Chest CT at a dose below 0.3 mSv: impact of iterative reconstruction on image quality and lung analysis

Abstract

Background

In chest computed tomography (CT), iterative reconstruction (IR) algorithms maintain diagnostic image quality (IQ) while significantly reducing the dose.

Purpose

To evaluate the impact of IR on IQ of chest CT at effective doses below 0.3 mSv.

Material and Methods

Twenty chest CT scans performed at effective dose below 0.3 mSv (CT1) were reconstructed varying three parameters: filtered back-projection and IR iDose⁴ algorithms; 512 × 512 and 768 × 768 matrices; and sharp and soft kernels, thus generating eight series per patient. The qualitative evaluation of the IQ was performed by ranking series from 1 to 8 (8 corresponding to the highest rank) which was subsequently compared to quantitative assessment of IQ by using an appropriated merit formula. Intra- and inter-reader IQ ranking reliability was also evaluated using Cohen’s kappa. Analysis of lung findings was finally compared between the best CT1 series and the reference CT (CT0).

Results

The best series in terms of qualitative and quantitative IQ was obtained using IR, 512² matrix and soft kernel. The best CT1 series detected nodules greater than 4 mm with an almost perfect match with CT0.

Conclusion

Chest CT performed at effective doses below 0.3 mSv may be used to confidently diagnose lesions greater than 4 mm using iDose⁴, soft kernel and 512 × 512 matrix.

Keywords

Reduced dose chest computed tomography image quality iterative reconstruction pulmonary nodules

Introduction

The increasing use of computed tomography (CT) examinations for medical exams is concerning because of the potential risk of radiation-induced malignancy. An increase of CT examinations by 600% has been noticed in the United States over the past 20 years. Approximately 70 million CT scans were performed in 2007 (1 –3). Iterative reconstruction (IR) algorithms, one of the latest advances in CT imaging, improve image quality (IQ) at reduced dose acquisitions by reducing noise, while maintaining spatial resolution (4 –14). If such algorithms were able to generate IQ allowing high diagnostic confidence at a dose close to that of a conventional chest X-ray, numerous clinical situations such as follow-up CT examinations could potentially be performed at these dose values. Good IQ may be defined as allowing a confident diagnosis with a noise level adapted to a specific diagnostic task.

The purpose of this study was to evaluate the impact of IR on IQ (first endpoint) and lung analysis (second endpoint) at acquisitions doses below 0.3 millisievert (mSv) corresponding to the lowest dose available on the CT scanner used, allowing the analysis of details greater than 1 cm, in accordance to ALARA principle and our experience. Qualitatively selected series with the best IQ were compared to the best quantitatively predicted series using a merit formula (15) that is frequently used in the development and testing of new reconstruction kernels and algorithms. The best series acquired at 0.3 mSv was then compared with a reference CT for lung analysis.

Material and Methods

Ethics statement

Local institutional review board approval was obtained for this retrospective study (CER-VD 375/13).

Inclusion criteria

Patients who underwent CT-guided interventions and received location scouting and/or immediate follow-up CT scans at 0.3 mSv (CT1) of thoracic lesions >1 cm with a recent (<4 months) reference chest CT (CT0) of good quality were included in this study. These procedures include transthoracic biopsy of parenchymal nodules (n = 12) and pleural masses (n = 2), pleural drainage (n = 5) including empyema (n = 4), pneumothoraces (n = 1), and a radiofrequency ablation of a pulmonary lesion (n = 1). From January 2013 to July 2013, 20 patients were selected. The mean age of the study population was 64 years (age range, 42–85 years), with six female patients and 14 male patients.

Exclusion criteria

Patients presenting a change of tumor size greater than 10 % between CT1 and CT0 or significant post-procedure complications such as alveolar hemorrhage and pneumothoraces were excluded to avoid compromising the comparison of lung lesion between CT1 and CT0. However, small pneumothoraces up to 2.5 cm pleural detachment (n = 5), localized alveolar hemorrhage (n = 6), and post-nodule biopsy without hindering the analysis of the residual pulmonary parenchyma were not excluded.

CT1 data acquisition

The CT1 was performed on a 64-slice CT scanner (Ingenuity, Philips Healthcare, Best, The Netherlands) for location scouting prior to an intervention (n = 8) or as a post-procedure control (n = 12). CT1 was performed without injection of contrast media. Acquisition was performed at 100 kV and 10 mAs, corresponding to a CTDI_vol of 0.4 mGy as measured on a 32 cm phantom. These parameters were determined optimal based on the size of the lesions related to the intervention, the non-diagnostic nature of the imaging and the existence of a recent reference diagnostic CT at our disposal. The constant of 0.017 mSv/mGy × cm was used to estimate the effective dose (16).

CT1 image reconstruction

CT1 was reconstructed with 0.8 mm-thick slices overlapped every 0.4 mm and three parameters were varied, generating a total of eight series:

CT reconstruction: standard filtered back-projection (FBP) algorithm and level 6 of iDose⁴ (Philips Health Care) (iD6), corresponding to the highest noise reduction for this IR algorithm.

Matrix size: 512 × 512 and 768 × 768 pixel elements.

Kernels: YA dedicated for spatial resolution and B designed for low contrast detectability which are both proprietary kernels of Philips Healthcare.

These eight image datasets were generated for each CT1 (B-512-FBP, B-512-iD6, B-768-FBP, B-768-iD6, YA-512-FBP, YA-512-iD6, YA-768-FBP, YA-768-iD6).

CT1 native image analysis

A blinded and randomized image analysis was performed using standard lung window settings (center = –600 Hounsfield units [HU], width = 1600 HU) under clinical viewing conditions on an EBW workstation (Philips Healthcare). The inclusion and anonymization of cases as well as the encoding of different series were performed by two of the study’s authors who did not take part in qualitative analysis.

Two board certified radiologists blinded to patient information (Reader 1, 24 years of experience in thoracic imaging, and Reader 2, 12 years of experience in radiology) interpreted all CT1 studies.

The analysis was focused only on the right lung, in order to avoid the cardiac-related kinetic artifacts observed on the left lung. The field of view (FOV) was identical for the eight series of each patient. Window width and level were fixed and non-modifiable. The best qualitative IQ, ranked from one (lowest IQ) to eight (best IQ) was based on the evaluation of the sharpness of the bronchial walls and of the pulmonary-vessel interface.

Image quality evaluation of postprocessing tools

According to the reading procedure used in our department, 2-mm thick slabs, 3-mm thick minimum intensity projection (MinIP) slabs, and 4-mm thick maximum intensity projection (MIP) slabs were assessed using a four-point scale as follows: 0 point, not interpretable with bad IQ; 1 point, affected diagnosis with poor IQ; 2 points, non-affected diagnosis with medium IQ; and 3 points, ideal for diagnostic analysis with optimal IQ.

An intra-observer analysis was performed in addition to the inter-observer analysis (Reader 1). To minimize recall bias, the two reading sessions were separated by 3 weeks for the intra-observer analysis.

Comparison with merit formula

The merit formula from ImPACT (15) was used to quantitatively predict IQ on the eight different reconstructions of each patient. Based on the relationships between IQ and dose, this formula is defined as follows:

Q_{2} = \sqrt{\frac{f_{av}^{3}}{σ 2 z_{1} {CTDI}_{vol}}}

σ = image noise, expressed by the standard deviation (SD), was measured within a 1 cm² homogeneous circular region of interest (ROI) located in the center of the ascending aorta, the shoulder muscles, and subcutaneous fat within the right sub-axillary region at the level of the carina on axial slices. The ROIs were placed at the same location on different image series for each patient. The average noise of the three anatomical regions was used for objective IQ assessment.

f_av = spatial resolution, given as (MTF₅₀ + MTF₁₀)/2, where MTF₅₀ and MTF₁₀ correspond to the spatial frequencies (in line pairs per cm) at 50% and 10% modulation transfer function (MTF) values, respectively. The MTF values were calculated using the phantom supplied by the vendor with a wire of 0.18 mm diameter and a radial MTF calculation method implemented in MATLAB®. The images were reconstructed using the same two reconstruction kernels as for the patient scans (B, YA).

z₁ = full width at half maximum (FWHM) of the imaged slice profile (z-sensitivity). We calculated the formula by using the z₁ values as stated by the manufacturer.

{CTDI}_{vol} = volumeweightedCTDI .

Assessment of the diagnostic value of the CT1 scan

The most recent CT scan of diagnostic quality (<4 months) with respect to the intervention was used as the reference CT0 to evaluate the diagnostic value of the CT1 scan. Nine exams were acquired after the administration of intravenous contrast medium (n = 9), while 11 were non-contrast enhanced (n = 11). These exams were acquired on 64-slice CT scanners: Ingenuity (n = 11) and Brilliance (n = 2) (Philips Healthcare), VCT (n = 5) and Discovery (n = 1) (GE Healthcare, Milwaukee, WI, USA), and on a 8-slice CT scanner: Lightspeed Ultra (n = 1) (GE Healthcare). All reference images were reconstructed with FBP using a lung kernel.

The best-judged reconstruction series was used to evaluate the diagnostic performance of CT1 compared with CT0. All nodules, masses, focal ground glass opacities (GGO) were recorded. Nodules were separated in two groups, >4 mm and ≤4 mm, according to the criteria of the Fleischner Society (17). The maximum number of nodules reported was limited to 10. The CT1 was analyzed first, to avoid the recall of lesions visualized on the CT0.

Statistical analysis

All data and statistical analysis were performed using Stata 12.0 software (Stata Corp., College Station, TX, USA). Continuous variables were presented as mean ± SD or mean (minimum–maximum). Inter- and intra-observer agreements for the classification of CT1 reconstructions datasets as well as for the numbering of parenchymal lung lesions (masses, nodules ≤4 mm, nodules >4 mm, and focal GGO) were assessed by computing Cohen’s kappa coefficient (k). Agreement between the image ranking qualitatively performed by the radiologists and the results of the merit formula was also estimated using this test. Agreement was interpreted using the Landis and Koch scale: k < 0 indicated no agreement; k = 0–0.20 slight agreement; k = 0.21–0.40 fair agreement; k = 0.41–0.60 moderate agreement; k = 0.61–0.80 substantial agreement; and k = 0.81–1 almost perfect agreement. Comparison of image noise between CT1 reconstructions datasets was performed using Kruskal-Wallis rank test. A P value <0.05 was considered statistically significant.

Results

Radiation doses and interval between CT1 and CT0

The mean dose-length product of CT1 was 15.72 ± 1.06 mGy × cm. Effective doses were estimated at 0.27 mSv. The mean CTDI and dose-length product of the CT0 were 6.35 ± 2.41 mGy and 244.63 ± 70.55 mGy × cm, respectively, with a mean effective dose of 4.16 mSv. The mean interval between the CT0 and the CT1 was 15.8 ± 27.68 days. Fifty-five percent of the CT0 scans were acquired on the same day as the CT1.

Image quality

All results are summarized in Table 1. Among the aforementioned eight CT1 series, the B-512-iD6 set was reported to exhibit the best qualitative IQ by the two readers with a substantial inter-observer concordance (k = 0.65, P < 0.0001) and a moderate intra-observer agreement (k = 0.47, P = 0.002). The YA-768-FBP series was found to exhibit the worst IQ by both readers, with an almost perfect inter-observer agreement (k = 0.98, P < 0.0001) and a perfect intra-observer agreement (k = 1.0, P < 0.0001). The reconstructions with iD6 and soft kernel B were always preferred to FBP and sharp kernel YA reconstructions for the choice of the best series. The matrix size 512 × 512 was considered superior to 768 × 768 (Fig. 1) in all but three cases for Reader 1 and in two cases for Reader 2 among the 20 cases. Among the four series reconstructed with the FBP algorithm, the B-512-FBP and the B-768-FBP series were the best ranked. In other words, the soft kernel B was always preferred except in one case by Reader 2.

Table 1.

Image quality assessment.

	B-512-iD6	B-768-iD6	YA-512-iD6	YA-768-iD6	B-512-FBP	B-768-FBP	YA-512-FBP	YA-768-FBP
Visual ranking*	8	7	6	5	4	3	2	1
Inter-observer agreement (k, P)	0.65, <0.0001	0.64, 0.001	0.98, 0.001	0.18, 0.05	–0.01, 0.5	–0.14, 0.8	–0.03, 0.6	1.0, <0.001
Intra-observer agreement (k, P)	0.47, 0.002	0.46, 0.007	1.0, <0.0001	0.22, 0.06	0.23, 0.04	0.04, 0.4	1.0, <0.0001	1.0, <0.001
Noise (HU)	40	46	113	166	102	117	308	457
Formula of merit
Q2	0.73	0.66	0.53	0.35	0.34	0.30	0.22	0.17
Rank*	8	7	6	5	4	3	2	1

8 being the best ranked series.

FBP, filtered back projection; HU, Hounsfield Unit, iD6, iDose 6.

Fig. 1.

Subjective image quality assessment with ranking of the eight series based on the evaluation of the sharpness of the bronchial walls and of the pulmonary-vessel interfaces. At this dose, the determining criterion of image quality was definitely the noise, that allowed to successively exclude series one at a time. (a) B-512-iD6, (b) B-768-iD6, (c) YA-512-iD6, (d) YA-768-iD6, (e) B-512-FBP, (f) B-768-FBP, (g) YA-512-FBP, (h) YA-768-FBP.

The IQ assessed by using the grading score (0–3 points) was optimal for average, MinIP, and MIP postprocessing tools in reconstructions with iD6 and kernel B, with a mean of 2.9 points for the matrix size 512 × 512 (B-512-iD6 set) and 2.8 points for the matrix size 768 × 768 (B-768-iD6 set), thus allowing for a confident level of diagnostic analysis of the scans. Regarding reconstructions using FBP, the mean grade of the average, MinIP and MIP postprocessing tools was assessed as 1.6, 1.8, and 2.1 for the B-512-FBP series and 1.6, 1.6, and 1.9 for the B-768-FBP series, respectively.

Considering all eight series together, there was a substantial inter- and intra-observer concordance (k = 0.61, P < 0.0001 and k = 0.69, P < 0.0001, respectively) for grading of IQ. Moreover, the ranking of the eight series after calculating the formula of merit perfectly agreed with the visual assessment of IQ (Table 1).

Noise measurements

The image noise was always lower (P = 0.0001) on the series B-512-iD6 compared to series B-768-iD6, regardless of the anatomic region analyzed. On average, the image noise was calculated at 40.5 and 45.6 HU, respectively, for these two series. These averaged values were significantly lower compared to those for B-512-FBP, YA-512-iD6, B-768-FBP, and YA-768-iD6 series, which had an image noise of 102.1, 113.0, 117.1, and 166.2 HU, respectively.

In accordance to the results of the qualitative IQ analysis, the two series reconstructed with FBP and the soft kernel B were on average less noisy than the YA-768-ID6 series. The YA-768-FBP series was always the noisiest one.

Assessing pulmonary lesions

The reported numbers of masses, solid nodules (>4 mm and ≤4 mm) and focal GGO are summarized in Table 2.

Table 2.

Diagnostic findings.

	Masses	Nodules >4 mm	Nodules ≤4 mm	GGO
Reference CT (CT0)
Reader 1	13	74	20	4
Reader 2	13	55	16	3
CT at a dose below 0.3 mSv (CT1)
Reader 1	13	78	20	9
Reader 2	13	62	19	4
CT0/CT1 agreement (k, P)
Reader 1	1.0, 0.0001	0.82, 0.0001	0.60, 0.0001	k = 0
Reader 2	1.0, 0.0001	0.54, 0.0057	0.39, 0.04	k = 0

CT, computed tomography; CT1, CT at a dose below 0.3 mSv; GGO, ground glass opacities.

The intra-observer agreement was perfect for the detection of masses on CT1 when compared to the CT0 (k = 1.0, P = 0.0001) for both readers (Fig. 2). The agreement was inversely proportional to size and density of lesions: for nodules >4 mm the agreement was almost perfect for Reader 1 (k = 0.82, P = 0.0001) and moderate for Reader 2 (k = 0.54, P = 0.0057), and for nodules ≤4 mm the agreement was moderate for Reader 1 (k = 0.6, P = 0.0001) and fair for Reader 2 (k = 0.39, P = 0.04). For GGO, there was no agreement for detection by both readers.

Fig. 2.

Cavitated part-solid nodule of the right lower lobe on native axial CT scan reconstructed with FBP, lung kernel, 512 × 512 matrix (a), 2-mm average slab reconstructed with FBP, soft kernel, 512 × 512 matrix (b), iDose 6, soft kernel, 512 × 512 matrix (c). The diagnostic quality may be considered as comparable between (b) and (c), despite less spatial resolution in (b). Note in particular the correct assessment of ground glass opacity at the lateral part of the nodule, despite the relative change of overall opacity at the same viewing parameters that could be related with the thickness of the slab in (b).

The inter-observer agreement was perfect for detection of masses on both CT1 and CT0 (k = 1.0, P = 0.0001). For nodules, the inter-observer agreement was perfect on CT1 (k = 1.0, P = 0.0001) and substantial on the CT0 (k = 0.76, P = 0.0011). Like the intra-observer agreement, the inter-observer agreement was lower for nodules ≤4 mm, being moderate on both CT1 and CT0 (k = 0.42, P = 0.0041 and k = 0.41, P = 0.0006, respectively), and there was also an inter-observer disagreement for GGO.

Discussion

In order to confirm the confidence in 0.3 mSv lung scans, we felt compelled to assess the resulting IQ. Such a dose is close to that of a conventional chest X-ray which is in the range of 0.03–0.1 mSv (18 –20).

Results of our study confirm the overall significant added value of IR over FBP reconstructions at a dose below 0.3 mSv, image noise being the most critical parameter determining IQ at such dose value. Interestingly, the matrix size 512 × 512 was visually preferred to 768 × 768. This suggests that the potential improved spatial resolution resulting from the 768 × 768 matrix was counterbalanced by the increased image noise at these sub-mSv doses. For that same reason, the soft kernel B was always preferred to the YA, despite its lower spatial resolution (21,22).

Our results also show that when IR algorithms are not available, FBP can benefit from the combined use of a soft kernel and average intensity projection tool for lung parenchyma analysis with an acceptable IQ that does not compromise the diagnostic confidence.

The perfect concordance of the ranking of the eight series between the visual assessment of IQ and the merit formula is interesting. Evaluation of IQ is in fact usually partly considered subjective, with each component of IQ, mainly spatial resolution versus noise, being difficult to strictly separate from each other. This appears to demonstrate that the human brain performs a synthesis of all these factors, downgrading IQ when one factor is unacceptable, most notably the noise.

This study shows, not surprisingly, a decrease in diagnostic performance of chest CT at a dose below 0.3 mSv in an inversely proportional relationship with the size and contrast of pulmonary lesions. The analysis of CT1 of Reader 1, a radiologist specialized in thoracic imaging, was more consistent with her CT0 analysis than the analysis of Reader 2, a radiologist specialized in abdominal imaging. Furthermore, the inter-observer agreement was almost equal for the CT1 and the CT0 scans. The inter-observer agreement for detection of nodules >4 mm was perfect at CT1 with better diagnostic performance than the CT0.

There were several limitations in this study. First, because patients were selected exclusively from the interventional imaging department, some lung entities such as interstitial lung disease were not represented. Second, some CT0 images were acquired within the month before or after the procedure. It could be argued that some lesions could have evolved thus changing their configuration and size. This was not the case when the two acquisitions were performed in the same day, as was done for more than half of the cases in our study. However, this procedure was judged as the most ideal for the patient because a routine-level dose CT followed by a reduced dose CT would have been unacceptable. To minimize this bias, we excluded cases with significant evolution or complications between the CT0 and the CT1. Third, the CT0 exams were acquired on several different machines, but all considered of diagnostic quality. Reconstruction parameters, such as slice thickness and kernel, were identical on both the 8-slice and 64-slice CT scanners, generating comparable IQ. Finally, this study only compares the diagnostic performance of chest CT at a dose below 0.3 mSv for analysis of the lung parenchyma, where lesions have a high intrinsic contrast relative to the surrounding pulmonary tissue. Our results need to be further evaluated using other thoracic pathologies and clinical situations along with a larger sample size.

In conclusion, chest CT performed at an effective dose below 0.3 mSv, close to that of a chest X-ray, achieves good IQ by using iDose⁴, soft kernel and 512 × 512 matrix. As compared to standard dose CT, these reconstruction parameters allow diagnosis of solid parenchymal disorders greater than 4 mm, and may thus be used in specific conditions such as post-interventional or parenchymal nodule >4 mm follow-up CT imaging.

Footnotes

Acknowledgments

The authors would like to thank Steven Hajdu, Alban Denys, Boris Guiu, and Brice Coisnon (Department of Radiology, Lausanne University Hospital, Rue du Bugnon 46, CH-1011 Lausanne, Switzerland) for CT1 acquisitions, and Frédéric Mieville (Department of Radio-oncology, Hospital Fribourg, Chemin des Pensionnats 26, CH-1708 Fribourg, Switzerland) for the MTF calculation method implemented in MATLAB®.

Conflict of interest

Three authors (Susanne Heinzer, Stéphane Montandon, and Andrei Feldman) are Philips employees.

Funding

This research received no specific grant from any funding agency in the public, commercial, or not-for-profit sectors.

References

Kuo

Ciet

Tiddens

. Monitoring cystic fibrosis lung disease by computed tomography. Radiation risk in perspective. Am J Respir Crit Care Med 2014; 189: 1328–1336.

Willemink

de Jong

Leiner

. Iterative reconstruction techniques for computed tomography, part 1: technical principles. Eur Radiol 2013; 23: 1623–1631.

Singh

Khawaja

Poujabbar

. Iterative image reconstruction and its role in cardiothoracic computed tomography. J Thorac Imaging 2013; 28: 355–367.

Vardhanabhuti

Loader

Mitchell

. Image quality assessment of standard- and low-dose chest CT using filtered back projection, adaptive statistical iterative reconstruction, and novel model-based iterative reconstruction algorithms. Am J Roentgenol 2013; 200: 545–552.

Kalra

Woisetschläger

Dahlström

. Sinogram-affirmed iterative reconstruction of low-dose chest CT: effect on image quality and radiation dose. Am J Roentgenol 2013; 201: 235–244.

Baumueller

Winklehner

Karlo

. Low-dose CT of the lung: potential value of iterative reconstructions. Eur Radiol 2012; 22: 2597–2606.

Pontana

Pagniez

Flohr

. Chest computed tomography using iterative reconstruction vs filtered back projection (Part 1): evaluation of image noise reduction in 32 patients. Eur Radiol 2011; 21: 627–635.

Pontana

Duhamel

Pagniez

. Chest computed tomography using iterative reconstruction vs filtered back projection (Part 2): image quality of low-dose CT examinations in 80 patients. Eur Radiol 2011; 21: 636–643.

Singh

Kalra

Gilman

. Adaptive statistical iterative reconstruction technique for radiation dose reduction in chest CT: a pilot study 1. Radiology 2011; 259: 565–573.

10.

Willemink

Leiner

de Jong

. Iterative reconstruction techniques for computed tomography, part 2: initial results in dose reduction and image quality. Eur Radiol 2013; 23: 1632–1642.

11.

Neroladaki

Botsikas

Boudabbous

. Computed tomography of the chest with model-based iterative reconstruction using a radiation exposure similar to chest X-ray examination: preliminary observations. Eur Radiol 2013; 23: 360–366.

12.

Kligerman

White

. Image quality and feasibility of an ultralow-dose high-pitch helical triple-rule-out computed tomography angiography acquired in the caudocranial direction. J Thorac Imaging 2014; 29: 50–59.

13.

Khawaja

Singh

Gilman

. Computed tomography (CT) of the chest at less than 1 mSv: an ongoing prospective clinical trial of chest CT at submillisievert radiation doses with iterative model image reconstruction and iDose4 technique. J Comput Assist Tomogr 2014; 38: 613–619.

14.

Katsura

Matsuda

Akahane

. Model-based iterative reconstruction technique for ultralow-dose chest CT. Invest Radiol 2013; 48: 206–212.

15.

ImPACT – Imaging Performance Assessment of CT Scanners. 2002. Four slice CT scanner comparison report. Available at: http://www.impactscan.org/reports/full/MDA02021.pdf (accessed 24 October 2014).

16.

European guidelines on quality criteria for computed tomography. EUR 16262. Available at: http://w3.tue.nl/fileadmin/sbd/Documenten/Leergang/BSM/European_Guidelines_Quality_Criteria_Computed_Tomography_Eur_16252.pdf (accessed 24 October 2014).

17.

MacMahon

Austin

Gamsu

. Guidelines for management of small pulmonary nodules detected on CT scans: a statement from the Fleischner Society. Radiology 2005; 237: 395–400.

18.

American College of Radiology. ACR Appropriateness Criteria, radiation dose assessment introduction. 2013. Available at: http://www.acr.org/∼/media/a27a29133302408bb86888eafd460a1f.pdf (accessed 24 October 2014).

19.

European Commission. Referral guidelines for imaging, radiation protection 118. 2008. Available at: http://ec.europa.eu/energy/nuclear/radioprotection/publication/doc/118_update_en.pdf (accessed 24 October 2014).

20.

American College of Radiology. ACR-AAPM practice guidelines for diagnostic reference levels and achievable dose in medical X-rays imaging. 2013. Available at: http://www.acr.org/∼/media/796DE35AA407447DB81CEB5612B4553D.pdf (accessed 24 October 2014).

21.

Von Falck

Galanski

Shin

. Informatics in radiology: sliding-thin-slab averaging for improved depiction of low-contrast lesions with radiation dose savings at thin-section CT. Radiographics 2010; 30: 317–326.

22.

Beigelman-Aubry

Hill

Guibal

. Multi-detector row CT and postprocessing techniques in the assessment of diffuse lung disease. Radiographics 2005; 25: 1639–1652.