Introduction

An optimal protocol of pediatric cardiac CT angiography examination should be performed in accordance with ALARA (as low as reasonably achievable) dose principles [1, 2] while maintaining diagnostic accuracy. Recently, rapid developments in CT technology have allowed reduced radiation exposure for children, including body-size-adaptive CT protocols, low tube voltage, low tube current and automated tube current modulation [3, 4]. However most of these strategies are limited by increased noise and reduced image quality with too-low radiation. Thus CT scanner manufacturers suggested several new image reconstruction methods based on iterative reconstruction techniques [2], including iDose4 and iterative model reconstruction (Philips Healthcare, Cleveland, OH) [5–7], adaptive statistical iterative reconstruction and model-based iterative reconstruction (GE Healthcare, Waukesha, WI) [5, 8], iterative reconstruction in image space and sinogram affirmed iterative reconstruction (Siemens Healthcare, Erlangen, Germany) [9], and adaptive iterative dose reduction (Toshiba Medical Systems, Otawara, Japan) [10]. Recently, pediatric CT studies have shown that compared with CT scans reconstructed with filtered back-projection, low-dose CT scans reconstructed with iterative reconstruction techniques maintain image quality with less noise [1, 6, 8, 10–16]. Few reports describe the technical and clinical feasibility of iterative reconstruction for children with congenital heart disease [9, 11, 17–19] or the optimal iDose4 for a low-dose protocol. Additionally, optimal iteration levels might be different depending on radiation dose for CT images. Thus we studied pediatric CT angiography with two radiation doses and assessed subjective and objective image quality with an advanced fourth-generation iterative reconstruction technique (iDose4, Philips Healthcare, Cleveland, OH).

Materials and methods

Patients

This prospective study was approved by our research ethics committees. The potential risks of contrast medium injection and radiation exposure were explained to the study children’s parents by a cardiac radiologist. Informed consent was obtained from parents of all children included in the study (n = 160; 114 boys). We included children younger than 1 year who had congenital heart disease previously assessed by echocardiography. All children were examined with cardiac CT angiography between February 2011 and February 2014. CT angiography scanning was used for preoperative assessment of cardiovascular anatomy or for evaluation of postoperative results. Exclusion criteria were allergies to iodine contrast medium, tachycardia (>180 beats per minute [bpm], exceeding the scanner limit for electrocardiograph gating), hyperthyroidism and impaired renal function (serum creatinine >1.4 mg/dl).

CT scanning technique and data acquisition

All CT angiography scans were performed using a 256-slice CT system (Philips Brilliance iCT; Philips Healthcare, Cleveland, OH) equipped with a filtered back-projection and hybrid iterative reconstruction (iDose4) reconstruction and post-processing package (Brilliance Workspace; Philips Healthcare, Cleveland, OH). All cases were performed by prospectively gated axial cardiac CT (Step & Shoot Cardiac; Philips Healthcare, Cleveland, OH), and all electrocardiograph electrodes were placed at standard positions. Short-term sedation of uncooperative children was achieved with chloral hydrate solution (0.1 mg/kg, per os; Brilliant Pharma, Chendu, Sichuan, China). No additional drugs were given to modify heart rate.

The following acquisition parameters were used: collimation, 96–128 × 0.625 mm; gantry rotation time, 270 ms; slice thickness, 0.8 mm; reconstruction interval, 0.4 mm. All raw data were obtained at 40–50% of the R-R interval; and three datasets were reconstructed at 40%, 45% and 50% of the R-R interval, and the best phase was chosen to be the evaluation objective. For evaluation of image quality and potential radiation dose reduction with hybrid iterative reconstruction techniques (iDose4) for pediatric cardiac CT angiography, we designed two image-acquisition protocols (full- and half-dose). Children were randomized to one group according to their registered time (i.e. 1st-full dose, 2nd-half dose, etc.). Children with parents who refused low-dose cardiac CT angiography were assigned to the full-dose group.

Based on findings from Paul et al. [20], we used a weight-based protocol (Table 1) as our full-dose protocol. For the half-dose group, we fixed tube voltages and adjusted tube currents (mAs) to the corresponding group. If CT angiography scans with either protocol failed, we planned to perform another full-dose protocol of CT angiography or MRI scan as a remedial measure within a week.

Table 1 Scanning protocol of full-dose group using 80 kVp
Full size table

Iodinated contrast medium (Ultravist 300; Bayer Schering Pharma, Berlin, Germany) was injected with a dual-syringe power injector (2.5 ml/kg, intravenously) followed by 5 ml saline. Flow rates were calculated as (weight [kg] × 2.5)/(post threshold delay + scan time +3) ml/s [21]. Bolus tracking was used in a region of interest within the descending aorta at the level of the carina, with an attenuation threshold of region of interest >100 Hounsfield units (HU) to trigger scanning after a 7-s delay. The coverage of the CT angiography scan was from the thoracic inlet to the lower end of the liver, and the scan course was accomplished in one or two passes based on patient size.

Image post-processing and quality analysis

To optimize iDose4 in pediatric CT angiography, we reconstructed all raw datasets of the best phase into four image sets, with the filtered back-projection and the hybrid iterative reconstruction (iDose4) at iteration levels 2, 4 and 6. Images were reconstructed using a XCB (standard) kernel with a thickness of 0.8 mm; the reconstruction interval was 0.4 mm. All images were transferred to an external workstation (Cardiac Viewer, Extended Brilliance Workspace 4.0; Philips Healthcare, Cleveland, OH) for interpretation. Multi-planar reformatting, curved planar reformatting, maximum-intensity projection and volume rendering were used to display cardiac abnormalities depending on target structure.

Subjective image-quality evaluation

Subjective image quality was independently evaluated with a five-point score by two experienced cardiovascular radiologists (L.Y. and H.L, 8 and 10 years of experience). The radiologists evaluated all images with mediastinum and lung window settings, and both radiologists were allowed to change the window width and level to their preference.

Overall subjective image quality was assessed by looking at cardiac and vascular structures (cardiac chambers, thoracic aorta, pulmonary arteries, pulmonary veins, coronary arteries) on 2-D axial images and other reformatted images (multi-planar reformatting, curved planar reformatting, maximum-intensity projection and volume rendering). Both radiologists were blinded to scanning parameters and patient characteristics (weight, age, gender). The scale used to assess subjective image quality was based on the one used by Huang et al. [21]. The grades were as follows; grades 3 or greater were considered sufficient for diagnostic purposes.

  • Grade 5: Excellent anatomical clarity and image quality

  • Grade 4: Good anatomical clarity; all structures clearly interpretable

  • Grade 3: Fair anatomical clarity; the anatomical relationships required clinically could be defined with confidence

  • Grade 2: Poor image quality or anatomical detail; incomplete demonstration of anatomical structures

  • Grade 1: No useful information obtained

Objective image-quality evaluation

To evaluate image quality objectively, the following data were measured: (1) attenuation and image noise, determined as the mean CT value and the mean standard deviation (SD) in the three regions of interest: ascending aorta (AA), main pulmonary artery (MPA), and myocardial walls (MW); (2) contrast-to-noise ratios (CNR), defined as the difference between CT value in the two regions of interest (ascending aorta and main pulmonary artery) and CT value in the myocardial walls divided by the standard deviation of the ascending aorta. The related formulae were CNRAA=(CT value AA – CT value MW) /SDAA, CNRMPA=(CT value MPA – CT value MW) /SDMPA. Regions of interest were adjusted to the area of interest (range from 15 mm2 to 137 mm2).

Assessment of radiation doses

CT dose index volume (CTDIvol) of CT scans was noted from the CT console after each scan. The size-specific dose estimates were calculated with methods recommended by the American Association of Physicists in Medicine (AAPM) report 204 [22]. First, we measured with on-screen calipers in centimeters the maximum lateral diameters (DLAT) of the chest on a standard axial image through the left inferior pulmonary vein. Second, we converted CTDIvol 32 cm and CTDIvol 16 cm to size-specific dose estimates values using the conversion factors listed in Tables 1 and 2 of the AAPM report [22].

Table 2 Patient demographics and radiation doses in the two dose groups
Full size table

Statistical analysis

Statistical analysis was performed with SPSS 20.0 software (IBM, Armonk, NY). We evaluated inter-observer agreement in subjective image-quality grading using kappa statistics, with 0.81 ≤ kappa ≤ 1.0 being excellent consistency, 0.61 ≤ kappa ≤ 0.80 good consistency, 0.41 ≤ kappa ≤ 0.60 moderate consistency, 0.21 ≤ kappa ≤ 0.40 fair consistency, and kappa < 0.20 poor consistency. The differences in age, heart rate and weight were compared between the two dose groups using a Student’s t-test. Subjective image-quality scores of the four image sets (filtered back-projection and three iDose4 levels) for groups were compared using a Kruskal-Wallis H test; if there were statistically significant differences, multiple comparisons were performed with a Bonferroni test to explore the optimal iDose4 iteration level. Mean CT values, noise, CTDIvol, contrast-to-noise ratios and size-specific dose estimates were compared using a Student’s t-test between filtered back-projection images and the iDose4-optimal iteration level images in each dose group. Diagnostic accuracy of CT angiography images was calculated based on the surgical or heart catheterization findings. Diagnostic accuracy differences among the four image sets were compared with Kruskal-Wallis H test. If statistically significant differences were found, multiple comparisons were studied with a Bonferroni test (P-value <0.05 was considered statistically significant).

Results

Patients

No serious adverse events were recorded and all children had successful cardiac CT angiography scanning. In four cases, parents refused to allow low-dose cardiac CT angiography, which meant that four children originally in the half-dose group underwent full-dose CT angiography, resulting in 84 children in the full-dose group and 76 in the half-dose group. Tables 1 and 2 depict patient data. There were significant differences in age (P < 0.05) but no significant differences in weight or heart rate between the two dose groups.

Radiation dose

Significant differences were noted in CTDIvol and size-specific dose estimates between the two dose groups (Table 2; all P < 0.05). Table 3 depicts data for cardiac deformities confirmed in both dose groups. Diagnostic accuracy for cardiac deformities for the four image sets were 92.6%, 94.6%, 95.1% and 96.2% (not significantly different).

Table 3 Surgical findings and conventional cardiac angiography (CCA) findings
Full size table

Subjective evaluation results

There was a good consistency of overall subjective image quality between independent observers (half-dose group, kappa = 0.74; full-dose group, kappa = 0.73). Diagnostic-quality images (score ≥3) were achieved in all cases (100%, 160/160) using iDose4 levels 4 or 6. Only 1 case (0.6%, 1/160) with transposition of the great arteries did not result in diagnostic-quality images using either filtered back-projection or iDose4 level 2, (this child was scanned with a half-dose protocol). Statistically significant differences were found in subjective scores among the four algorithms for each dose group (Table 4). Data show that the best subjective image quality was reconstructed with iDose4 level 4 for both groups (Figs. 1 and 2).

Table 4 Comparisons of subjective image-quality scores using different algorithms with full- and half-radiation dose
Full size table
Fig. 1
figure 1

Subjective image quality in the full-dose group. Effects of iDose4 on visibility of left anterior descending coronary artery (arrow in a) in a 10-month-old girl (full-dose group). a–d Different image construction algorithms: (a) filtered back projection, (b) iDose4 level 2, (c) iDose4 level 4, (d) Dose4 level 6. Image reconstructed with iDose4 level 6 (d) displays over-smoothing, or plastic-appearing edges. iDose4 level 4 (c) was considered optimal by both reviewers

Full size image
Fig. 2
figure 2

Subjective image quality in the half-dose group. Ventricular septal defect (arrow in a) in a 7-month-old girl child (half-dose group). a–d Different image construction algorithms: (a) filtered back-projection, (b) iDose4 level 2, (c) iDose4 level 4, (d) Dose4 level 6. Image reconstructed with iDose4 level 6 (d) displays over-smoothing, or plastic-appearing edges. iDose4 level 4 (c) was considered optimal by both reviewers

Full size image

Objective evaluation results

Attenuation

No statistically significant differences in mean CT values were found between filtered back-projection and iDose4 level 4 for the same evaluated anatomical region (ascending aorta, main pulmonary artery and myocardial walls) for each group. iDose4 level 4 caused minimal changes in CT numbers in the same regions of interest in the same dose group (Table 5).

Table 5 The attenuation, image noise, contrast-to-noise ratio in various anatomical regions of interest
Full size table

Image noise

Mean noise reduction with iDose4 level 4 was 28% in the ascending aorta (range, 11–38%), 27% in the main pulmonary artery (range, 7–42%) and 24% in the myocardial walls (range, 8–33%) in the half-dose group; 27% in the ascending aorta (range, 13–46%), 21% in the main pulmonary artery (range, 8–38%) and 26% in the myocardial walls (range, 8–45%) in the full-dose group, compared to filtered back-projection. Data in Table 5 show that iDose4 level 4 reduced image noise compared to filtered back-projection for both groups.

Contrast-to-noise ratio

Significant differences in contrast-to-noise ratios were observed between filtered back-projection and iDose4 level 4 for the same evaluated anatomical region (ascending aorta or main pulmonary artery) in each dose group (all P < 0.05) (Table 5). Mean contrast-to-noise ratios increased with iDose4 level 4, including 40% in the ascending aorta (range, 11–62%) and 38% in the main pulmonary artery (range, 13–86%) in the half-dose group; 38% in the ascending aorta (range, 4–72%) and 27% in the main pulmonary artery (range, 4–64%) in the full-dose group, compared with filtered back-projection.

Discussion

The traditional filtered back-projection technique provides poor image quality when CT scanning with low-radiation protocols because of the limits of its mathematical model. More recent hybrid iterative reconstruction technologies, such as iDose4, allow for radiation reduction while maintaining overall diagnostic quality. However, limited studies about optimization of hybrid iterative reconstruction for pediatric CT scanning exist in the literature [6, 17, 23]. Karmazyn’s [6] group compared filtered back-projection and five hybrid iterative reconstruction levels (levels 2–6 of iDose4) for pediatric body CT and found that hybrid iterative reconstruction levels 3 or 4 were optimal for most studies. Mieville et al.’s [17] work suggests that cases performed with filtered back-projection and several hybrid iterative reconstruction levels cause structure conspicuity decreases exceeding 50% of the hybrid iterative reconstruction levels. Their data indicate that ~20–40% hybrid iterative reconstruction level (e.g., levels 1–3 in iDose4) provides the best images [17]. Brady and colleagues [23] also reported that this technique (adaptive statistical iterative reconstruction, GE Healthcare, Waukesha, WI) decreased noise variance and increased graininess with increasing hybrid iterative reconstruction level. Based on the results of phantom and clinical observations, they concluded that 40% hybrid iterative reconstruction level (e.g., iDose4 level 3) provided the best results [23]. However, all of these studies were performed with a full- or single-dose protocol, so whether the optimal iteration is the same with full- or half-dose radiation was unclear until now.

In our study, we compared filtered back-projection and three levels of iDose4 (levels 2, 4, 6) in two groups (full- and half-dose radiation) to find the optimal iteration for each group (Figs. 1 and 2). iDose4 levels 1, 3, 5 and 7 are available, but limited experience with these suggests that differences in any two consecutive levels were minimal. We found that iDose4 level 4 was optimal for both dose groups, indicating that the level of hybrid iterative reconstruction is most important for subjective evaluation results. Data from prior studies suggest that many images reconstructed with iDose4 level 6 appeared unusually smooth, which resulted in degradation of image quality, and both radiologists in this study noted that most of images for both dose groups reconstructed with iDose4 level 6 displayed a plastic appearance (Figs. 1 and 2).

Recently, pure iteration methods, such as model-based iterative reconstruction and iterative model reconstruction, have been developed, and these can be more easily compared than those acquired with a hybrid iterative reconstruction technique [14, 16, 19, 24]. However clinical application of model-based iterative reconstruction and iterative model reconstruction is limited by a longer reconstruction time compared with hybrid iterative reconstruction or filtered back-projection [19] methods, and, as reported by Mieville and colleagues [5], structure conspicuity in these techniques is decreased by more than 50% as compared to hybrid iterative reconstruction. So, whether similar results are obtained with pure iteration images requires more studies with large sample sizes. For this pediatric CT angiography study of hybrid iterative reconstruction, neither radiologist noticed artifacts or alterations in image appearance for any regions with iDose4 level 4.

A reliable iterative reconstruction method should not change the attenuation (HU). Our results demonstrate that there were no statistically significant differences in attenuation between filtered back-projection and iDose4 level 4 images for the full- or the half-dose group, which indicates that iDose4 level 4 does not influence the attenuation. A similar result was found by Zheng et al. [11].

There were no statistically significant differences for diagnostic accuracy among the four image sets (full- and half-dose filtered back-projection and optimal hybrid iterative reconstruction level images). This may be because there were limited effects of half-doses of radiation or that the half-dose protocol might be still higher than needed, indicating that further radiation reduction is possible. Our data show that the hybrid iterative reconstruction technique (iDose4) allowed a reduction of size-specific dose estimates by half without a loss of diagnostic information. Mieville’s group [17] evaluated the benefits of adaptive statistical iterative reconstruction on diagnostic image quality in pediatric cardiac CT examinations and indicated that a 36% radiation reduction is possible for a 2- to 3-year-old child when using 40% adaptive statistical iterative reconstruction. Tricarico and colleagues [9] assessed image quality of simulated half-dose pediatric cardiovascular CT angiography, and they reconstructed all raw data with filtered back-projection and iterative reconstruction. They reported that iterative reconstruction improved image noise, contrast-to-noise ratios, and subjective image quality compared with filtered back-projection for low-radiation-dose pediatric CT angiography and might allow for further radiation reductions without compromising diagnostic image quality. Zheng and co-workers [11] decreased radiation by 53.8% for 62 pediatric cardiovascular CT angiography exams and reported that a combination of prospectively electrocardiograph-triggered high-pitch spiral acquisition, low tube current, and iterative reconstruction technologies offered diagnostic images for pediatric cardiovascular CT angiography with effective dose <0.1 mSv. Compared with prior studies, we documented similar or higher dose reductions with hybrid iterative reconstruction in children with congenital heart disease. Effective dose or size-specific dose estimate values from our study might be greater than those of published reports because we used a large scanning range (from the thoracic inlet to the lower end of the liver) and large phase tolerance around the 45% phase (40–50%).

Our study has several limitations. First, the study only evaluates one vendor’s reconstruction technique, and optimization parameters might not be easily translated to other vendor technology. Second, during the subjective image assessment, although we removed all study information from CT angiography images, the typical smoothing effect of hybrid iterative reconstruction on CT angiography images limited a true blinding — radiologists can differentiate these from traditional filtered back-projection images and this might introduce some bias. Finally, the sample size was moderately small, with only 76 children enrolled in the half-dose group. Although our data indicate that hybrid iterative reconstruction in pediatric CT angiography examinations reduced radiation without affecting image quality, more studies with larger sample sizes are required to confirm our findings before routine application in the field. We plan to conduct more studies with larger sample sizes to evaluate the stability of the half-dose protocol. If our findings are confirmed, we plan to implement the half-dose protocol with iDose4 level 4 as a routine protocol for all pediatric CT angiography exams.

Conclusion

iDose4 level 4 was optimal for most patients in both dose groups, indicating that iDose4 might significantly reduce radiation used for pediatric cardiovascular CT angiography studies. A low-dose protocol with iDose4 level 4 and prospective electrocardiograph-triggered acquisition permitted a 53% reduction in size-specific dose estimates, without significantly affecting image quality and diagnostic confidence. This work might be helpful for promoting hybrid iterative reconstruction for pediatric cardiovascular CT angiography and thereby decreasing radiation exposure.