The Impact of Different Levels of Adaptive Iterative Dose Reduction 3D on Image Quality of 320-Row Coronary CT Angiography: A Clinical Trial

Purpose The aim of this study was the systematic image quality evaluation of coronary CT angiography (CTA), reconstructed with the 3 different levels of adaptive iterative dose reduction (AIDR 3D) and compared to filtered back projection (FBP) with quantum denoising software (QDS). Methods Standard-dose CTA raw data of 30 patients with mean radiation dose of 3.2 ± 2.6 mSv were reconstructed using AIDR 3D mild, standard, strong and compared to FBP/QDS. Objective image quality comparison (signal, noise, signal-to-noise ratio (SNR), contrast-to-noise ratio (CNR), contour sharpness) was performed using 21 measurement points per patient, including measurements in each coronary artery from proximal to distal. Results Objective image quality parameters improved with increasing levels of AIDR 3D. Noise was lowest in AIDR 3D strong (p≤0.001 at 20/21 measurement points; compared with FBP/QDS). Signal and contour sharpness analysis showed no significant difference between the reconstruction algorithms for most measurement points. Best coronary SNR and CNR were achieved with AIDR 3D strong. No loss of SNR or CNR in distal segments was seen with AIDR 3D as compared to FBP. Conclusions On standard-dose coronary CTA images, AIDR 3D strong showed higher objective image quality than FBP/QDS without reducing contour sharpness. Trial Registration Clinicaltrials.gov NCT00967876

(F/23/08, F/27/10), the Joint Program from the German Research Foundation (DFG) and the German Federal Ministry of Education and Research (BMBF) for meta-analyses (01KG1013, 01KG1110, 01KG1110), GE Healthcare, Bracco, Guerbet, and Toshiba Medical Systems. The CARS-320 study has received grants from Bracco. Prof. Dewey has received lecture fees from Toshiba Medical Systems, Guerbet, Cardiac MR Academy Berlin, and Bayer (Schering-Berlex). Prof. Dewey is a consultant to Guerbet and one of the principal investigators of multi-center studies (CORE-64 and 320) on coronary CT angiography sponsored by Toshiba Medical Systems. He is also the editor of Coronary CT Angiography and Cardiac CT, both published by Springer, and offers hands-on workshops on cardiovascular imaging (www.ct-kurs.de). Prof. Dewey is an associate editor of Radiology and European Radiology. Dr. Jörg Blobel and Dr. Joanne Schuijf are full-time employees of Toshiba Medical Systems Europe. Institutional master research agreements exist with Siemens Medical Solutions, Philips Medical Systems, and Toshiba Medical Systems. The terms of these arrangements are managed by the legal department of Charité -Universitätsmedizin Berlin. The authors of this manuscript declare no relationships with any companies, whose products or services may be related to the subject matter of the article. The authors have full control control of all primary data. All funders (TMS, Guerbet, CMR Bayer, Siemens) provided support in form of salaries, but did not have any role in the study design, data collection and analysis, decision to publish, or preparation of the manuscript. The specific roles of these authors are articulated in the "author contributions`section".
for standard dose coronary CTA by using different quantitative parameters: signal intensity/attenuation (CT number [HU]), noise, signal-to-noise ratio (SNR), CNR and contour sharpness.

Ethics statement
This clinical trial has been registered (www.clinicaltrials.gov; Study NCT00967876; August 27, 2009). The study is based on the Declaration of Helsinki and in accordance with the TREND guidelines [22] because of the non-randomized study design. Find attached the TREND checklist (S1 Checklist). We received written informant consent from all patients for the study. The study protocol is available as S1 Protocol. It was approved by the Charité IRB (EA1/133/08) and the Federal Office for Radiation Protection (BfS Z5-22462/2-2008-057) and all authors confirm that all related trials for this intervention are registered.

Study design and patient collective
This study was a non-prespecified substudy of the prospective intention to diagnose study Coronary Artery Stent Evaluation with 320-row CT [23]. In this image quality substudy we included patients with suspected in-stent restenosis and stable symptomatic. clinicaltrials.gov for further details). Clinical follow-up was performed 6 months, 12 months and 24 months after CT examination (last follow-up in 11/2013). The included patients were selected with regards to their gender (15 male patients, 15 female patients) and their body mass index (BMI male 29.1±2.5 kg/m²; BMI female 28.5±5.2 kg/m²) that should exclude outliers from the analysis. All patients underwent a coronary CTA. The exclusion and inclusion criteria, the CT protocol and the clinical indications for the examinations have already been published [23,24]. The study protocol was approved by the local ethics committees and all patients gave written informed consent after verbal and written information of the study.

Coronary CT angiography (CTA)
CTA was performed on a 320-row CT (0.5 mm detector collimation and 350 ms rotation time, Aquilion ONE, Toshiba Medical Systems, Otawara Japan) and occurred within 14 days after the inclusion in the study. The CTA was performed with electrocardiogram (ECG) triggering covering 70-80% of the RR interval if the heart rate (HR) was 65 bpm, and in case of a HR ranging from 66 to 79 bpm ECG triggering was performed between 40-80% of the RR interval (no patient had a HR >79 bpm). Immediately before the scan, nitroglycerine was sublingually administered (1.2 mg Nitrolingual N Spray, Pohl-Boskamp, Hohenlockstedt, Germany). In case of a HR 65 bpm the patients received oral beta blockers one hour before CTA (atenolol, Tenormin, Astra-Zeneca), or directly before the scan beta blockers were intravenously administered (esmolol, Brevibloc, Baxter). The amount of the contrast agent (Iomeron 400, Bracco Imaging, Milan, Italy) was 50-70 ml, and the amount and the flow were adjusted to the patients`body weight [23,24]. By using a tube voltage of 120 kV for each patient the remaining scan parameters were adapted to the patients`BMI [23]. The tube current in the study protocol varied between 250 mA and 450 mA.

Reconstruction
The reconstruction was performed with a field of view (FOV) of 180 mm and a kernel of FC 05 for both IR 3D and FBP. Slice thickness was 0.5 mm by using a reconstruction interval of Toshiba Medical Systems. He is also the editor of Coronary CT Angiography and Cardiac CT, both published by Springer, and offers hands-on workshops on cardiovascular imaging (www.ct-kurs. de). Prof. Dewey is an associate editor of Radiology and European Radiology. Dr. Jörg Blobel and Dr. Joanne Schuijf are full-time employees of Toshiba Medical Systems Europe. Institutional master research agreements exist with Siemens Medical Solutions, Philips Medical Systems, and Toshiba Medical Systems. The terms of these arrangements are managed by the legal department of Charité -Universitätsmedizin Berlin. The authors of this manuscript declare no relationships with any companies, whose products or services may be related to the subject matter of the article. The authors have full control control of all primary data. All funders (TMS, Guerbet, CMR Bayer, Siemens) provided support in form of salaries, but did not have any role in the study design, data collection and analysis, decision to publish, or preparation of the manuscript. The specific roles of these authors are articulated in the "author contributions`section". This does not alter the authors' adherence to PLOS ONE policies on sharing data and materials. Two employees from Toshiba Medical System (JB and JS) gave their input into the manuscript and, due to that, are listed as authors.
0.25 mm. First, FBP/QDS was reconstructed using automated "BestPhase"reconstruction (Pha-seXact, Toshiba, Tokyo, Japan) that automatically selects the phase with fewest motion influence in the 3D dataset [24,25]. Additional reconstructions were conducted in 5% steps of the available RR interval. Subsequently, the phase with the least motion artefacts for the coronary arteries was visually chosen and reconstructed with IR 3D mild, IR 3D standard and IR 3D strong by using the reconstruction parameters as described above.

Measurement points
The data analysis was performed using a dedicated workstation (Vitrea fx, Toshiba, Tokyo, Japan). A total of 21 measurement points per patient was defined for the analysis. The first measurement point was placed in the ascending aorta directly cranial of the origin of the right coronary artery (RCA). Two additional measurement points were localised in the proximal left main coronary artery 5 mm behind the origin of the sinus of Valsalva in the vessel lumen and in the surrounding tissue at the same slice position as used for the measurement in the vessel (Fig 2). Per coronary artery (RCA, left anterior descending coronary artery [LAD], and left circumflex artery [LCX]) 6 measurement points were placed. The proximal measurement point was manually placed 5 mm distal of the origin of the vessel, the medial measurement point was placed 5 mm distal of the first branch in the major vessel, and the distal measurement point was located 5 mm distal of the second branch in the major vessel. The corresponding measurements in the surrounding tissue were performed at the same slice position and in spatial proximity to the measurement point in the vessel.

Analysis of signal and noise
The analysis of signal and noise were performed by using a region of interest (ROI) placed into the vessel lumen and the corresponding surrounding tissue on axial images (Fig 2) for the 21 measurement points. The ROI for the measurement in the ascending aorta was standardised to an area of 1 cm². The signal intensity corresponded to the grey value. The noise was defined as the standard deviation of the signal intensity of the ROI in the vessel lumen or the epicardial tissue measurement point. The SNR was calculated as follows: signal intensity in the measurement point in HU divided by the noise in the ascending aorta. The CNR was calculated as the difference between the signal intensity in the epicardial tissue and the signal intensity in the corresponding vessel lumen in HU divided by the noise in the ascending aorta. For each patient, the data sets that were reconstructed with IR 3D mild, standard, strong and combined FBP/QDS were simultaneously loaded, and the ROIs were copied from one of the four datasets into the remaining three datasets. As a consequence, all the ROIs were located at exact the same spatial position in the four datasets.

Analysis of the contour sharpness
The methods of the contour sharpness analysis have already been published [26]. We performed the measurement by gathering two different parameters: 1) analysis of the distance between 25% and 75% of the maximal grey value in mm and 2) evaluation of the maximal slope in the contour in percent that conformed to the quotient of the maximal difference of grey values between two contiguous pixels and the difference between minimal and maximal grey value in the measurement (Fig 3). The measurement of the contour sharpness was performed only in the 4 proximal measurement points in the vessel lumen (Fig 3) and the relevant area in the proximal vessel was slanted in the coronal and sagittal slices along the vessels course. A screenshot was created for the four simultaneously loaded, equally slanted images of the datasets at the same slice position. The resulting images were exported from Vitrea fx as DICOM formats. Orthogonally to the vessels course, a straight line was placed in the image connecting the epicardial tissues on both sides of the vessel for the analysis of the contour sharpness without integrating calcified, non-calcified plaques or stents by using ImageJ Version 1.4 (Fig 3). By loading the images being reconstructed with FBP/QDS, IR 3D mild, IR 3D standard and IR 3D strong simultaneously, the measurements with ImageJ were performed at the same spatial position for the four reconstructions.

Statistical analysis
The values are given as arithmetic mean (standard deviation), if not defined otherwise. First, Kolmogorov-Smirnov test was performed to test normal distribution. A p-value of 0.05 was considered to indicate statistical significance. An overall analysis was performed with the Repeated Measures ANOVA testing each measurement point and the four reconstructions as dependent variable. Only if p-value was 0.05 that was defined as statistical significant, the ANOVA test was separately conducted for each measurement point over the four reconstructions. If the p-value was 0.05, the single tests were pairwise performed for each measurement point and the separate reconstructions by using the t-test for dependent variables. The significance level was automatically adapted by SPSS version 20 for the multiple testing with the following 6 single tests by using Bonferroni correction: FBP/QDS-IR 3D mild, FBP/QDS-IR 3D standard, FBP/QDS-IR 3D strong, IR 3D mild-IR 3D standard, IR 3D mild-IR 3D strong, IR 3D standard-IR 3D strong. Dependent on the number of measurement points for each variable, a further correction of the significance level was manually performed by using Bonferroni correction: for the signal and noise measurements, a p-value 0.002 was considered statistical significant (21 measurement points), for the SNR and the CNR a p-value 0.005 was defined as statistical significant (10 measurement points) and the p-value for the contour sharpness was corrected to a value of 0.01 (4 measurement points). The influence of the vessel segments was tested over all 30 patients by ANOVAs test, comparing the relative SNR and CNR differences of the three analysed vessel segments for RCA, LAD and LCX. These relative mean differences were calculated as the difference between each IR 3D level value and the FBP value divided by the FBP value. The means for each vessel segment were calculated and the relative deviation from the means over all vessel segments is illustrated in Blant-Altman Plots for SNR and CNR of each IR 3D level.  Table for statistical results; a/e: combined filtered back projection and quantum denoising filtering software; b/f: adaptive iterative dose reduction threedimensional (AIDR 3D) mild; c/g: AIDR 3D standard; d/h: AIDR 3D strong; i: The analysis of contour sharpness is based on the difference between 25% and 75% of the maximal gray value (max.) and the maximal slope of gray values between two pixels in percent of the difference between minimal (min.) and maximal density. doi:10.1371/journal.pone.0125943.g003

Patient collective
We included 30 patients with a male-to-female ratio of 1:1 (15 females, 15 males; Table 1) and a mean age of 64 years (64.1 ± 9.1 years). The patients were pre-obese with a mean BMI of 29 kg/m² (28.8 ± 4.0 kg/m²). During the CT examination the average HR was <60 bpm (56.2 ± 7.3 bpm). By using a tube voltage of 120 kV the average tube current was 342 mA (341.7 ± 60.0 mA). Mean radiation dose of CTA was 3.2 mSv (3.2 ± 2.6 mSv). In this image quality substudy no measurement was excluded from the analysis, and all measurement points were classified as evaluable.

Signal
The comparison of the signal intensity in the vessel lumen and the surrounding tissue showed no significant difference between FBP/QDS and IR 3D mild (S2 Table; Fig 4). In the vessel lumen signal intensity was decreased for IR 3D standard in comparison to IR 3D mild by 2% and for IR 3D strong by 2% (at 10 of 11 measurement points p<0.001), while in the surrounding tissue the comparison of signal between IR 3D mild and IR 3D standard (p0.007 at 10/10 measurement points), and also for IR 3D strong versus IR 3D standard (p0.018 at 7/10 measurement points) showed no statistical significant differences (S2 Table).

Noise
Half of the measurement points in the surrounding tissue showed an increase of noise for IR 3D mild compared to FBP/QDS (p0.002; S3 Table; Table 2). The remaining half of the measurement points in the surrounding tissue, and the 11 measurement points in the vessel lumen showed the same tendency, but without statistical significance after Bonferroni correction (p n. s. ; Fig 4). Noise was higher for IR 3D mild against IR 3D standard (p<0.001 at 21 measurement points) and also for IR 3D standard compared to IR 3D strong (p0.001 at 20/ 21 measurement points). In the ascending aorta noise increased for IR 3D mild compared to FBP/QDS by 8.6 ± 11.7%, and, noise decreased for IR 3D standard in comparison to FBP/QDS by 9.6 ± 9.0%, and by 21.7 ± 7.8% with IR 3D strong as compared to FBP/QDS.

SNR and CNR
Eight of 10 measurement points were each characterised by a reduction of both SNR and CNR for IR 3D mild against FBP/QDS (SNR reduction of 10% and CNR reduction of 9%), all 10 measurement points showed an increase of SNR and CNR by using IR 3D standard in comparison to IR 3D mild (SNR increase of 19% and CNR increase of 18%) and also for IR 3D strong as compared to the IR 3D standard (p0.004, respectively; Fig 4; S4 Table; Table 2; SNR and CNR increase of 14%). The two measurement points that showed no statistical significance in the comparison between IR 3D mild and FBP/QDS were located in the distal LCX and LAD (p>0.005).
Testing the 3 vessel segments over all 30 patients due to the relative SNR, the significances were determined for IR 3D mild p>0.754 in RCA, p>0.114 in LAD, p>0.634 in LCX; for IR 3D standard p>0.349 in RCA, p>0.459 in LAD, p>0.371 in LCX and for IR 3D strong p>0.011 in RCA, p>0.923 in LAD and p>0.012 in LCX and due to the relative CNR the significances were determined for IR 3D mild p>0.451 in RCA, p>0.032 in LAD, p>0.605 in LCX; for IR 3D standard p>0.255 in RCA, p>0.222 in LAD, p>0.432 in LCX and for IR 3D strong p>0.004 in RCA, p>0.340 in LAD and p>0.020 in LCX. In summary slight differences were found, using IR 3D mild for CNR in LAD and using IR 3D strong for SNR and CNR in RCA and LCX. With IR 3D standard no differences in SNR or CNR were found. The relative deviations from the FBP means over all analysed vessel segments, as illustrated in Blant-Altman Plots of Fig 5are confirming visually these non-significant differences of the distal vessels (open marked dots) compared with the middle and proximal vessels (filled dots) of RCA, LAD, LCX, LM and aorta using the three IR 3D levels. Values are given in arithmetic mean (SD); reconstruction with filtered back projection/ quantum

Contour sharpness
The intraindividual comparison between the 4 reconstructions showed no significant difference regarding the contour sharpness based on the distance between 25% and 75% of the maximal grey value (S5 Table; Fig 6; p>0.08). The comparison of the maximal slope of grey values in the contour showed a deterioration by using the IR 3D strong as compared to the FBP/QDS at the LCX measurement point (p = 0.004). The remaining single comparisons showed no statistical significance (S5 Table).

Discussion
The use of AIDR 3D strong as compared to FBP/QDS in standard dose coronary CTA improved the following image quality parameters: strongest noise reduction, highest values for SNR and CNR. Despite the expectation of reduced contour sharpness for AIDR 3D, there was no disadvantage in comparison to FBP/QDS. The use of AIDR 3D mild resulted in similar are shown as open marked dots. Mainly, the relative deviation of SNR and CNR in distal segments is not significantly different compared to that of proximal segments; a: AIDR mild; relative deviation is less than 2% of the mean relative deviation of all segments, b: AIDR standard; Relative deviation is less than 1.5% of the mean relative deviation of all segments, c: AIDR strong; Relative deviation is less than 4% of the mean relative deviation of all segments.
doi:10.1371/journal.pone.0125943.g005 Fig 6. Comparison of contour sharpness. No significant difference was found for the contour sharpness based on the difference between 25% and 75% of the maximal gray value (a) and the maximal slope of gray values between two pixels in percent of the difference between minimal and maximal density (b); comparison between FBP/QDS, AIDR 3D mild, AIDR 3D standard (STD) and AIDR 3D strong (STR). image quality as compared to the combination of FBP/QDS when evaluating these parameters at a noise level adapted for FBP/QDS. Coronary CTA is able to detect significant coronary stenosis with a high diagnostic accuracy [1]. In patients with acute coronary syndrome and a low to intermediate likelihood, CTA is a safe and efficient alternative to the CCA [27]. With due regard to the diagnostic accuracy and cost-effectiveness in comparison to CCA, CTA may represent an first-step-diagnostic alternative in those patients [27].
In patients with known CAD coronary stents may decrease the diagnostic accuracy [23]. Instead of the FBP reconstruction also IR can be used in those patients to improve the image quality [21].
The major components of AIDR 3D include noise reduction processing in the projection data domain and iterations in the reconstruction domain [15]. Since the former is most efficient at high noise levels, noise reduction by AIDR 3D is most effective at low mAs settings (<100 mAs). The strength of the noise reduction processing in the projection data domain and the number of iterations in the image data domain are increased from AIDR 3D mild to AIDR 3D strong, resulting in highest noise reduction with AIDR 3D strong. For each application, the algorithm has been developed using as a reference FBP reconstructions at noise levels that are routine for FBP/QDS. At corresponding standard dose levels, SD values of images reconstructed with AIDR 3D standard should be similar to those of FBP/QDS reconstructions. In contrast, when data acquisition is performed at considerably lower radiation dose, lower SD values are expected for AIDR 3D as compared to FBP/QDS. Moreover, due to a nonlinear relationship, the difference in noise levels between AIDR 3D and FBP/QDS will increase with decreasing radiation exposure.
In our current study, we performed a systematic evaluation of the effect of the different levels of AIDR 3D on image quality of coronary CTA studies acquired at tube current levels currently routine for FBP/QDS. Increased noise reduction was observed with each level of AIDR 3D from mild to strong. As expected, the use of AIDR 3D mild at these high mA levels did not improve image quality parameters as compared to FBP/QDS. Nevertheless, the use of AIDR 3D strong within this particular mA range still provided improvement in image quality. This was reflected by the strongest noise reduction as well as highest values for SNR and CNR, all significantly improved as compared to FBP/QDS. No effect of AIDR 3D on contour sharpness was observed in this study. Slight differences were found for SNR and CNR between distal and proximal coronary segments, because the intensively smoothing in AIDR mild and advanced enhancement in AIDR strong influences the CT numbers in the proximal segments more intensively than in in the mid and distal segments.
Other studies that compared the image quality between iterative reconstructions and FBP showed similar results: the raw data based HIR is characterised by higher image quality parameters based on higher values for both SNR and CNR, while noise is reduced without degrading signal intensity values [11,28]. SAFIRE shows a reduction of noise and also higher subjective image quality compared to FBP [29]. In recent studies that compare AIDR 3D to FBP, noise was reduced, while SNR, CNR and subjective interpretation capacity were improved for AIDR 3D [30][31][32][33].
A recent study of Yang et al. that was performed as an intraindividual comparison, compared the image quality of five iteration levels of the raw data based SAFIRE as compared to the conventional FBP in a collective including 304 patients by using dual source CT [17]. This study showed an association between increase of the filter strength and reduction of noise, while CNR and SNR were improved without degrading signal intensity values. Our analysis showed the same results for the different levels of the AIDR 3D, with exception of mild levels. The usage of different iterative reconstruction methods could be a relevant factor for the discrepancy of the results of Yang et al. compared to that of our analysis. Additionally, in our study higher tube current was used as compared to the study from Yang et al. (43 ± 12 mA versus 342 ± 60 mA) and the patients in our study showed a higher BMI that is associated with an increase of noise, in general (23 ± 3 kg/m² versus 29 ± 4 kg/m²).
Despite the value of subjective image quality analysis for evaluating the diagnostic accuracy, we performed a systematic and objective evaluation that shows the advantage to avoid potential observer bias. Based on recent reviews [34,35], the quantification of objective image quality parameters is commonly performed by gathering SNR, CNR and noise (20 min per patient). In our analysis, for the first time a differentiation between proximal, mid and distal coronary segments was performed for the analysis of the CNR and the SNR (60 min per patient). Additionally, the evaluation of the contour sharpness was used to scientifically analyse the image quality (2.5 h per patient) based on a recent work that has already been published in PLoS ONE [26].
The small patient collective is a relevant limitation of our study. The comparison of the contour sharpness showed no statistical significant difference in our analysis. Due to the small patient collective we decided to perform an intraindividual comparison instead of a randomised trial that may show differences in one patient at the same position in the coronary segment, and not only between groups. Thus, our patients underwent the CT by using the same scanning conditions for both, AIDR 3D and FBP. Performing the AIDR 3D with lower radiation dose might have influenced the contour sharpness. In our analysis, FBP reconstruction was combined with QDS for additional noise reduction. Due to this difference in reference standard, comparison to recent studies is only partially possible. If ROIs are placed manually into distal coronary segments with a small diameter, the vessel wall could be included in the measurement and, as a result, the statistical noise is influenced by the profile of CT number and noise increases. In addition, we summarized the LM and proximal LAD, LCX and RCA to proximal measurement points though, in general, the LM vessel size is superior to that of LAD, LCX and RCA. While in our analysis there was no difference of SNR and CNR between proximal and distal segments, the image quality might me influenced by small changes of the vessel size. Although coronary CTA may be used to specify the different plaque entities [36], we did not include a plaque analysis in our manuscript. Finally, our standard scanning conditions were above 100 mAs. Even for standard clinical acquisition protocols of coronary CTA that acquire with mAs values above 100 mAs, AIDR 3D remains effective compared to original FBP to reduce SD values and improve the image quality, however the effect is less outspoken when compared to the former QDS+ technique of Toshiba. As the noise reduction effects of these iterative reconstruction techniques are expected more pronounced at lower mAs values, it will be important to explore coronary CTA protocols at lower mAs in future studies.
In our analysis, the image quality was optimised by using AIDR 3D strong that showed highest values for both SNR and CNR, while reducing noise most efficiently. Accordingly, when used at the same radiation dose, reconstructing with AIDR 3D strong will result in improved image quality as compared to FBP/QDS. Alternatively, if the aim is to have similar image quality, radiation dose can be reduced using AIDR 3D strong as compared to FBP/QDS. In contrast, the use of AIDR 3D mild resulted in similar image quality as compared to the combination of FBP/QDS when evaluating these parameters at a noise level adapted for FBP/QDS and thus may not provide additional benefit when applied in these scanning conditions. Mainly, in the distal segments of the RCA, LAD and LCX vessels the relative deviations in CNR/ SNR are nearly similar to changes in the middle and proximal vessels.