Cardiopulmonary Imaging
Original Research
Estimated Radiation Dose Reduction Using Adaptive Statistical Iterative Reconstruction in Coronary CT Angiography: The ERASIR Study
Audio Available | 
Share
OBJECTIVE. The objective of our study was to assess the impact of Adaptive Statistical Iterative Reconstruction (ASIR) on radiation dose and study quality for coronary CT angiography (CTA).
SUBJECTS AND METHODS. We prospectively evaluated 574 consecutive patients undergoing coronary CTA at three centers. Comparisons were performed between consecutive groups initially using filtered back projection (FBP) (n = 331) and subsequently ASIR (n = 243) with regard to patient and scan characteristics, radiation dose, and diagnostic study quality.
RESULTS. There was no difference between groups in the use of prospective gating, tube voltage, or scan length. The examinations performed using ASIR had a lower median tube current than those obtained using FBP (median [interquartile range], 450 mA [350–600] vs 650 mA [531–750], respectively; p < 0.001). There was a 44% reduction in the median radiation dose between the FBP and ASIR cohorts (4.1 mSv [2.3–5.2] vs 2.3 mSv [1.9–3.5]; p < 0.001). After adjustment for scan settings, ASIR was associated with a 27% reduction in radiation dose compared with FBP (95% CI, 21–32%; p < 0.001). Despite the reduced current, ASIR was not associated with a difference in adjusted signal, noise, or signal-to-noise ratio (p = not significant). No differences existed between FBP and ASIR for interpretability per coronary artery (98.5% vs 99.3%, respectively; p = 0.12) or per patient (96.1% vs 97.1%, p = 0.65). CONCLUSION. ASIR enabled reduced tube current and lower radiation dose in comparison with FBP, with preserved signal, noise, and study interpretability, in a large multicenter cohort. ASIR represents a new technique to reduce radiation dose in coronary CTA studies.
Keywords: adaptive statistical iterative reconstruction, coronary CT angiography, iterative reconstruction, radiation dose
| Introduction |   |
|---|
Adaptive Statistical Iterative Reconstruction (ASIR, GE Healthcare) is an alternative to filtered back projection (FBP) for image reconstruction in coronary CT angiography (CTA). ASIR incorporates statistical modeling to reduce image noise, which may permit preserved image quality with reduced tube current, thereby permitting lower radiation dose.
With the advent of 64-MDCT, coronary CTA has become an important tool in the assessment of coronary artery disease. It is considered an appropriate test for a number of indications including the evaluation of symptomatic patients. Coronary CTA has a high diagnostic accuracy to detect and exclude coronary artery disease in patients with a low to intermediate probability of obstructive coronary disease [1–3]. Although coronary CTA is clinically efficacious, its utility must be weighed against the association with possible future radiation-induced malignancies. Recently published data suggest that a radiation dose from coronary CTA can be significant even when performed at highly specialized centers [4, 5].
Multiple techniques have been reported to reduce radiation dose and thereby ensure that coronary CTA is performed in accordance with the as low as reasonably achievable (ALARA) principle. These techniques include the use of ECG-based tube current modulation with helical scanning [6, 7], prospective ECG-triggered sequential scanning [4, 8–10], and high pitch spiral and reduced tube voltage [4, 7, 11].
Iterative reconstruction was introduced as an image reconstruction algorithm for CT [12] and provides an alternative to traditional filtered back projection (FBP). One implementation of iterative reconstruction is ASIR. Iterative reconstruction is widely used in PET and was used in CT when it was originally introduced [13]. However, FBP is used for CT because it is fast and requires substantially less time and computational power to perform than other reconstructions.
ASIR uses iterative comparisons of each acquired projection to a synthesized projection incorporating modeling of both system optics and system statistics, permitting more efficient image updating. Compared with images reconstructed using FBP, images reconstructed using ASIR have lower image noise [14–17] (Fig. 1A, 1B). Because of its noise reduction capability, ASIR in theory can be used to reconstruct images with similar image quality (noise and signal-to-noise ratio [SNR]) but with a substantial reduction in tube current, resulting in a reduction in overall radiation dose. This hypothesis was recently studied in abdominal CT applications by Hara et al. [18]: They found that the use of ASIR was associated with effective dose reductions of 32–65% compared with routine FBP-reconstructed abdominal imaging without degradation of image quality. In coronary CTA, the noise reduction capabilities of ASIR were recently reported [19], but the clinical implications of ASIR in coronary CTA has not been studied to our knowledge.
Our hypothesis was that ASIR would permit a reduction in tube current, resulting in incident radiation dose reduction for coronary CTA with preserved image noise and study interpretability.
| Subjects and Methods | |
|---|
For the Estimated Radiation Dose Reduction Using Adaptive Statistical Iterative Reconstruction for Coronary CT Angiography (ERASIR) study, consecutive patients undergoing clinically indicated coronary CTA were enrolled in a prospective fashion. All adult patients with coronary CTA were included without regard to baseline heart rate. Exclusion criteria included renal insufficiency (estimated glomerular filtration rate < 60 mL/min/1.73 m2), known pregnancy, or prior iodinated contrast reaction. Patients were limited to those undergoing routine coronary CTA; studies that included evaluation of coronary artery bypass grafts, pulmonary arteries, or the thoracic aorta were not included. Each site had the approval of its respective institutional review board with a waiver of informed consent. Where applicable, sites were compliant with HIPAA. Patient demographics were prospectively acquired via patient query by a dedicated health care provider.
 View larger version (154K) | Fig. 1A —Effect of Adaptive Statistical Iterative Reconstruction (ASIR, GE Healthcare) on image signal and noise. CT angiographic images show ascending aorta in 48-year-old man. Images were reconstructed using both filtered back projection (FBP) (A) and 40% ASIR (B). Window width and level are same for each. Signal and noise measurements were taken from 1-cm2 circular region of interest in same location. Signal was similar between FBP and ASIR (718.6 vs 719.3 HU, respectively), whereas noise decreased between FBP and ASIR (52.3 vs 38.5 HU). |
 View larger version (165K) | Fig. 1B —Effect of Adaptive Statistical Iterative Reconstruction (ASIR, GE Healthcare) on image signal and noise. CT angiographic images show ascending aorta in 48-year-old man. Images were reconstructed using both filtered back projection (FBP) (A) and 40% ASIR (B). Window width and level are same for each. Signal and noise measurements were taken from 1-cm2 circular region of interest in same location. Signal was similar between FBP and ASIR (718.6 vs 719.3 HU, respectively), whereas noise decreased between FBP and ASIR (52.3 vs 38.5 HU). |
Because ASIR was expected to result in the use of a lower tube current and given the potential for decreased study interpretability, a power analysis was performed. Our hypothesis was that per-patient study interpretability would be similar between groups. An a priori two-tailed power analysis was therefore performed to detect any difference in independent proportions for interpretability. Assuming a baseline interpretability of 95%, to detect a 10% absolute difference at 80% power, the study was estimated to require 141 studies in each group for adequate power.
Comparisons were performed regarding patient and scan characteristics, radiation dose, signal and noise, and study interpretability. Study interpretability was assessed on a per-patient basis and per-artery basis. An artery was considered interpretable if image quality was adequate for evaluation of coronary plaque and stenosis in all segments ≥ 1.5 mm. On a per-patient basis, all four coronary arteries were required to be interpretable for a patient's examination to be considered interpretable.
Each site enrolled a consecutive and sequential series of patients in two nonoverlapping phases. The first phase consisted of consecutive coronary CTA studies performed using reconstruction with FBP. The second phase consisted of coronary CTA studies performed with ASIR. Importantly, these study cohorts represent consecutive patients presenting for coronary CTA at the same three institutions. All FBP studies were performed on a 64-MDCT system (LightSpeed VCT XT, GE Healthcare), whereas all ASIR studies were performed using a different 64-MDCT system that was equipped with ASIR (Discovery HD 750, GE Healthcare).
For both groups, each site used independent body mass index (BMI)-based protocols in which tube current and tube voltage were individualized on the basis of patient body habitus rather than using a single standard coronary CTA protocol. Tube voltage was typically reduced from 120 to 100 kV for patients with a BMI of < 30 kg/m2 and tube current was also tailored to the patient, with tube current ranging from 275 to 800 mA for patients examined. After the introduction of a scanner that permitted reconstruction with ASIR, all studies were reconstructed with ASIR. The scanner allows a blend of ASIR reconstruction with the more commonly used FBP. All sites selected 40% ASIR and 60% FBP for image reconstruction according to the vendor's recommendations and our initial experience with the reconstruction algorithm. In the ASIR group, each study site was permitted to alter their protocols to balance radiation dose and image quality and sites were aware of the potential of ASIR-enabled tube current reduction [14, 15, 18, 19]. This was done to better assess the real-world applicability of ASIR rather than on the basis of predetermined standardized scan protocols.
Patients with a heart rate of > 65 beats per minute (bpm) received oral or IV β-blockers (or both). Those without contraindications were also administered 0.4 mg of sublingual nitroglycerin immediately before the study. Two sites used a triple-phase contrast protocol: 60 mL of iodixanol (Visipaque, GE Healthcare), followed by 75 mL of a 50:50 mixture of iodixanol and saline, followed by a 50-mL saline flush. The third site used a dual-phase protocol with 75 mL of iodixanol followed by a 50-mL saline flush.
The scanning parameters included a rotation time of 350 milliseconds, 64 × 0.625 mm collimation, tube voltage of 100 or 120 kV, and tube current of 275–800 mA. All sites adjusted tube current and voltage on the basis of the BMI for both cohorts. Sequential scanning, as has been described previously [8], was used at each site when possible during the entire study period unless specifically requested by the ordering physician for assessment of left ventricular function, when the heart rate was ≥ 65 bpm despite medication administration, or when the heart rate was highly variable (≥ 5 bpm). For sequential scans, padding or additional tube “on” time beyond the minimum acquisition window was manually entered at the time of the study and at the discretion of the site depending on heart rate and heart rate variability as previously described in the literature [20]. ECG-based tube current modulation was used for all helical studies. FBP studies were reconstructed with FBP; all studies obtained on the ASIR-capable scanner were reconstructed in 40% ASIR, which represents a composite of 40% ASIR and 60% FBP.
For the ASIR-capable scanner, studies could be acquired in standard or high resolution; collimation was identical for both, although the in-plane spatial resolution was 0.625 and 0.230 mm2 for standard- and high-resolution scans, respectively. In the high-resolution mode, 2,400 views per segment are acquired instead of the routine 900 views per segment in standard resolution. The higher number of samples or views enables multiple projections of the same object, thereby enabling a higher degree of image clarity and improvement in spatial resolution.
High-resolution imaging was used at the discretion of each site and was chosen frequently when available given the theoretic benefit of enhanced spatial resolution. High resolution was used for 204 of the 243 ASIR patients; it was not available on the scanner used for the FBP cohort. Scanner settings at each of the sites were not altered on the basis of the use of standard- or high-resolution imaging.
Study interpretability was determined by three experienced level 3–certified coronary CTA imagers, each with experience reading several thousand studies. Diagnostic study quality was graded at a per-artery and per-patient level, and the study was deemed diagnostic if every anatomically present segment (≥ 1.5 mm) could be assessed for the presence of atherosclerosis and the presence and severity of stenosis. Studies were read using workstations (Advantage AW 4.3–4.4, GE Healthcare). The use of axial data sets, maximum intensity projections, curved multiplanar reformations, and other postprocessing tools was left to the discretion of each reader. All images were reconstructed using a 0.625-mm slice thickness for both scanners and at all sites. The signal and noise were measured in the aortic root at the level of the left main coronary artery on axial images using a 1.0-cm2 area to measure the mean signal value in Hounsfield units and SD (noise).
Radiation dose for coronary CTA was determined by the dose–length product (DLP); the DLP was converted to millisieverts by multiplying it by the conversion factor of 0.014 mSv × mGy–1 × cm–1 [21].
Comparisons between groups were performed using Student's t tests for continuous variables with normal distributions and the Mann-Whitney U test for continuous variables with nonnormal distributions. The chi-square test was used for categoric variables.
For multivariate analyses, stepwise linear regression models were used that assessed patient or scan characteristics expected to have associations with radiation dose, signal and noise, and tube current. The stepwise regression used an entry criterion of p < 0.05 and a removal criterion of p > 0.10. Patient characteristics included age, sex, BMI, and heart rate; scan characteristics included sequential versus helical acquisition, tube voltage, ASIR versus FBP reconstruction, and scan length. To prevent overfitting of the multivariate models, only variables with a p < 0.10 on univariate linear regression were entered into the final model.
To identify variables associated with radiation dose, stepwise linear regression models respectively considered patient characteristics only and scan characteristics only. The separate models were used because of the intrinsic interaction between patient and scan variables. Because there was a nonnormal distribution of radiation dose, the dose in millisieverts was converted to the natural logarithm of millisieverts to obtain a normal distribution for regression analysis. Univariate linear regression models were used first to determine the relationship between radiation dose (ln of mSv) and each patient or scan variable individually. Multivariate regression models, using stepwise methods, were used to look at the relationship between radiation dose (ln of mSv) and each patient or scan variable while adjusting for all other patient and scan variables. Because padding duration is only applicable to sequential scanning, an additional model was limited to sequential scans and considered the effect of scan variables, including padding duration, on radiation dose.
Because ASIR has been proposed to permit a reduction in radiation dose by means of decreased tube current, we used linear regression models to assess independent associations of patient and scan characteristics as well as scan characteristics only on tube current. Scan length was not considered in these models because it is not expected to relate to tube current. Univariate linear regression was used to first determine the independent relationship between tube current (mA) and ASIR. Multivariate linear regression looked at the association of tube current and ASIR while adjusting for all significant patient and scan variables. Another multivariate linear regression assessed the relationship between tube current and ASIR while adjusting for significant scan variables only.
Additional stepwise linear regression models, similar to those described, assessed variables associated with study signal, noise, and SNR. Candidate variables included patient and scan characteristics; the latter included acquired in-plane spatial resolution due to a potential effect on image signal and noise. Scan length was not included because it is not expected to relate to study signal and noise. These models were further assessed with scan characteristics only.
All analyses were performed with statistics software (SPSS version 17.0, SPSS) for Microsoft Windows. A two-tailed p value of < 0.05 was deemed significant.
| Results | |
|---|
Five hundred seventy-four patients were recruited, consisting of two consecutive nonoverlapping cohorts: FBP (n = 331) and ASIR (n = 243). Patient and scan characteristics are provided in Table 1. ASIR patients were more likely to be male and have higher heart rates. There was no difference between groups in terms of age or BMI. There was no difference in the use of sequential scanning or scan length between groups. In comparison with examinations of the FBP cohort, examinations of the ASIR cohort had increased use of 100-kV tube voltage and lower tube current (Table 1).
 View larger version (140K) | Fig. 2A —Images of patient at normal weight: 55-year-old woman who presented with chest pain (body mass index, 22 kg/m2). Scanning parameters were as follows: 100 kV, 325 mA, 0 milliseconds padding, and scan length of 13.9 cm. Effective radiation dose was 0.56 mSv. Left anterior descending (A) and right coronary artery (B) are displayed with curved multiplanar formations without stenosis or visualized plaque. |
 View larger version (137K) | Fig. 2B —Images of patient at normal weight: 55-year-old woman who presented with chest pain (body mass index, 22 kg/m2). Scanning parameters were as follows: 100 kV, 325 mA, 0 milliseconds padding, and scan length of 13.9 cm. Effective radiation dose was 0.56 mSv. Left anterior descending (A) and right coronary artery (B) are displayed with curved multiplanar formations without stenosis or visualized plaque. |
The use of ASIR was associated with reduced tube current with univariate linear regression (mean, –149 mA; 95% CI, –126 to –171; p < 0.001). A similar reduction in current was noted after adjustment for significant patient and scan characteristics (–141 mA; 95% CI, –121 to –161; p < 0.001) and after adjustment for significant scan characteristics only (–143 mA; 95% CI, –121 to –165; p < 0.001) (Fig. 2A, 2B).
The exact reconstruction times were not recorded; however, all reconstructions required less than 5 minutes to perform.
The unadjusted and adjusted percentage effects of scan variables on radiation dose are provided in Table 2. When only patient variables were considered, decreased heart rate (–22% per –10 bpm; 95% CI, –16% to –27%; p < 0.001) and decreased BMI (–5% per –1 kg/m2; 95% CI, –3% to –6%; p < 0.001) were each independently associated with reduced radiation dose; age and sex were not associated with radiation dose. All scan variables were independently associated with dose reduction. After adjusting for all significant scan variables, all variables except voltage remained associated with dose reduction (Table 2).
To assess the effect of padding duration on radiation dose, another multivariable model considered only sequential coronary CTA studies. After adjustment for scan variables, decreased padding duration (–55% per –100 milliseconds; 95% CI, –51% to –59%; p < 0.001), reduced tube voltage (–32% for 100 vs 120 kV; 95% CI, –24% to –40%; p < 0.001), ASIR versus FBP (–17%, 95% CI, = –11% to –22%, p < 0.001), and shorter scan length (–6%; 95% CI, = –4% to –8%, p < 0.001) were associated with reduced radiation dose.
No significant differences in the frequency of interpretable studies or arteries were observed between groups (Table 1). With univariate linear regression, ASIR studies were independently associated with reduced signal (–15 HU; 95% CI, –31 to 1; p = 0.07), increased noise (8 HU; 95% CI, 6–10; p < 0.001), and lower SNR (–2; 95% CI, –2 to –3; p < 0.001). Despite the significant reduced tube current used with ASIR studies, after adjustment for patient and scan characteristics, there was no significant difference in ASIR versus FBP studies with regard to signal (p = 0.61), noise (p = 0.73), or SNR (p = 0.19). After adjustment for scan characteristics only, the results were similar, and ASIR was not associated with a difference in signal (p = 0.26), noise (p = 0.88), or SNR (p = 0.37).
| Discussion | |
|---|
Several radiation dose reduction techniques have been recently described and are now used in clinical practice including ECG-based tube current modulation for helical examinations, axial scanning, reduction in the duration of padding [7–11], 100-kV tube imaging, and reduction of z-axis scan length [22, 23]. To our knowledge, this study represents the first validation of the new ASIR algorithm as a radiation dose reduction tool in coronary CTA. ASIR is a unique CT reconstruction algorithm that differs from the standard CT reconstruction algorithm of FBP. Unlike FBP, ASIR makes fewer assumptions regarding the distribution of noise and utilizes an iterative process of mathematic modeling to identify and selectively reduce noise [15–19].
Our study results show that ASIR permits reduction in tube current while imparting a statistically significant reduction in radiation dose due to the direct relationship between tube current and dose. The impact of ASIR is in fact incremental to established dose reduction techniques. Tube current reduction in CT would be expected to result in increased image noise due to the decreased number of photons; however, despite the tube current reduction, the studies reconstructed with ASIR had no significant increase in image noise when compared with the FBP studies. After adjustment for patient and scan characteristics or for scan characteristics in isolation, there is no difference in signal or noise between groups despite the use of lower tube current in the ASIR cohort. Although sequential scanning had the greatest impact on dose, ASIR had a statistically significant and incremental effect and was associated with a 27% reduction while adjusting for other radiation dose reduction techniques.
The significant computational power required to perform image reconstruction with ASIR has become available only recently in clinical coronary CTA applications. The studies in our ERASIR cohort were not observed to suffer from excessively long reconstruction times or to have a negative impact on clinical efficiency.
The use of iterative reconstruction techniques is expected to increase in CT as computational processing improves and algorithms become more robust and easy to apply. Because more powerful iterative reconstruction algorithms are emerging, the impact of these techniques may show greater noise reduction and thereby permit further reductions in current and radiation dose.
There are several limitations to this study. Although study interpretability was maintained in our large multicenter cohort, these interpretations were largely observational clinical interpretations by experienced coronary CTA readers without coronary catheterization correlation to assess diagnostic accuracy. Additionally, the ASIR algorithm was implemented only on a newer ASIR-capable scanner that uses a more advanced detector technology than the scanner used for FBP studies. The detector counting efficiency of the two scan platforms are comparable; however, our experimental design cannot exclude the possibility that differences in image quality and radiation dose may be partially attributable to the difference in the material composition of the detectors. In addition, the patients in our study were not randomized and thus interpretations blinded to the image reconstruction algorithm was not possible. Each study was evaluated by a single experienced reader, so interobserver variability could not be assessed. ASIR was not available across all platforms for CT at the time of the completion of this study; however, various versions of iterative reconstruction are being developed for use on multiple platforms.
In conclusion, ASIR permits the use of lower tube current with similar image noise and study interpretability in comparison with coronary CTA studies performed with FBP and higher tube current. ASIR represents a novel method of radiation dose reduction that appears additive to existing techniques.
B. Heilbron and J. K. Min receive research support from GE Healthcare. J. Leipsic, J. K. Min, and J. P. Earls serve on the medical advisory board and speaker's bureau for GE Healthcare.
Address correspondence to J. Leipsic ([email protected]).
| References | |
|---|
