Radiol med DOI 10.1007/s11547-014-0382-3

RADIOBIOLOGY AND SAFETY

Assessment of dose exposure and image quality in coronary angiography performed by 640-slice CT: a comparison between adaptive iterative and filtered back-projection algorithm by propensity analysis Ernesto Di Cesare • Antonio Gennarelli • Alessandra Di Sibio • Valentina Felli • Alessandra Splendiani • Giovanni Luca Gravina Antonio Barile • Carlo Masciocchi

•

Received: 9 August 2013 / Accepted: 1 October 2013 Ó Italian Society of Medical Radiology 2014

Abstract Purpose This study was performed to confirm, by propensity score matching, whether the use of adaptive–iterative dose reduction (AIDR 3D) with a built-in automatic exposure control system provides clinical and dosimetric advantages with respect to the traditional filtered backprojection (FBP) algorithm without automatic exposure modulation. Materials and methods A total of 200 consecutive patients undergoing coronary computed tomography (CT) angiography on a 640-slice CT scanner were studied. A protocol with exposure parameters based on patient body mass index (BMI) and with images reconstructed using FBP (group A) was compared with a protocol with images acquired using tube current decided by an automatic exposure control system and reconstructed using AIDR (group B). Mean effective dose and image quality with both objective and subjective measurements were assessed. Results Mean effective dose was 23.6 % lower in group B than in group A (2.56 versus 3.34 mSv; p \ 0.0001).

E. Di Cesare (&) Division of Cardiac Radiology, Department of Biotechnological and Applied Clinical Sciences, University of L’Aquila, Via Vetoio 1, 67100 L’Aquila, Italy e-mail: [email protected] A. Gennarelli
A. Di Sibio
V. Felli
A. Splendiani
A. Barile
C. Masciocchi Division of Radiology, Department of Biotechnological and Applied Clinical Sciences, University of L’Aquila, L’Aquila, Italy G. L. Gravina Laboratory of Radiobiology, Division of Radiotherapy, Department of Biotechnological and Applied Clinical Sciences, University of L’Aquila, Via Vetoio 1, 67100 L’Aquila, Italy

Noise was significantly lower in group B with consequent higher signal-to-noise (SNR) and contrast-to-noise (CNR) (p \ 0.0001) compared with group A. Subjective quality parameters were also significantly higher in group B. Conclusions Comparative analysis by propensity score matching confirms that AIDR 3D with automatic exposure control is able to reduce significantly the mean radiation dose and improve the image quality compared with traditional FBP without exposure modulation. Keywords Coronary CT angiography
Adaptive– iterative reconstruction
Image noise
Radiation dose
Image quality

Introduction Coronary artery disease (CAD) is the leading cause of death in developed countries [1, 2], and cardiac catheterisation remains the standard of reference for the evaluation of the coronary arteries. Coronarography comes at a considerable cost and, although complications may be infrequent, it accounts for well-known procedure-related morbidity [3]. Several studies have demonstrated that coronary computed tomography (CT) is an effective alternative to coronarography for detecting significant coronary stenosis in patients with suspected CAD [4]. However, coronary CT using 64-slice scanners, the most used technology for coronary evaluation, has relatively high radiation exposure (median 12 mSv, range 8–16 mSv) [5], and the cancer risk associated with radiation exposure with this technique [2] is worrisome. As a result, all major scanner manufacturers have developed different systems to reduce radiation exposure, such as electrocardiogram-based tube current modulation for

123

Radiol med

helical examinations, prospective gating, noise reduction filters, high-pitch spiral acquisition, reduction of tube voltage and reduction of z-axis scan length [6]. These methods provide significant reductions in radiation dose and have been reported to result in similar image quality in multicentre studies [6]. Three recent reports [7–9] have indicated that adaptive– iterative dose reduction (AIDR 3D) with built-in automatic exposure control (SureExposure 3D) has better performance than traditional filtered back-projection (FBP) without automatic dose modulation. However, the main weakness of these reports was related to the inherent nature of a prospective observational study. In order to mitigate methodological biases, the image quality and mean radiation doses of AIDR 3D with SureExposure 3D were compared with traditional FBP without automatic dose modulation using a 640-slice CT scanner and a one-beat prospective acquisition by propensity score matching. This powerful statistical approach, attempting to reduce the bias due to confounding variables that could be found in an estimate of the outcome measure obtained from simply comparing outcomes among units belonging to different study groups, simulates a randomisation process by creating an absolute comparability of clinical characteristics among groups. Thus, the aim of this study was to confirm, using this powerful statistical approach, whether the use of AIDR 3D with SureExposure 3D really provides clinical and dosimetric advantages with respect to traditional FBP without dose modulation.

Materials and methods Patients From October 2011 to December 2012, 272 consecutive patients scheduled for coronary CT were prospectively studied with a 640-slice CT scanner (Aquilion One, Toshiba Medical Systems). Inclusion criteria were atypical chest pain and/or dyspnoea or equivocal or inconclusive stress test. Exclusion criteria were impaired renal function (serum creatinine [1.5 mg/dl), previous coronary artery interventions, heart rate [65 beats per minute (bpm) after beta-blocker premedication and severe calcification of the major coronary arteries. Patients were divided into two groups based on the reconstruction algorithm and method of tube current selection. A protocol with exposure parameters based on patient body mass index (BMI) (Table 1) and with images reconstructed using FBP (group A) was compared with a protocol with images acquired using tube current decided by the automatic exposure control system (SureExposure) and reconstructed using AIDR 3D (group B; 69 males and 29 females, mean age

123

Table 1 Computed tomography coronary angiography (CCTA) with body mass index (BMI)-based parameters used in group A

BMI

mA

100 kV 13

300

15

350

17

400

19

450

21

500

23

550

120 kV 23

400

25

450

30

500

35–40

580

135 kV 40?

510

62.2 ± 9.5 years; mean BMI 26.2 ± 3.5). Images were evaluated for the presence of significant stenosis, which was defined as a narrowing of the coronary lumen equal to or exceeding 50 % of the cross-sectional area. Institutional review board approval was obtained and all patients provided informed consent before the procedures. Acquisition protocol All CT scans were acquired using a 640-slice CT scanner. One hour before imaging, the patients’ blood pressure and heart rate were recorded. Patients with a pre-scan heart rate of 65 bpm or higher were given 5–20 mg of intravenous atenolol (Tenormin, AstraZeneca, Sweden). Just before the acquisition, 0.3 mg of sublingual nitroglycerin (isosorbide dinitrate, Carvasin 5 mg, Wyeth Lederle) was administered to all patients [10]. Sixty millilitres of nonionic contrast medium (Iomeron 400, Bracco Diagnostic, Milan, Italy) was injected into the antecubital vein at 5 ml/s, followed by 40 ml of saline solution at the same flow rate, using a dual power injector (Stellant, Medrad, Indianola, PA, USA). The timing to start the CT acquisition was determined using bolus tracking (Sure Start) in the descending aorta with a scan start threshold of 300 Hounsfield units (HU). Images were obtained during inspiratory breath holding. Gantry rotation time was 350 ms, with the best temporal resolution of 175 ms. Mid-diastolic prospective scanning with an electrocardiographically (ECG) gated window of 70–80 % of the R–R interval was performed. A medium field of view (FOV) was selected to allow correct visualisation of all cardiac structures in each patient. A FOV up to 16 cm in the z-axis was obtained, covered by 320 detector rows of 0.5 mm each. Reconstruction was performed with data from a single heartbeat. For each patient, the phase with the least artefacts was automatically

Radiol med

determined by the system. A motion map of the sinogram was traced by the system and the phase with the least movement was chosen (PhaseXact, Toshiba Medical Systems, Tochigi, Japan). The maximum number of reconstructed slices was 640 with 0.5 mm thickness and 0.25 mm interval, obtained by means of the proprietary double-slice technique and cone-beam reconstruction algorithm (ConeXact, Toshiba Medical Systems, Tochikiken, Japan). Reconstruction was performed with data from a single heartbeat. Subjective evaluation of image quality Subjective evaluation of the images was carried out by two experienced cardiac radiologists. The readers were blinded to the details of the CT datasets and the images were analysed independently and in a random fashion. Any discordant final results were reviewed in consensus. Issues such as motion and poor gating were not taken into account in the subjective assessment, as they could not be ascribed to the reconstruction algorithm. The coronary arteries were subdivided into 15 segments according to the American Heart Association classification [11]. The intermediate artery, when present, was designated as

segment 16 [7]. The 15 coronary artery segments were categorised into three segmental classes as follows: (1) Proximal [proximal right coronary artery (RCA), left main coronary artery (LM), proximal left circumflex artery (LCx), proximal left anterior descending artery (LAD)]; (2) Mid [mid RCA, distal RCA, ramus intermedius (R. intermedius), obtuse marginalis (OM), first diagonal branch (D1), mid LAD]; and (3) Distal segmental classes [posterior descending artery (PDA), distal LCx, second diagonal branch (D2), distal LAD] [8]. Mean scores were calculated for each segmental class as well as for all of the three segmental classes as a whole [8]. All evaluable coronary artery segments were identified and analysed using a 5-point scale as follows: 0 (nondiagnostic) = impaired image quality with excessive image noise; 1 (adequate) = evident limitations in vessel wall definition and contrast resolution with severe image noise; 2 (good) = minimal limitations in vessel wall definition and contrast resolution with moderate image noise; 3 (very good) = good attenuation of the vessel lumen and wellmaintained vessel wall definition with minimal image noise and; 4 (excellent) = excellent attenuation of the vessel lumen and clear vessel wall definition with barely perceived image noise [10, 12, 13].

Fig. 1 a, b Measurement of signal and noise in the ascending aorta, above the coronary ostia, in two patients with the same body mass index of 26. a, c Computed tomography (CT) image obtained with filtered backprojection (FBP) without an automatic exposure control system (SureExposure 3D). Noise: 29.59. b, d CT image obtained with adaptive-iterative dose reduction 3D (AIDR 3D) with built-in SureExposure 3D. Noise: 27.96

123

Radiol med

Objective evaluation of image quality For the objective evaluation of image quality of the proximal coronary arteries, four parameters were selected. Noise (N), CT density (HU), signal-to-noise ratio (SNR) and contrast-to-noise ratio (CNR) were analysed by two experienced cardiac radiologists (E.D.C. and A.G.), as previously described [12, 14]. Image noise was defined as the standard deviation of CT density measured at the ascending aorta and was determined by placing the largest possible region of interest (ROI) in the aortic root cranial to the left coronary ostium with care to avoid inclusion of the aortic wall [8, 13, 15] (Fig. 1a, b). The signal (VI) of the proximal coronary arteries was measured as the mean attenuation value (HU) within circular ROIs drawn in the central portion of both the left main coronary artery and right coronary artery, with care to avoid inclusion of the coronary vessel wall [8]. CT density of the epicardial fat (VE) surrounding the artery was measured by placing a ROI immediately next to the artery (both left main coronary artery and right coronary artery) [10]. SNR and CNR were calculated as follows: SNR = VI/N, CNR = (VI-VE)/N; SNR was calculated by dividing the density by the image noise and, as for the CNR, CT density of the epicardial fat (VE) was subtracted from signal (VI), which was then divided by the image noise [10]. Estimation of radiation dose For each patient radiation dose was assessed by multiplying the DLP (dose length product) by the conversion coefficient for the chest (0.014 mSv/mGy cm) [16].

differences between matched pairs were evaluated using the signed rank test for continuous variables and McNemar’s test for binary data. Only subjects with overlapping propensity score values were compared and included in the final analysis. p values \0.05 were considered statistically significant. Interobserver agreement for image quality was calculated with the Cohen k statistic [18], which was interpreted as poor (k \ 0.20), fair (k = 0.21–0.40), moderate (k = 0.41–0.60), good (k = 5 0.61–0.80), very good (k = 0.81–0.90) or excellent (k [ 0.91). SPSS version 13.0 was used for statistical analysis and graphic presentation.

Results A total of 272 patients were scheduled for coronary CT. Of these, 200 were included in the final analysis. Twenty-one subjects were excluded due to previous coronary artery interventions (15 patients with coronary artery by-pass graft) and HR [64 bpm after beta-blocker premedication (6 patients). Five patients with severe calcification of all three major coronary arteries were excluded due to limitations in the assessment of image quality. Finally, 46 subjects were excluded from the final analysis since they did not present overlapping propensity score values. The patients’ demographic and clinical characteristics are listed in Table 2. In group A, 24 patients (24 %) did not have significant stenosis, 36 (36 %) had single-vessel disease, 18 (18 %) had two-vessel disease, and 22 (22 %) had threevessel disease. In group B, 26 patients (26.5 %) did not have significant stenosis, 31 (31.6 %) patients had single-

Statistical analysis

Table 2 Clinical and demographic characteristics according to propensity analysis

Continuous variables were condensed by means, standard deviation (SD), standard error (SE) or 95 % confidence intervals (CI) as appropriate. Differences in continuous variables were analysed with Student’s t test. Differences in categorical variables were compared with the v2 or Fisher exact test, when appropriate. To minimise bias related to the nonrandom assignment of patients, the propensity score method was used [17]. The significance level of post hoc analysis was adjusted by the Bonferroni method. In order to reduce treatment selection bias and determine treatment effect, a case–control matched propensity analysis was performed. Multivariate logistic regression was used to calculate the predicted probability of the dependent variables as well as the propensity score for all observations in the dataset. A propensity score was arranged using gender, age, weight and BMI as covariates. A 1:1 matched analysis was performed where one case was matched with one control. For the matched analysis,

Parameter

Group A

Group B

Number of patients

102

98

69

69

33

29

123

p

Gender (%) M F

0.87

Age

66.9 ± 7.5

65.2 ± 9.5

0.09

Weight (kg)

75.7 ± 11.4

73.8 ± 10.7

0.23

BMI (kg/m2)

26.9 ± 3.6

26.2 ± 3.5

0.16

b-Blocker (%)

59

64

0.56

HR \65 bpm (%)

100

100

1.00

Cardiovascular risk factors Hypertension (%)

45

43

0.89

Hypercholesterolaemia (%)

64

68

0.65

Smoking (%) Diabetes (%)

35 10

32 12

0.76 0.82

Family history (%)

41

43

0.89

BMI body mass index, HR heart rate, bpm beats per minute

\0.0001

\0.0001 \0.0001

\0.0001

\0.0001

p value

Radiol med

vessel disease, 17 (17.4 %) had 2-vessel disease and 24 patients had 3-vessel disease (24.5 %).

98 0 \0.0001 155 0 \0.0001 0 410

\0.0001

157

55 268 177 0 \0.0001 \0.0001 76 277 204 107 \0.0001 \0.0001 220 108 168 749

\0.0001 \0.0001

37 204

2

22 210

41 \0.0001

\0.0001 5

0 24

180 \0.0001

1.0 0

2

0

72

\0.0001

\0.0001

2

29

Interobserver agreement on subjective image quality was interpreted as ‘‘very good’’ for both groups (k = 0.81 for group A and 0.84 for group B). Among 1,600 segments analysed, 257 segments in group A and 242 in group B were considered to be unevaluable because the segment was absent (224 and 215 segments, respectively), too small (diameter \1.5 mm; 30 and 23 segments, respectively) or completely occluded (3 and 4 segments, respectively). In group A, no segments were scored as 4, 215 segments (16 %) as 3, 601 segments (45 %) as 2, 462 segments (34 %) as 1 and 65 segments (5 %) as 0. In group B, 410 segments (30.2 %) were scored as 4, 749 segments (55.2 %) as 3, 168 segments (12.4 %) as 2, and 29 segments (2.1 %) as 1 and 2 segments (0.1 %) as 0 (Table 3). Statistical analysis performed with the v2 test with Bonferroni post hoc analysis indicated that the number of segments with scores 4 and 3, regardless of segmental class, was significantly higher in group B compared with group A (p \ 0.005) (Fig. 1c, d). Interestingly, the number of distal segments with scores 4 and 3 was 0 % (0/428) and 0 % (0/428), respectively, in group A and 22 (98/445) and 60 % (268/445), respectively, in group B, with a significant difference in favour of group B (p \ 0.005). The mean value of the total subjective image quality scores for all of the three segmental classes was significantly higher for group B datasets than for group A datasets (3.11 ± 0.71 versus 1.72 ± 0.78, p \ 0.0001) (Table 4). The mean score calculated for each of the three segmental classes was also significantly higher in AIDR 3D datasets than in FBP datasets (Table 4). The mean differences in the proximal, intermediate and distal segmental classes between the groups were 1.2 (95 % CI of mean difference 1.11–1.30), 1.37 (95 % CI of mean difference 1.28–1.46) and 1.67 (95 % CI of mean difference 1.57–1.76). Indeed, improvement of the image quality was especially observed in the analysis of the intermediate and distal segmental classes. For these segmental classes an improvement of

Table 4 Mean subjective image quality scores of the different segmental classes in the two groups

Total

Group A

Group B

p

1.72 ± 0.78

3.13 ± 0.71

\0.0001

0

Proximal

2.09 ± 0.66

3.29 ± 0.65

\0.0001

Mid

1.76 ± 0.83

3.13 ± 0.68

\0.0001

Distal

1.31 ± 0.64

2.98 ± 0.76

\0.0001

4

601 215 2 3

65 0

462

Segmental classes

1

Group A (N = 428) Group A (N = 515) Group B (N = 400) Group A (N = 400) Group A (N = 1,343)

Group B (N = 1,358)

p value

Proximal Total Score

Table 3 Segment-based subjective image quality scores in the two groups

p value

Mid

Group B (N = 513)

p value

Distal

Group B (N = 445)

Subjective evaluation of image quality

123

Radiol med Table 5 Objective image quality and radiation dose Parameter

Group A

Group B

p

Mean difference

Standard error

95 % CI of mean difference

-64.1

12.8

-38.9 to -89.3

HU aorta

443 ± 87

507.1 ± 93.5

0.0007

Aorta noise

30.6 ± 5.4

27.6 ± 3.9

0.0018

481 ± 85.5

0.0001

-3.03 -67.2

0.95

413.8 ± 72.3

SNR LM

13.9 ± 3.1

17.7 ± 3.5

0.0001

-3.77

0.67

CNR LM

16.2 ± 3.5

20.6 ± 3.6

0.0001

-4.40

0.71

HU RCA

383.6 ± 79.4

440.4 ± 83.3

0.0007

SNR RCA

12.7 ± 2.7

16.2 ± 3.3

0.0001

-3.42

0.61

CNR RCA

15 ± 3

19.1 ± 3.6

0.0001

-4.15

0.66

-2.80 to -5.40

Dose (mSv)

3.34 ± 0.8

2.55 ± 1

0.0001

0.78

0.18

1.13 to 0.42

-56.8

11.2

-4.90 to -1.15

HU LM

11.5

-45.15 to -89.25 -2.45 to -5.09 -2.98 to -5.81 -34.11 to -79.49 -2.21 to -4.63

CI confidence interval, HU Hounsfield units, SNR signal-to-noise ratio, CNR contrast-to-noise ratio, LM left main, RCA right coronary artery

56.3 and 43.9 %, respectively, was observed when AIDR 3D automatic exposure control was used. Objective evaluation of image quality The results of objective image quality analysis are described in Table 5. Noise was significantly higher (p = 0.0018) in group A while the SNR and CNR were significantly higher in group B (Table 5). In particular, in group B we observed a 3.03 (95 % CI of mean difference -4.9 to -1.15) average noise reduction, a 3.77 (95 % CI of mean difference 2.45–5.09) average SNR increase and a 4.40 (95 % CI of mean difference 2.98–5.81) CNR increase at the level of the LM, and a 3.42 (95 % CI of mean difference 2.21–4.63) average SNR increase and a 4.15 (95 % CI of mean difference 2.80–5.40) average CNR increase at the level of RCA (Table 5). Mean effective dose was 3.34 ± 0.8 mSv (range 1.2–6 mSv) in group A and 2.55 ± 1 (range 0.8–4.9) in group B (Table 5) which represents 76.4 % of the mean radiation dose described in group A.

Discussion The development of 320-detector CT makes it possible to cover the whole heart in a single CT snapshot and eliminates stair-step artefacts inherent in 64-slice technology [19]. A previous study reported that coronary CT using a 320-detector scanner has the potential to significantly reduce the radiation dose while maintaining high diagnostic accuracy [20]. Moreover, the AEC systems now available for multidetector CT scanners are able to adjust the radiation dose according to the patient’s attenuation while sustaining diagnostic image quality [21]. AEC adjusts the tube current to maintain a user-specified noise level in the image data [21]. SureExposure (Toshiba Medical, Tokyo, Japan), the AEC system that we used, provides both patient-size

123

and z-axis AEC [21]. This system determines the tube current on the basis of projection data obtained from a scanogram [21]. The appropriate tube current is applied at the maximum water equivalent diameter to achieve the selected standard deviation (noise level) [21]. In our series we used one the most recent iterative reconstruction process, the AIDR 3D. The AIDR 3D works in both the raw data and the image domain. First, the algorithm works on the raw data performing a noise reduction process of the sinogram based on a projection noise estimation considering a noise model and a scanner model. The noise model analyses the electronic and quantum noise in the raw data. The scanner model analyses the physical properties of the CT system during the acquisition. This analysis involves not only noise reduction but also artefact correction. After the raw data elaboration, an initial input FBP image is reconstructed. This input image is processed with a sophisticated technique to minimise the noise in the image preserving the sharp details. This correction takes into account the specific body region scanned. At each iteration the SNR is improved and the input image is compared with the output image. At the end of the iterative process, a weighted blending between the original image and the image obtained by the process is automatically applied by the system. AIDR 3D improves the SNR, preserving and enhancing the edge of the anatomical structures and the spatial resolution. A limit of a general iterative reconstruction is the unnatural look of the images; AIDR 3D gives images a natural look by increasing the SNR and the spatial resolution. With AIDR 3D it is possible to maintain a target level of image noise while significantly reducing the exposure dose to the patient. The amount of dose reduction with a fixed level of noise is automatically determined by the SureExposure3D with AIDR 3D integrated into the acquisition protocols. In this study, we evaluated the effect of AIDR 3D and AEC on image quality and mean radiation dose in comparison with conventional FBP that employs exposure parameters based on patient BMI, on a 640 slice CT scanner

Radiol med

[8, 22]. FBP algorithms were the technique traditionally used for CT reconstruction and represent the most widespread system [7]. The main advantage of FBP includes fast reconstruction while the main disadvantages are related to the inevitable increase in image noise following tube current reduction [7]. Statistical iterative reconstruction algorithms are an old idea and were successfully adopted by emission tomography in nuclear medicine due to small amounts of projection data [23]. Its application in CT was limited by the time-consuming repetitive reconstruction process until a few years ago, when advances in computer performance made iterative reconstruction feasible also for CT, which produces much more projection data than emission tomography [23]. In the present study, AEC and AIDR 3D showed significant reduction of the radiation dose with improvement of the image quality. Indeed, both subjective and objective image quality analysis showed significantly higher quality in the AEC–AIDR-3D group. Subjective analysis showed significant improvement of the image quality in the AIDR group images for all the segments evaluated. Separate analysis of proximal, intermediate and distal segments also confirmed the higher quality in the AEC–AIDR-3D group in comparison to the FBP group. Our results are similar to those of Tomizawa et al. [7] but, while they used only AIDR 3D, our data combine the improvement due to iterative reconstruction with those due to AEC. Yoo et al. [8] found an improvement in image quality using AIDR 3D with a built-in AEC system but they did not evaluate the effects of these technologies on radiation doses. Williams et al. [9] also found both a significant improvement in image quality and a reduction in radiation dose using AIDR 3D but the quality assessment was made on a per-patient rather than a per-segment level. This is particularly important because our results showed improvement of the image quality in the AIDR 3D group, especially in the intermediate and distal segmental class, which are generally the segments with lowest image quality, primarily due to the small calibre of these vessels (\1.5 mm in diameter). By combining AIDR 3D and AEC we also observed a significant exposure reduction. Indeed, the AIDR–3D-AEC group showed a significant reduction in effective dose (3.34 ± 0.8 vs. 2.55 ± 1 mSv) with a 23.6 % reduction in comparison with the FBP group. Also, we observed a 3.03 ± 0.95 noise reduction with a 0.78 ± 0.18 mSv dose reduction. The dose reduction is mainly due to a significant decrease of tube current at the same scanning conditions because of the introduction of AIDR 3D. The AEC system, taking into account the AIDR3D reconstruction, automatically selects the tube current on the basis of the body profile and the attenuation information derived from the scanogram, to reach the target standard deviation of pixel values for a specified target image quality. Moreover, AIDR 3D improves the SNR.

The main limitation of this study is clearly the estimation of the current reduction generated by the introduction of the AIDR 3D with respect to the current chosen from the BMI table. To evaluate the real current reduction it is necessary to fix the image quality (S.D). The BMI table does not give a complete correlation with the quality of the images and S.D. expected. BMI-based dose exposure is less accurate. It does not consider the patient’s morphology or mass distribution. Therefore, the introduction of AEC allows for a better image quality because it enables dose optimisation. The same BMI, indeed, may be present in a patient with a very large abdomen and small chest diameter. In our study we investigated only the dose reduction due to the contribution of AIDR 3D. In order to evaluate this contribution we regarded as fixed all the other variables that influence image quality. The results in dose reduction obtained with AIDR 3D suggest that coronary CT imaging could be further improved by modifying the reconstruction kernel [24] and the contrast injection protocol [25]. It is worthy of note that AIDR 3D is more efficient when the tube voltage is low and, in this study, we use a protocol that employed a tube voltage of 120 kV. Therefore, we are introducing a protocol that employs a lower voltage (100 kV) to obtain a higher contrast. The next step will be the possible introduction of 80 kV. Conclusions The main strength of our study lies in the use of propensity score analysis which helped us to obtain two groups of patients virtually randomised for important clinical characteristics. Thus, the comparative analysis by propensity-matched pairs made the results less prone to methodological biases compared with other usual statistical methods and allowed us to confirm that AIDR 3D with AEC is able to reduce significantly the mean radiation dose and improve the image quality with respect to traditional FBP without dose modulation. Acknowledgments The present work benefited from the input of Dr. Eng. Alessandro Zappata, of Toshiba Medical System, who provided valuable assistance to the undertaking of the research summarised here. Conflict of interest Ernesto Di Cesare, Antonio Gennarelli, Alessandra Di Sibio, Valentina Felli, Alessandra Splendiani, Giovanni Luca Gravin, Antonio Barile and Carlo Masciocchi declare no conflict of interest.

References 1. Lloyd-Jones D, Adams RJ, Brown TM et al (2010) Executive summary: heart disease and stroke statistics—2010 update: a report from the American Heart Association. Circulation 121:948–954

123

Radiol med 2. Einstein AJ, Henzlova MJ, Rajagopalan S (2007) Estimating risk of cancer associated with radiation exposure from 64-slice computed tomography coronary angiography. JAMA 298:317–323 3. Hamon M, Biondi-Zoccai GG, Malagutti P et al (2006) Diagnostic performance of multislice spiral computed tomography of coronary arteries as compared with conventional invasive coronary angiography: a meta-analysis. J Am Coll Cardiol 48: 1896–1910 4. Di Cesare E, Carbone I, Carriero A et al (2012) Clinical indications for cardiac computed tomography. From the Working Group of the Cardiac Radiology Section of the Italian Society of Medical Radiology (SIRM). Radiol Med 117:901–938 5. Hausleiter J, Meyer T, Hermann F et al (2009) Estimated radiation dose associated with cardiac CT angiography. JAMA 301: 500–507 6. Labounty TM, Earls JP, Leipsic J et al (2010) Effect of a standardized quality-improvement protocol onradiation dose in coronary computed tomographic angiography. Am J Cardiol 106: 1663–1667 7. Tomizawa N, Nojo T, Akahane M et al (2012) Adaptive iterative dose reduction in coronary CT angiography using 320-row CT: assessment of radiation dose reduction and image quality. J Cardiovasc Comput Tomogr 6:318–324 8. Yoo RE, Park EA, Lee W et al (2013) Image quality of adaptive iterative dose reduction 3D of coronary CT angiography of 640-slice CT: comparison with filtered back-projection. Int J Cardiovasc Imaging 29:669–676 9. Williams MC, Weir NW, Mirsadraee S et al (2013) Iterative reconstruction and individualized automatic tube current selection reduce radiation dose while maintaining image quality in 320-multidetector computed tomography coronary angiography. Clin Radiol 68:e570–e577 10. Chun EJ, Lee W, Choi YH et al (2008) Effects of nitroglycerin on the diagnostic accuracy of electrocardiogram-gated coronary computed tomography angiography. J Comput Assist Tomogr 32:86–92 11. Austen WG, Edwards JE, Roe BB et al (1975) A reporting system on patients evaluated for coronary artery disease. Report of the Ad Hoc Committee for Grading of Coronary Artery Disease, Council on Cardiovascular Surgery, American Heart Association. Circulation 51:5–40 12. Leipsic J, Labounty TM, Heilbron B et al (2010) Adaptive statistical iterative reconstruction: assessment of image noise and image quality in coronary CT angiography. AJR Am J Roentgenol 195:649–654 13. Park EA, Lee W, Kim KW et al (2011) Iterative reconstruction of dual-source coronary CT angiography: assessment of image

123

14.

15.

16.

17. 18. 19.

20.

21.

22.

23.

24.

25.

quality and radiation dose. Int J Cardiovasc Imaging 28: 1775–1786 Bittencourt MS, Schmidt B, Seltmann M et al (2011) Iterative reconstruction in image space (IRIS) in cardiac computed tomography: initial experience. Int J Cardiovasc Imaging 27: 1081–1087 Hara AK, Paden RG, Silva AC et al (2009) Iterative reconstruction technique for reducing body radiation dose at CT: feasibility study. AJR Am J Roentgenol 193:764–771 Bongartz G, Golding SJ, Jurik AG et al (2004) European Commission Website. European guidelines for multislice computed tomography: appendix C. Funded by the European Commission, March 2004. Contract No. FIGM-CT2000-20078-CT TIP Rubin DB (1997) Estimating causal effects from large data sets using propensity scores. Ann Intern Med 127(8 Pt 2):757–763 Cohen J (1960) A coefficient of agreement for nominal scales. Educ Psychol Meas 20:37–46 Hoe J, Toh KH (2009) First experience with 320-row multidetector CT coronary angiography scanning with prospective electrocardiogram gating to reduce radiation dose. J Cardiovasc Comput Tomogr 3:257–261 Dewey M, Zimmermann E, Hamm B et al (2009) Non invasive coronary angiography by 320-row computed tomography with lower radiation exposure and maintained diagnostic accuracy: comparison of results with cardiac catheterization in a head-tohead pilot investigation. Circulation 120:867–875 Lee CH, Goo JM, Ye HJ et al (2008) Radiation dose modulation techniques in the multidetector CT era: from basics to practice. Radiographics 28:1451–1459 Chen MY, Steigner ML, Leung SW et al (2013) Simulated 50% radiation dose reduction in coronary CT angiography using adaptative iterative dose reduction in three-dimensions. Int J Cardiovasc Imaging 29:1167–1175 Silva AC, Lawder HJ, Hara A et al (2010) Innovations in CT dose reduction strategy: application of the adaptative statistical iterative reconstruction algorithm. AJR 194:191–199 Cademartiri F, La Grutta L, Runza G et al (2007) Influence of convolution filtering on coronary plaque attenuation values: observations in an ex vivo model of multislice computed tomography coronary angiography. Eur Radiol 17:1842–1849 La Grutta L, Galia M, Gentile G et al (2013) Comparison of iodinated contrast media for the assessment of atherosclerotic plaque attenuation values by CT coronary angiography: observations in an ex vivo model. Br J Radiol 86:20120238