Evaluation of an iterative model-based reconstruction of pediatric abdominal CT with regard to image quality and radiation dose

Abstract

Background

In pediatric patients, computed tomography (CT) is important in the medical chain of diagnosing and monitoring various diseases. Because children are more radiosensitive than adults, they require minimal radiation exposure. One way to achieve this goal is to implement new technical solutions, like iterative reconstruction.

Purpose

To evaluate the potential of a new, iterative, model-based method for reconstructing (IMR) pediatric abdominal CT at a low radiation dose and determine whether it maintains or improves image quality, compared to the current reconstruction method.

Material and Methods

Forty pediatric patients underwent abdominal CT. Twenty patients were examined with the standard dose settings and 20 patients were examined with a 32% lower radiation dose. Images from the standard examination were reconstructed with a hybrid iterative reconstruction method (iDose⁴), and images from the low-dose examinations were reconstructed with both iDose⁴ and IMR. Image quality was evaluated subjectively by three observers, according to modified EU image quality criteria, and evaluated objectively based on the noise observed in liver images.

Results

Visual grading characteristics analyses showed no difference in image quality between the standard dose examination reconstructed with iDose⁴ and the low dose examination reconstructed with IMR. IMR showed lower image noise in the liver compared to iDose⁴ images. Inter- and intra-observer variance was low: the intraclass coefficient was 0.66 (95% confidence interval = 0.60–0.71) for the three observers.

Conclusion

IMR provided image quality equivalent or superior to the standard iDose⁴ method for evaluating pediatric abdominal CT, even with a 32% dose reduction.

Keywords

Computed tomography (CT)iterative reconstruction image quality pediatric abdomen dose reduction

Introduction

It is important to be aware of the radiation dose delivered to pediatric patients in computed tomography (CT) examinations, because children are more radiosensitive than adults, and thus, they are at higher risk of developing radiation-induced cancer. This difference in radiation sensitivity is based on the high cellular mitotic activity and longer life expectancy in children (1 –4). The International Commission of Radiological Protection recommends that individual doses should be as low as reasonably achievable to minimize the risk to the patient (5).

As awareness of the importance of reducing radiation exposure has grown, many steps have been taken to reduce the radiation dose in CT examinations. Dose-reducing technologies include hardware improvements, such detector sensitivity, adaptive collimators, anti-scatter grids, and dose modulations, in addition to software solutions, such as post-processing image filters and iterative reconstruction. These improvements have contributed to the feasibility of reducing the radiation dose without compromising image quality (6). One important recent development was the reintroduction of iterative reconstruction methods to replace the conventional filtered back projection procedure. The first reconstruction technique used in a clinical CT scanner was algebraic reconstruction, which is a type of iterative reconstruction (7). A report from the Summit on the Management of Radiation Dose in Computed Tomography stated that iterative reconstruction algorithms claimed to play an important role in achieving routine sub-milliSievert CT scans (8). The iterative reconstruction algorithms attempt to reconstruct CT images with low noise and improved or preserved resolution of anatomical structures. Iterative algorithms allow the reconstruction process to incorporate statistical models in addition to models of the CT system and the acquisition process. The iterative process can be applied to image data, projection (raw) data, or a combination of both. Different CT manufacturers use different methods to reconstruct images; these algorithms are typically classified into either hybrid or model-based iterative reconstructions (9).

The purpose of this study was to investigate whether a new iterative model-based reconstruction (IMR) method could maintain or improve the image quality of pediatric abdominal CT compared to the current reconstruction method, when the radiation dose was reduced.

Material and Methods

Patients

The local ethics committee approved this study and the requirement for informed consent was waived. Forty patients (age range = 1–15 years; weight range = 10.7–45.0 kg) were included in the study. The patients were either referred from our emergency department or scheduled for oncological follow-ups. Children weighing more than 50 kg were excluded from the study. The participants were divided into two groups: one group underwent a standard CT examination of the abdomen with intravenous injection of contrast medium; the second group underwent the same examination, but with 32% lower radiation dose. In both groups, the dosage of contrast medium was 2 mL/kg (iodoxanol, Visipaque®, 270 mg I/mL, GE Healthcare, Oslo, Norway), with injection speeds of 1.5–3.0 mL/s. A summary of patient characteristics and the scanning protocol parameters are shown in Table 1.

Table 1.

Descriptions of the patients and the scan parameters used in this study. The only difference between Techniques II and III was the reconstruction technique. Age, weight, tube load, and CTDI_vol are given as the median (range); other values were constant. The CTDI_vol values are given for the reference CTDI body phantom, 32 cm in diameter.

Group	Technique I	Technique II/ Technique III
Patients (n)	20	19
Age (years)	10 (2 –12)	6 (1 –15)
Weight (kg)	33 (15.1–43.5)	25 (10.7–45)
Tube voltage (kV)	100	100
Dose right index	17	14
Tube load (mAs)	68 (35–106)	50 (31–89)
Pitch	1.2	1.2
Collimation	64 × 0.625	64 × 0.625
Reconstruction technique	iDose⁴ level 4	iDose⁴ level 4/ IMR level 1
Kernel	C	C/soft tissue
Slice thickness (mm)	5	5/0.9
CTDI_vol (mGy)	2.7 (1.4–4.2)	1.9 (1.2–3.5)

Scan parameters

All CT scans were performed on a Brilliance Ingenuity 128 system (Philips Healthcare, Cleveland, OH, USA). The image quality reference parameter for the Philips automatic exposure control system was the Dose Right Index (DRI) (10). A lower DRI indicated a reduced radiation dose and, therefore, increased image noise. Each DRI step lowered the tube load by 12%. The first 20 patients were scanned at DRI 17, which was the dose level currently used clinically, and the other 20 patients were scanned at DRI 14, i.e. a 32% dose reduction (Table 1).

Reconstruction settings

All patient images were reconstructed in 5-mm slice thicknesses with 2.5-mm increments. For the hybrid iterative reconstruction with iDose⁴ (SW version 4.1.6) (9), the reconstruction was performed at level 4 with kernel C, which corresponded to a sharp filter (11). The first set of 20 patient images was designated Technique I (standard dose level) and the next set of 20 patient images (low dose level) was designated Technique II. In addition, the latter patient images were also reconstructed with the IMR (9), in 0.9-mm slice thicknesses with increments of 0.45 mm, and with a soft tissue reconstruction algorithm. This image set was designated Technique III.

Image quality assessment

An objective assessment of image quality was performed by measuring the image noise (standard deviation of the CT numbers), in a reasonably homogeneous part of the liver. This was measured in all image sets. The position of the region of interest (ROI) is shown in Fig. 1.

Fig. 1.

Abdominal CT images (axial view) of a 12-year-old male patient, acquired at low dose. (a) IMR image reconstructed with 0.9-mm-thick slices (Technique III). (b) iDose⁴ level 4 image reconstructed with 5-mm-thick slices (Techniques I and II). The images show the position of the region of interest (red circles) for measuring the image noise in the liver. Notice the different appearance of noise in the two images.

In addition, image quality was subjectively assessed by three pediatric radiologists in a visual grading study. The observers had 19–26 years of experience in diagnostic radiology with 13–21 years in pediatric radiology. The stacks of axial images were presented individually in randomized order on a PACS workstation. Observers used the viewing and scoring software, ViewDEX v2.0 (Viewer for Digital Evaluation of X-ray images, The Sahlgrenska Academy at University of Gothenburg, Gothenburg, Sweden) for evaluations (12). The window level and width were pre-set to 40 and 400, respectively. The observers were able to adjust the setting as desired. In a training session before the image quality evaluation, four demo images of patients that were not included in the study were shown to the observer.

The anatomical structures used for the image quality evaluations were based on modified visualization criteria from the European guidelines (13). The anatomic structures (questions A–E, Table 2) were selected to include a wide range of important aspects of quality. For example, good image quality included both high spatial resolution in the reproduction of intrahepatic vessels and low contrast in the pancreas. A four-point grading scale was used to describe how well the criteria were fulfilled: “clearly” indicated that the structure had a distinct shape; “acceptably” indicated that the structure was moderately well delineated; “poorly” indicated that the structure was vaguely delineated; and “not at all” indicated that the structure could not be discerned. It was also possible to use a fifth option, “not applicable,” when in any case the structure was missing.

Table 2.

Subjective evaluations of image quality and corresponding rating scales.

Subjective evaluation
A	How well are the liver parenchyma and intrahepatic vessels reproduced?
B	How well are the pancreatic contours reproduced?
C	How well is the aorta delineated?
D	How well is the differentiation between the adrenal glands and adjacent structures reproduced?
E	How well is the spleen delineated?
F	Rate the overall image quality.
G	Rate the diagnostic confidence in the image appearance.
Rating scale for questions A–E
4	Clearly
3	Acceptably
2	Poorly
1	Not at all
N/A	Not applicable
Rating scale for questions F–G
4	Excellent
3	Sufficient
2	Poor
1	Unacceptable

The overall image quality and the diagnostic confidence in the image appearance (questions F and G) were divided into four ratings on the visual grading scale, including “excellent,” “sufficient,” “poor,” and “unacceptable” quality. A summary of all questions and rating scales is presented in Table 2.

Statistical analysis

The visibility of anatomic structures (questions A–E, Table 2) was evaluated with a visual grading characteristics (VGC) analysis (14,15). The VGC Analyzer software v1.0 (The Sahlgrenska Academy at University of Gothenburg, Gothenburg, Sweden), written in IDL (Research System, Inc, Boulder, CO, USA), was used to evaluate the results (16,17). The program produced a statistical analysis of the rating data from studies performed with multiple readers, where multiple cases were assessed, and all readers assessed all cases. The area under the VGC curve (AUC_VGC) was the result of the VGC study. A non-parametric method was applied for statistical tests, including a bootstrapping resampling technique and a permutation resampling technique, to determine the confidence interval (CI) and the P value for testing the null hypothesis. When the two methods compared were equivalent, the analysis returned an AUC_VGC = 0.5. The VGC Analyzer calculated the bootstrap-averaged AUC_VGC, the non-parametric 95% confidence interval (CI) of the AUC_VGC, and the P value for the null hypothesis. When the 95% CI did not include the value 0.5, a statistically significant difference at the 95% level between the two compared methods was established.

For questions F and G, which covered overall image quality and image appearance, the mean value was calculated and the distribution of the rating scores was analyzed.

An intraclass correlation coefficient (ICC) (18,19) was calculated to assess agreement among the three observers, i.e. it determined whether the three observers graded questions equally. Inter-observer variability refers to systematic differences among the observers. ICC was calculated with SPSS Statistics (Statistical Package for the Social Science, IBM SPSS Statistics for Macintosh, Armonk, NY, USA) with a two-way mixed model of ICC. The reliability of the assessment was determined as follows: an ICC < 0.4 was considered “poor;” an ICC of 0.4–0.59 was considered “fair;” an ICC of 0.60–0.74 was considered “good;” and an ICC > 0.75 was considered “excellent” (19).

The intra-observer variability was based on duplicate evaluations of eight cases. Variability was determined by the number of equivalent ratings between the first and second set of evaluations for a specific observer.

Results

For one patient, the images scanned with the low dose protocol (Techniques II and III) had extensive motion artifacts; thus, those images were excluded. Consequently, 19 patients were included in all analyses performed with Techniques II and III.

Noise

The image noise measured in the liver was lower with the IMR method (Technique III) compared to the standard and low dose examinations performed with iDose⁴ (Techniques I and II) (Fig. 1). The mean noise measurements were 13 HU for Technique I, 16 HU for Technique II, and 8 HU for Technique III. Image noise values from individual examinations are shown in Fig. 2. These results confirmed results from a previous study performed on anthropomorphic phantoms (20).

Fig. 2.

Image noise (HU) measured in the liver as a function of patient body weight. Technique I employed the standard abdominal CT protocol and images were reconstructed with iDose⁴ level 4 in 5-mm-thick slices. Techniques II and III were scanned with a 32%-lower radiation dose and reconstructed with iDose⁴ level 4 (Technique II) or with iterative model-based reconstruction (Technique III), with slice thicknesses of 5 mm and 0.9 mm, respectively.

Visual grading characteristic analysis

The ratings for questions A–E were not significantly different between the images from Technique I (standard-dose abdominal CT reconstructed with iDose⁴) and Technique III (32% low-dose abdominal CT reconstructed with IMR). For all questions, the AUC_VGC was >0.5, except for question C (the aorta delineation) (Fig. 3a). Techniques II and III are compared in Fig. 3b; Technique III provided significantly higher image quality than Technique II, for aspects covered in questions B and D. A comparison of Techniques I and II showed that the former technique had a higher image quality for questions C and E (Fig. 3c).

Fig. 3.

Visual grading characteristic (VGC) analysis for subjective evaluation questions A–E (Table 2). The area under the VGC curve (AUC_VGC) was used to compare image quality between techniques and equivalent ratings are indicated by a CI (whiskers) that includes 0.5. (a) Technique I (standard abdominal CT protocol, reconstruction with iDose⁴ level 4) compared to Technique III (32% lower radiation dose, IMR). No significant difference is apparent. (b) Technique II, (32% lower radiation dose, reconstruction with iDose⁴ level 4) compared to Technique III. IMR was rated significantly better than iDose⁴ for questions B and D. (c) Technique I compared to Technique II. The higher radiation dose provided significantly better ratings for questions C and E.

Mean score and distribution

The distributions of scores for overall image quality and for confidence in the image appearance are shown in Fig. 4. The highest scores (“excellent” and “sufficient”) were given to Techniques I and III more frequently than Technique II. Two cases were rated “unacceptable” for Technique II, regarding the overall image quality and the diagnostic confidence in the image appearance. The mean scores for overall image quality and image appearance were highest for IMR (Technique III) and lowest for Technique II.

Fig. 4.

Distributions of the rating scores from subjective evaluations (Table 2). Technique I employed the standard CT abdomen scanning protocol, and images were reconstructed with iDose⁴; Techniques II and III employed 32% lower dose scans, and images were reconstructed with iDose⁴ (Technique II) or with the IMR (Technique III). (a) Overall image quality (question F). (b) Confidence in image appearance (question G).

Inter-observer variability

The ICCs for the ratings were 0.71 (95% CI = 0.62–0.79) for images performed with Technique I; 0.67 (95% CI = 0.56–0.76) for images performed with Technique II; and 0.71 (95% CI = 0.62–0.79) for images performed with Technique III. The ICCs were in the range of 0.60–0.74 for all three techniques, which indicated good agreement (19). For all images included in the study, the ICC was 0.66 (95% CI = 0.60–0.71) for the three observers.

Intra-observer variability

Evaluation of intra-observer variability was based on duplicate evaluations of eight images. Each observer was investigated for variability on each of the seven questions (i.e. 56 questions). The rating from the first evaluation was compared to the rating of the second evaluation of the same image. The agreement levels were 80% for observer 1, 86% for observer 2, and 91% for observer 3.

Discussion

The main purpose of this study was to evaluate if new model-based iterative reconstruction method and the potential to maintain or improve image quality of pediatric abdominal CT compared to the current reconstruction method, when the radiation dose was reduced with 32%. The results from the VGC analysis showed no differences in image quality between the current reconstruction method with standard dose and the new model-based reconstruction with lower radiation dose.

The standard abdominal examination according to Technique I was an optimized examination routinely used in the clinic. The image quality was acceptable to radiologists and the radiation dose was considered to be as low as reasonably achievable. When introducing a new examination technique (in this case a reconstruction method), the image quality must be equivalent or better than that achieved with the previous technique, and the radiation dose must be equivalent, or preferably, reduced. In this study, we reduced the radiation dose by 32%, and evaluated the new reconstruction method to determine whether the images were equivalent or better in quality at the lower dose level, compared to the current reconstruction technique at the higher dose level.

No image quality difference was found between Techniques I and III, which indicated that the new reconstruction method could produce high quality images at a substantially reduced radiation dose (Fig. 3a). A reduction of the radiation dose with the standard reconstruction method (Techniques II vs. I) resulted in inferior image quality, as judged by the observers (Fig. 3b). The delineations of the aorta and spleen were significantly impaired on the lower-dose scan. With IMR (Technique III), images were reconstructed in 0.9-mm-thick slices with increments of 0.45 mm, which differed from the reconstructions performed with iDose⁴ (Techniques I and II; 5-mm-thick slices with 2.5-mm increments). There is a balance between the slice thickness and the amount of noise in the image. A previous study (21) showed that IMR could be used for analyzing images at thinner slice thicknesses, which improved image quality by suppressing the partial volume effect. Currently, IMR has been introduced in some routine examination protocols for adults in our clinic and images are analyzed in thin slices.

A comparison of the images evaluated with Techniques II and III showed that IMR provided better image quality than iDose⁴ for questions B and D. Question B corresponded to pancreatic contours, and question D corresponded to adrenal gland differentiation from adjacent structures. IMR provided better resolution with low contrast than iDose⁴, particularly for low-dose scans, and this feature could explain the different ratings for questions B and D.

The delineation of the aorta with IMR was rated mediocre compared to Technique I. This might be explained by the fact that the aorta appeared to have jagged edges with IMR, because the noise in the structure was enhanced with IMR. As aortic pathology in children is uncommon, the pediatric radiologists were disturbed by the jagged edge of the aorta. The radiologists realized that the jagged edge was an artifact and assigned a lower scoring for IMR, compared to iDose⁴. In contrast, with the standard iDose⁴ reconstruction, the reader could overlook the noise and interpret the shape as a soft curve. The aorta was contrast-filled, and in the axial view, it appeared nearly perfectly circular-shaped; this round shape may be more appreciable with some noise. The structures of the aorta and the detection of contrast filled vessels could probably be enhanced by using an IMR kernel (sharper filter) that is optimized for vessels. The optimization of kernels for IMR was outside the scope of this study.

Objective measurements of the noise in the liver showed that IMR produced the lowest noise among the three techniques, which was expected, based on previous studies (20,21). Technique III received the highest mean ratings for overall image quality and for diagnostic confidence in image appearance, closely followed by Technique I. We also asked the observers to rate the appearance of the IMR images (question G), because the image appearance rendered with the model-based reconstruction technique is typically different from the appearance rendered with hybrid reconstruction techniques. In a previous study, the noise power spectrum (NPS) was calculated to evaluate the image appearance with IMR and iDose⁴ techniques (20). An NPS curve reveals the amount of noise and the characteristics of the image noise. iDose⁴ images have a coarse graininess and IMR images are smoother. In this study, we found that the appearances of IMR images were mostly scored “sufficient” or “excellent;” only 5% were rated “poor.” The observers had some previous experience with CT images reconstructed with IMR; thus, they were somewhat familiar with the appearance of IMR images. The ratings for overall image quality and diagnostic image appearance (questions F and G) would probably have been different, if IMR images had been completely new to the observers.

One limitation of this study was that the groups compared were different in patient age and weight. Since automatic exposure control were used, the difference in patient size was compensated by the tube current adaption. The patients evaluated in Technique I were older and weighed more than the patients evaluated in Techniques II and III. Smaller children lack fat between tissues and organs, which complicates the diagnosis. Thus, the larger proportions of smaller children evaluated with Techniques II and III represented a disadvantage compared to Technique I. Another limitation of this study is recognition bias since the observers could identify the images reconstructed with IMR due to the thinner slice thickness. A limitation using visual grading methods is that the diagnostic accuracy was not evaluated. However, an assumption with VGC analysis is that there is a correlation between the reproduction of anatomical and pathologic features, but further studies are needed to determine the diagnostic performance of IMR.

In conclusion, the implementation of IMR for pediatric abdominal CTs acquired with a 32% dose reduction provided acceptable image quality without degradation.

Footnotes

Acknowledgments

The authors thank pediatric radiologist Pär Wingren for assessing the images.

Declaration of Conflicting Interests

The author(s) declared no potential conflicts of interest with respect to the research, authorship, and/or publication of this article.

Funding

The author(s) received no financial support for the research, authorship, and/or publication of this article.

References

Pierce

Shimizu

Preston

et al.

Studies of the mortality of atomic bomb survivors. report 12, part I. Cancer: 1950-1990. 1996. Radiat Res 2012; 178: 61–87.

Brenner

Elliston

Hall

et al.

Estimated risks of radiation-induced fatal cancer from pediatric CT. Am J Roentgenol 2001; 176: 289–296.

Shah

Platt

. ALARA: is there a cause for alarm? Reducing radiation risks from computed tomography scanning in children. Curr Opin Pediatr 2008; 20: 243–247.

Pearce

Salotti

Little

et al.

Radiation exposure from CT scans in childhood and subsequent risk of leukaemia and brain tumours: a retrospective cohort study. Lancet 2012; 380: 499–505.

Charles

. ICRP Publication 103: Recommendations of the ICRP , Ottawa, ON: ICRP, 2008.

Kalender

Buchenau

Deak

et al.

Technical approaches to the optimisation of CT. Phy Med 2008; 24: 71–79.

Hounsfield

. Computerized transverse axial scanning (tomography): Part I. Description of system. 1973. Br J Radiol 1995; 68: 166–172.

McCollough

Chen

Kalender

et al.

Achieving routine submillisievert CT scanning: report from the summit on management of radiation dose in CT. Radiology 2012; 264: 567–580.

Mehta

Thompson

Morton

et al.

Iterative model reconstruction: simultaneously lowered computed tomography radiation dose and improved image quality. Med Phys Int J 2013; 2: 147–155.

10.

Vlassenbroek

Mehta

Yanof

. CT Radiation Dose: Philips Perspective. Radiation Dose from Multidetector CT, Berlin: Springer, 2012, pp. 617–632.

11.

Olsson M-L, Norrgren K, Söderberg M. Improvement of CT image quality with iterative reconstruction iDose⁴. Conference paper ECR. Vienna: Proc of European Congress of Radiology, 2014:C-0387.

12.

Börjesson

Håkansson

Båth

et al.

A software tool for increased efficiency in observer performance studies in radiology. Radiat Prot Dosimetry 2005; 114: 45–52.

13.

Jessen

Panzer

Shrimpton

et al.

EUR 16262: European guidelines on quality criteria for computed tomography , Luxembourg: European Commission, 2000.

14.

Båth

Månsson

. Visual grading characteristics (VGC) analysis: a non-parametric rank-invariant statistical method for image quality evaluation. Br J Rad 2007; 80: 169–176.

15.

Båth

. Evaluating imaging systems: practical applications. Radiat Prot Dosimetry 2010; 139: 26–36.

16.

Båth

Hansson

. VGC analyzer: A software for statistical analysis of fully crossed multiple-case visual grading characteristics studies. Radiat Prot Dosimetry 2016; 169: 46–53.

17.

Hansson

Månsson

Båth

. The validity or using ROC software for analysing visual grading characteristic data: an investigation based on the novel software VGC analyzer. Radiat Prot Dosimetry 2016; 169: 54–59.

18.

Landers R. Computing Intraclass Correlations (ICC) as Estimates of Interrater Reliability in SPSS. The Winnower 2015. DOI:10.15200/winn.143518.81744.

19.

Hallgren

. Computing inter-rater reliability for observational data: an overview and tutorial. Tutor Quant Methods Psychol 2012; 8: 23–34.

20.

Aurumskjöld

Ydström

Tingberg

et al.

Improvements to image quality using hybrid and model-based iterative reconstructions: a phantom study. Acta Radiol 2017; 58: 53–61.

21.

Aurumskjöld

Ydström

Tingberg

et al.

Model-based iterative reconstruction enables the evaluation of thin-slice Computed tomography images without degrading image quality or increasing radiation dose. Radiat Prot Dosimetry 2016; 169: 100–106.