated lung tissue in patients with lung emphysema [25, 26]. Furthermore, a recent experimental study performed in an ovine model of ARDS demonstrated that a dose reduction of up to 70 % did not affect the accuracy of lung qCT [27].

It is, however, important to underline that a time-consuming manual drawing of CT images is required and dedicated software needs to be used in order to perform lung qCT. These aspects greatly limit the clinical applicability of lung qCT [28]. Recently, an alternative to qCT to estimate lung recruitability has been described, i.e., the visual anatomical analysis [29]. This approach does not require dedicated software, significantly shortens image processing time, and has been found to be sufficiently accurate in discriminating between patients with higher and lower potential for lung recruitment [29].

The aim of the present study was therefore to evaluate (1) the accuracy of qCT analysis and (2) the accuracy of visual anatomical analysis of low-dose chest CT scans performed on patients with ARDS.

We hypothesized that low-dose CT would provide accurate quantitative and visual anatomical results in patients with ARDS.

Materials and methods

Study population

The study was approved by the institutional review board of our hospital (no. 3102/2012) and was registered at clinicaltrials.gov (NCT01926093). Informed consent was obtained in accordance with Italian national regulations. Patients admitted to the intensive care unit (ICU) of our institution, fulfilling the criteria for ARDS [30, 31] and requiring chest CT for clinical indications were studied. Exclusion criteria were age less than 16 years, pregnancy, and documented barotrauma.

Study design, CT image acquisition, and reconstruction

Clinical characteristics, respiratory variables, and ventilatory settings of the patients were recorded in the ICU, before the study. All patients were sedated, paralyzed, and maintained in supine position.

Subsequently, patients were transported to the radiological department. Immediately before each chest CT scan (Siemens Somatom Definition Flash, Siemens Healthcare, Forchheim, Germany), a recruitment maneuver was performed as previously described [21]. Thereafter, the endotracheal tube was clamped during a respiratory hold performed with the ventilator (Servo 900; Siemens, Solna, Sweden) and two consecutive scans of the whole lung were performed during the same breath hold. For each couple of scans two different mAs were applied: in the first 24 patients 110 mAs was coupled with 60 mAs; in the following 21 patients 110 mAs was coupled with 30 mAs. All other parameters of the CT scanner were kept unaltered during image acquisition (tube voltage 120 kVp, rotation time 0.5 s, collimation 128 × 0.6 mm, pitch 0.85, reconstruction matrix 512 × 512). Contrast material was not used during image acquisition. An automatic tube current modulation technique (Care Dose 4, Siemens, Medical Solutions) allowing for a dynamic reduction of dose radiation during CT examination was applied [32–34]. In each patient, two sets of two consecutive chest CT scans were performed either at airway pressures of 45 and 15 or at airway pressures of 45 and 5 cmH2O, according to clinical indication. Images were reconstructed using a 5-mm section width, a 5-mm interval, and a medium smooth + recon kernel (B31f).

Quantitative lung CT analysis

Lung boundaries were manually drawn in each CT slice using a dedicated image analysis software (Maluna 3.15, Göttingen, Germany). Mediastinal structures and pleural effusions were excluded from the region of interest [21]. Thereafter, total lung volume and lung weight were computed. Based on the degree of aeration/density of lung tissue, four compartments were computed: hyperinflated tissue (−1,000 to −901 Hounsfield units [HU]), normally aerated tissue (−900 to −501 HU), poorly aerated tissue (−500 to −101 HU), and non-aerated tissue (−100 to +500 HU) [5, 27].

Lung recruitability was computed, as previously described [21], as the ratio between the difference in weight of not inflated lung tissue passing from 5 to 45 cmH2O of airway pressure and total lung weight measured at 5 cmH2O of airway pressure.

Visual anatomical lung CT scan analysis

Two radiologists, unaware of the quantitative results, analyzed all CT scans using a Dicom image viewer (Efilm, workstation, Canada). Briefly, for each bronchopulmonary segment, the percentage of collapsed/consolidated tissue, defined as a zone of increased attenuation in which the vessels and the airway walls are not recognizable, was evaluated [5, 35]. The obtained percentage of tissue was than multiplied by the estimated percentage of volume of the corresponding bronchopulmonary segment with respect to total lung volume [29]. Bronchopulmonary segments were identified as previously described [29, 36]. The total percentage of consolidated/collapsed lung tissue was obtained as the sum of the individual percentages. Lung recruitability was computed as the difference in the percentage of consolidated/collapsed tissue between scans performed at 5 and scans performed at 45 cmH2O of airway pressure [29]. In addition, the presence/absence of pleural effusion was assessed in each patient.

Except for the assessment of correlation between radiologists (see below), the average of the results of the two radiologists was used for analysis.

Radiation dose and image noise

Effective mAs and dose–length product (DLP) of each CT scan were provided by the scanner. Effective dose was computed using the DLP method, as previously described [27, 37]. Image noise levels for each applied mAs were calculated as the mean standard deviation (SD) of tissue density in a given, uniform region of interest (within the thoracic aorta) [27, 38].

Statistical analysis

Data are expressed as mean ± SD or as median and interquartile range in the case of variables not normally distributed. The agreement between results obtained at standard (110 mAs) and lower doses (60 and 30 mAs) was assessed using Bland–Altman analysis [39] and linear regression. The percentage error was calculated as limits of agreement divided by the mean value measured with the standard dose of 110 mAs [40, 41]. The agreement between lung recruitability estimated by different radiologists was assessed using the concordance correlation coefficient (CCC) [42]. One-way analysis of variance was used to compare effective mAs, DLP, effective dose, and image noise of the different applied mAs. Statistical significance was defined as p < 0.05. Analysis was performed with SAS 9.2 (SAS Institute Inc, Cary, North Carolina, USA) and SigmaPlot 11.0 (Systat, Chicago, IL, USA).

Results

The main characteristics of the whole population (45 patients) are summarized in Table 1. No significant differences were observed between patients of group 1 (110–60 mAs) and patients of group 2 (110–30 mAs) (see Table E1 of the Electronic Supplementary Material).

Table 1 Baseline characteristics of the whole study population
Full size table

A total of 180 chest CT scans, i.e., 90 comparisons, were performed. Six CT scans, i.e., three comparisons (one in the 110–60 mAs group, two in the 110–30 mAs group), were excluded because of large differences (>15 %) in lung volume computed with qCT between coupled scans, likely due to the presence of air leakage during image acquisition, leaving 47 comparisons between 110 and 60 mAs (performed in 24 patients) and 40 comparisons between 110 and 30 mAs (performed in 21 patients). Visual anatomical analysis was performed in 32 patients in which CT images were acquired at 45 and 5 cmH2O of airway pressure (15 comparisons between 110 and 60 mAs and 17 comparisons between 110 and 30 mAs; patients characteristics are reported in Table E2 of the Electronic Supplementary Material).

Quantitative lung CT analysis

Overall, good agreement was observed between quantitative results obtained at 110 and 60 mAs and between quantitative results obtained at 110 and 30 mAs (Table 2), regardless of the presence or absence of pleural effusion (Fig. E1 of the Electronic Supplementary Material). Lung weight computed at 110 mAs was highly correlated with lung weight computed both at 60 mAs (r 2 = 0.99, p < 0.001) and at 30 mAs (r 2 = 0.99, p < 0.001). Bias for the comparisons of the measurement of lung weight obtained at 110 and 60 mAs was 14.7 g (0.7 %) with limits of agreement (LOA) of −62.7 to 92.1 g (−3.4 to 4.8 %) and a percentage error of 10 % (Fig. 1, upper panel). Bias of lung weight measurement obtained at 110 and 30 mAs was −3.7 g (−0.3 %) with LOA of 85.7–93.0 g (−6.4 to 5.8 %) and a percentage error of 11 % (Fig. 1, lower panel). Small biases (all <1 %) with tight LOA were observed also for the quantification of differently aerated lung compartments (Table 2). Additional Bland–Altman plots, linear regressions, and frequency distributions of CT numbers are reported in the Electronic Supplementary Material (Figs. E3–E7).

Table 2 Comparisons between quantitative results obtained at 110 and 60 mAs and between quantitative results obtained at 110 and 30 mAs
Full size table
Fig. 1
figure 1

Bland–Altman analysis of lung weight computed with the quantitative analysis on CTs performed at 110 and 60 mAs (upper panel) and 110 and 30 mAs (lower panel). X -axis represents the mean of the two measurements,Y -axis represents the difference between lung weight (expressed as percentage of total lung weight) computed at 110 and 60 mAs (upper panel) and 110 and 30 mAs (lower panel). Bias is represented as the horizontal solid line; standard errors of the bias are represented as horizontal dotted lines; limits of agreement (±1.96 SD) are represented as horizontal dashed lines. Upper panel slope = 0.0, r 2 = 0.05, p = 0.13; lower panel slope = 0.00, r 2 = 0.00, p = 0.99

Full size image

Visual anatomical lung CT scan analysis

The overall CCC between the two radiologists for estimating the amount of consolidated/collapsed lung tissue and lung recruitability was 0.96 (95 % confidence interval 0.95–0.97) and 0.95 (95 % confidence interval 0.94–0.97), respectively.

A high correlation was observed between lung recruitability measured visually on CT scans performed at standard dose (110 mAs) and both 60 and 30 mAs (r 2 = 0.97, p < 0.001 for both correlations). Related biases and LOA were −0.1 % (−5.9 to 5.6 %) between 110 and 60 mAs (see Fig. 2, upper panel) and −1.5 % (−6.6 to 3.6 %) between 110 and 30 mAs (see Fig. 2, lower panel). Biases and LOA for the measurement of collapsed/consolidated lung tissue between 110 mAs and 60 or 30 mAs are summarized in Table 3.

Fig. 2
figure 2

Bland–Altman analysis of lung recruitability estimated with the visual anatomical analysis computed on CTs performed at 110 and 60 mAs (upper panel) and 110 and 30 mAs (lower panel). X-axis represents the mean of the two measurements, Y-axis represents the difference between lung recruitability computed at 110 and 60 mAs (upper panel) and 110 and 30 mAs (lower panel). Bias is represented as the horizontal solid line; standard errors of the bias are represented as horizontal dotted lines; limits of agreement (±1.96 SD) are represented as horizontal dashed lines. Upper panel slope = −0.10, r 2 = 0.25, p = 0.06; lower panel slope = −0.10, r 2 = 0.27, p = 0.05

Full size image
Table 3 Comparisons between visual anatomical analysis results obtained at 110 and 60 mAs and between visual anatomical analysis results obtained at 110 and 30 mAs
Full size table

Dose radiation and image noise

Effective mAs applied during image acquisition were 111 ± 32, 58 ± 17, and 31 ± 8 which differed slightly from the set values (110, 60, and 30 mAs) because of the applied tube current modulation technique. Related DLP was 257 ± 76, 136 ± 41, and 70 ± 16 mGy cm−1 for 110, 60, and 30 mAs, respectively (p < 0.001). Lowering the applied mAs from 110 to 60 and 30 mAs progressively reduced the effective dose from 5.3 ± 1.6 to 2.8 ± 0.8 and 1.4 ± 0.3 mSv (p < 0.001). These variations corresponded, compared to the standard dose of 5.3 mSv recorded at 110 mAs, to a reduction in effective dose of 47 and 74 %, respectively. Finally, image noise significantly and progressively increased when reducing mAs from 110 to 60 and 30 (8.7 ± 3.9 HU for 110 mAs, 10.6 ± 4.8 HU for 60 mAs, and 18.4 ± 8.2 HU for 30 mAs, p < 0.001).

Discussion

The primary findings of this study performed in patients with ARDS are (1) the reduction of the applied tube current–time product up to 30 mAs (approximately 70 % effective dose reduction) did not affect the accuracy of quantitative lung CT analysis; (2) visual anatomical analysis of collapsed/consolidated lung tissue and lung recruitability performed on low-dose CT is sufficiently accurate. The application of lower doses during CT image acquisition could reduce the risks associated with radiation exposure and therefore allow a more frequent use of chest CT in patients with ARDS, in order to evaluate the time course of lung disease and potentially tailor mechanical ventilation.

Quantitative lung CT analysis

The accuracy of lung qCT performed on low-dose CT has been recently demonstrated in an ovine model of oleic acid-induced ARDS [27]. To investigate its accuracy in a population of patients with ARDS, we compared lung qCT results obtained at a standard dose of 110 mAs with two progressively lower doses in 45 patients admitted to our general ICU. Initially, 110 mAs was compared with 60 mAs and a good agreement was observed (Table 2). Thereafter, given the proven accuracy at 60 mAs, a lower dose (30 mAs) was tested and similar results were found (Table 2).

As previously discussed [27], by lowering the applied mAs during CT image acquisition in ARDS patients, two effects could be expected. First, the reduction in mAs would cause an increase in image noise with the related effects on the measurement of lung densitometry [27, 43] (Figs. E2 and E7 of the Electronic Supplementary Material); second, as a result of the worsening of image quality and in light of the similarity of densities of mediastinal structures and collapsed/consolidated lung parenchyma, the ability of the operators to recognize the lung boundaries and exclude thoracic structures and pleural effusions could be jeopardized. This fact could in turn potentially lead to macroscopic variations in the manual drawing of the regions of interest and could therefore significantly affect quantitative results.

We observed a very good agreement with tight LOA and low percentage errors [40] between quantitative results obtained on CT images acquired at a standard dose (110 mAs) and those obtained at progressively lower doses (60 and 30 mAs). Therefore, the overall effects of image noise on lung densitometry (Fig. E7 of the Electronic Supplementary Material) were negligible and the increased image noise did not hinder the proper recognition of lung and mediastinal structures, i.e., manual drawing was not affected. Interestingly, similar results were found regardless of the presence or absence of pleural effusion (Figs. E1 and E9 of the Electronic Supplementary Material). This subgroup analysis was performed, as the presence of pleural effusion could have worsened the ability to recognize the lung boundaries on CT images acquired with lower doses.

Visual anatomical lung CT analysis

The ‘visual anatomical analysis’ was developed by our group in an attempt to simplify and shorten the time required to estimate lung recruitability in patients with ARDS and discriminate between patients with higher and lower potential for lung recruitment [29]. Until now, we had performed the visual anatomical analysis only on standard-dose CTs. Although the LOA of the present study might appear quite high (Table 3) and a trend toward underestimation was observed for lower doses (60 and 30 mAs) in patients with a high amount of recruitable lung (Fig. 2), these results are similar to those previously reported [29], suggesting that equivalent results can be obtained despite a reduction of tube current–time product up to 30 mAs. This implies, that the increased image noise associated with the dose reduction did not impair the ability of the two radiologists to recognize and evaluate the amount of collapsed/consolidated lung tissue.

Radiation dose

When dealing with the exposure to ionizing radiation the ALARA concept (as low as reasonably achievable) should always be kept in mind. As mentioned above, no standardized protocol is currently available for CT acquisition and widely variable mAs have been reported. In the present study, the effective dose was reduced by approximately 70 % without affecting quantitative and visual anatomical results significantly. On the basis of our results, we might therefore strongly suggest the use of low-dose chest CT if the aim is to perform a quantitative or a visual anatomical analysis. Furthermore, effective dose could be further reduced by coupling low-dose CT with the method based on the extrapolation of whole lung results from ten CT slices [28]. Finally, the use of new techniques based on iterative reconstruction algorithms could lower image noise and therefore allow a further dose reduction [44].

Limitations

A limitation of our study that requires discussion is the fact that doses lower than 30 mAs were not investigated. Indeed, as a worsening of the accuracy has been previously described for dose reductions beyond 70 % [27], we chose not to explore these mAs values. However, it might be of interest to investigate, in future studies, the use of iterative reconstruction algorithms as it could allow a further dose reduction. Furthermore, we need to mention the small sample size included in the visual anatomical analysis (15 patients in the 110–60 mAs and 17 patients in the 110–30 mAs group) that possibly contributed to the observed large limits of agreement.

Conclusions

In patients with ARDS the reduction of tube current–time product up to 30 mAs (70 % reduction in effective dose) during CT image acquisition does not affect the accuracy of quantitative and visual analysis results. Low-dose CT, with the related quantitative and visual anatomical analysis, is a valuable tool to characterize and potentially monitor lung disease in patients with ARDS.