European Radiology

, Volume 22, Issue 1, pp 138–143

Adaptive statistical iterative reconstruction versus filtered back projection in the same patient: 64 channel liver CT image quality and patient radiation dose

Authors

  • Lee M. Mitsumori
    • Department of RadiologyUniversity of Washington School of Medicine
    • Department of RadiologyUniversity of Washington School of Medicine
  • Janet M. Busey
    • Department of RadiologyUniversity of Washington School of Medicine
  • Orpheus Kolokythas
    • Department of RadiologyUniversity of Washington School of Medicine
  • Kent M. Koprowicz
    • Department of RadiologyUniversity of Washington School of Medicine
Computed Tomography

DOI: 10.1007/s00330-011-2186-3

Cite this article as:
Mitsumori, L.M., Shuman, W.P., Busey, J.M. et al. Eur Radiol (2012) 22: 138. doi:10.1007/s00330-011-2186-3

Abstract

Objectives

To compare routine dose liver CT reconstructed with filtered back projection (FBP) versus low dose images reconstructed with FBP and adaptive statistical iterative reconstruction (ASIR).

Methods

In this retrospective study, patients had a routine dose protocol reconstructed with FBP, and again within 17 months (median 6.1 months), had a low dose protocol reconstructed twice, with FBP and ASIR. These reconstructions were compared for noise, image quality, and radiation dose.

Results

Nineteen patients were included. (12 male, mean age 58). Noise was significantly lower in low dose images reconstructed with ASIR compared to routine dose images reconstructed with FBP (liver: p < .05, aorta: p < 0.001). Low dose FBP images were scored significantly lower for subjective image quality than low dose ASIR (2.1 ± 0.5, 3.2 ± 0.8, p < 0.001). There was no difference in subjective image quality scores between routine dose FBP images and low dose ASIR images (3.6 ± 0.5, 3.2 ± 0.8, NS).Radiation dose was 41% less for the low dose protocol (4.4 ± 2.4 mSv versus 7.5 ± 5.5 mSv, p < 0.05).

Conclusions

Our initial results suggest low dose CT images reconstructed with ASIR may have lower measured noise, similar image quality, yet significantly less radiation dose compared with higher dose images reconstructed with FBP.

Keywords

Image qualityLow dose CTAdaptive statistical iterativeReconstructionFiltered back projectionRadiation dose reduction

Introduction

Computed tomography images have been reconstructed from raw data using filtered back projection (FBP) for several decades. The standard FBP algorithm is based on several fundamental assumptions about CT geometry and is a compromise between reconstruction speed and image noise. Different assumptions about CT geometry combined with multiple iterations of reconstruction—termed adaptive statistical iterative reconstruction (ASIR)—may result in slightly longer reconstruction time but also in less image noise from the same raw data. ASIR reduces image quantum noise without having an impact on spatial or contrast resolution through more complex modeling of detector response and of the statistical behavior of CT measurements [110].

When 64 channel CT is performed with automated tube current modulation (ATCM), image noise is determined in part on some CT systems by operator selection of a noise index (NI), defined as the standard deviation of the image noise resulting from the range of mA employed by the ATCM. Patient dose may be decreased by increasing the operator-selected noise index (lowering the mA range), resulting in greater image noise [11]. If this greater image noise could then be modulated with ASIR, low dose CT imaging might be possible with resultant image noise comparable to the image noise from routine dose imaging reconstructed with FBP [3, 5].

We hypothesized that a low dose protocol performed with a higher noise index and ASIR reconstruction might produce liver images of comparable noise and subjective diagnostically acceptable image quality yet with less patient radiation dose compared with a routine dose protocol performed with a lower noise index and FBP reconstruction. To test this hypothesis, we retrospectively identified a sequential group of liver cirrhosis patients who had been examined by CT twice: first with our routine dose FBP protocol and then at a later date with a low dose ASIR protocol. We compared these two CT examinations for quality, noise, and radiation dose.

Materials and methods

This single-center study was approved by our University institutional review board; informed consent was not required because of the retrospective design of the research.

Patient population

Adaptive statistical iterative reconstruction became available at our institution for clinical CT imaging in March 2009. By searching electronic CT logs during the following two months, we identified a sequential series of 19 patients awaiting liver transplant who had undergone imaging for the indication of advanced cirrhosis with two CT examinations: the first performed with our previous routine standard dose clinical protocol using FBP and the second performed after the introduction ASIR with our new, modified routine low dose clinical protocol. As a part of our routine clinical practice, usual CT exclusion criteria included patients with a GFR of less than 30 or with severe allergy to contrast material. To assemble this retrospective study of sequential patients who underwent imaging twice for clinical indications, we checked for exclusion criteria of interval development of significant ascites, interval marked change in body weight, or interval therapeutic intervention such a chemoembolization; none of these was encountered in this series.

CT technique

All examinations were performed on 64 channel CT (Lightspeed VCT, GE Healthcare, Milwaukee, WI, USA). All patients had circulation time from the antecubital fossa to the abdominal aorta estimated by a timing bolus of 15 ml of IV contrast material (Omnipaque 350, GE Healthcare, Chalfont St. Giles, UK) followed by 15 ml of saline administered through a dual-head power injector (Stelland D, Medrad, Warrendale, PA, USA) at 5 ml per second. Subsequently, for the diagnostic imaging, 150 ml of contrast material followed by 30 ml of saline was administered at 5 ml per second with the start of imaging of the upper abdomen determined by the sum of the timing bolus peak plus 20 s for the late arterial phase, by the beginning of the injection plus 65 s for the portal venous phase, and at five minutes after the beginning of the injection for the delayed phase. In order to do a focused assessment of measured image noise and overall subjective image quality (as opposed to lesion identification) we analyzed only the late arterial phase images, according to the method of Marin et al. [4]. Other CT parameters are described in Table 1; only the noise index and the reconstruction algorithm were different between the first and the second CT examinations. For each image, the noise index was relative to the first sequence slice thickness reconstructed (0.625 mm). The first CT examination in each patient was obtained with a routine dose protocol noise index of 30 and reconstructed with 100% FBP. The second examination in each patient was performed with a low dose protocol at a later date using an empirical increase of the noise index to 40. This second examination was reconstructed two ways: with 100% FBP and then with a blend of 50/50% ASIR/FBP. For each protocol in each patient, Z-axis coverage, patient dose-length product (DLP), and CT dose index volume (CTDI vol) from the late arterial phase imaging were recorded along with patient weight.
Table 1

CT and post processing parameters

 

Routine dose protocol

Low dose protocol

FBP

FBP

ASIR

Noise Index (on 0.625 slice thickness)

30

40

40

Reconstruction Algorithm

FBP

FBP

50% ASIR

Detector collimation

0.625

0.625

0.625

Reconstruction section thickness

2.5

2.5

2.5

Reconstruction section interval

2.5

2.5

2.5

Pitch

1.375

1.375

1.375

kVp

120

120

120

Tube Current (mA) control

ATCM

ATCM

ATCM

mA minimum/maximum setting

100/700

100/700

100/700

FBP Filtered Back Projection, ATCM Automatic Tube Current Modulation, ASIR Adaptive Statistical Iterative Reconstruction

Image analysis

To measure image noise using a commercially available workstation (Advantage Windows 4.4, GE Healthcare, Milwaukee, WI,USA) one of the authors placed three regions of interest (ROI) approximately 200 mm2 in size in the right lobe of the liver, carefully avoiding intra-hepatic vessels, hypervascular foci, and artifacts according to the method of Marin et al. [4]. A liver ROI was placed at the level of the portal bifurcation, at the level of the celiac artery, and at level of the origin of the superior mesenteric artery; mean noise was calculated. The noise in the abdominal aorta was measured once with an ROI in the central two thirds at the level of the main portal vein.

Subjective image quality scoring was performed by two experienced body imaging radiologists (10 and 13 years’ experience) reviewing together in consensus. Examinations were randomized for both patients and reconstructions and were de-identified of clinical information, protocol type, and reconstruction type. Reviewers looked at the 114 image sets (19 patients times two examinations per patient times three reconstructions per examination) in a random order using a standardized window/level setting of 400/40 with the monitor on a one-on-one cine format. For each image set, the reviewers graded image quality on a 5 point Likert scale: (1) Image quality considered poor and non-evaluable because of high image noise, marked artifact, distortion of spatial or contrast resolution, or poor edge definition; (2) Image quality considered fair but significantly compromised by moderate image noise, some image artifact, or some distortion of spatial resolution or contrast resolution; (3) Image quality considered good and only minimally compromised by image noise, artifact, or minimal distortion of spatial resolution or contrast resolution; (4) Image quality considered very good, displaying image noise typically encountered in our routine clinical practice, and not compromised by artifact or distortion of spatial or contrast resolution, and (5) Image quality considered superior, with striking lack of image noise and absence of any artifact or distortion. Reviewers were instructed to use scores of 1 or 2 if they considered an image set diagnostically unacceptable in a clinical setting and to use scores of 3, 4, or 5 if diagnostically acceptable.

Radiation dose assessment

Computed tomography dose index volume in milligray (mGy) and dose-length product were recorded from the dose page for each scan. Effective patient radiation dose in millisieverts (mSv) was estimated from the dose-length product for each protocol using a conversion factor for the abdomen of 0.015 [12].

Statistical analysis

Data were analyzed using Excel (version 12.2.3, Microsoft Corporation, Redmond, WA, USA) and SAS (version 9.2, SAS Corporation, Cary, NC, USA). Means and standard deviations for patient age, weight, z-axis length, time between the two protocols, and radiation dose were calculated. Overall differences in mean image noise and image quality among routine dose FBP images, low dose FBP images, and low dose ASIR images were evaluated with non-parametric Kruskal-Wallis tests followed by Dunn’s tests for pairwise comparisons. A p value of less than 0.05 was considered statistically significant.

Results

Twelve patients were male and seven were female. Mean patient age was 58 ± 9 years. Mean patient weight at the time of the routine dose protocol was 81 kg ± 16 kg (range 57 kg to 122 kg) and at the time of the low dose protocol was of 85 kg ± 16 kg (range 58 kg to 123 kg); there was no statistical difference in weight among individual patients between the routine dose protocol and the low dose protocol. None of the patients had developed significant ascites in the interval nor did they have a therapeutic intervention such as chemoembolization. Elapsed time in days between the routine dose protocol and low dose protocol averaged 184 days (6.1 months; range 42 days to 517 days).

Quantitative mean image noise in the liver and the aorta on the hepatic late arterial phase images was significantly lower for the routine dose protocol FPB images (NI = 30, 100% FBP reconstruction) compared with the low dose protocol FBP images (NI = 40, 100% FBP). There was significantly lower measured noise in both the liver and the aorta for the low dose ASIR images (NI 40, 50/50% ASIR/FBP) compared with the routine dose protocol FBP images (NI = 30, 100% FBP; Table 2).
Table 2

Image noise, subjective image quality scores, and effective dose for three image sets

 

Routine dose protocol

Low dose protocol

P Value

FBP

FBP

ASIR

Routine dose FBP vs. low dose FBP

Routine dose FBP vs. low dose ASIR

Image Noise (HU)

 Liver

25 ± 4

65 ± 10

20 ± 4

<0.001

05

 Aorta

30 ± 4

40 ± 9

24 ± 5

<0.001

<0.001

Image Quality Score

3.6 ± 0.5

2.1 ± 0.5

3.2 ± 0.8

<0.001

NS

CTDI Vol (mGy)

15.2 ± 10

9.1 ± 5

9.1 ± 5

<0.001

<0.001

Effective Dose (mSv)

7.5 ± 5.5

4.4 ± 2.4

4.4 ± 2.4

<0.001

<0.001

Data are means ± standard deviations. NS not significant

Subjective image quality mean scores for hepatic late arterial phase images showed a significant difference between the routine dose FBP images and the low dose FBP images (3.6 versus 2.1), with the mean of low dose FBP images in the diagnostically unacceptable range. There was no statistical difference between the mean image quality scores of routine dose FBP images and low dose ASIR images, with the mean image quality scores for both of these image sets within the diagnostically acceptable range (3.6 and 3.2; Table 2)

The Z axis image length for the routine dose protocol (25.8 ± 4.2 cm, range 17–28 cm) was not significantly different compared with the low dose protocol (25.3 ± 2.3 cm, range 21–29 cm; p < 0.05). The CTDI vol for the low dose protocol (9.1 ± 5.0 mGy) was significantly less than for the routine dose protocol (15.2 ± 10 mGy; 40% less, p < 0.001). The effective dose for the low dose protocol (4.4 ± 2.4 mSv) was significantly less than for the routine dose protocol (7.5 ± 5.5 mSv; 41% less dose, p < 0.001; Table 2). When patients were divided into three weight categories, reduction in patient radiation dose was least in the lowest weight category (35%) and similar in the higher two weight categories (45 and 43%; Table 3).
Table 3

Radiation dose reduction by weight category using routine and low dose CT protocols

Patient weight

CT protocol

CTDI Vol (mGy)

Effective dose (mSv)

<70 kg (n = 6)

Routine

10.4 ± 5.4

5.1 ± 3.4

Low Dose

6.8 ± 3.2

3.3 ± 1.8

Percentage Dose Reduction

35%

35%

70–90 kg (n = 6)

Routine

17.4 ± 15.9

8.7 ± 3.4

Low Dose

9.6 ± 4.9

4.8 ± 2.8

Percentage Dose Reduction

45%

45%

>90 kg (n = 7)

Routine

17.4 ± 6.4

7.9 ± 3.0

Low Dose

10.5 ± 5.4

4.5 ± 2.3

Percentage Dose Reduction

40%

43%

Data are means ± standard deviations

Discussion

Previous reports have stated that ASIR reconstruction has no discernable impact on low contrast resolution, on spatial resolution, or on edge sharpness [24, 10]. However, images reconstructed with 100% ASIR have such low noise and such a high degree of spatial correlation that they may appear to have a “waxy” texture, unfamiliar to radiologists accustomed to images reconstructed with FBP. To lessen this unfamiliar appearance, ASIR reconstruction of raw data in clinical practice is often blended with FBP reconstruction. We chose a 50/50% blend of ASIR/FBP reconstruction for our low dose clinical protocol based on our own early experience, on the CT manufacturer’s recommendation, and on recent publications [8, 10].

Noise index is a manufacturer-specific term, which is defined as the standard deviation of the noise in the central region of a CT image of a uniform 20-cm water phantom when imaged and reconstructed with a standard reconstruction algorithm. When a CT technologist is setting up imaging, the manual selection of a noise index number controls the range of the mA values over which the tube current varies during automated tube current modulation in order to approximate the predicted standard deviation of the noise for the resulting image [11]. With the advent of ASIR, we increased the noise index from 30 to 40 in our routine clinical abdominal CT protocol in order to lower patient radiation dose and also added the 50/50 blend of ASIR with FBP. In this investigation, the noise index selection was always based on the 0.625 mm slice thickness.

To compare routine dose (NI 30, FBP) and low dose (NI 40, 50% ASIR) clinical CT protocols, we identified a sequential group of liver cirrhosis patients who had undergone CT using a routine dose protocol and who then were examined by CT again at a later date using a low dose protocol. We found that CT images reconstructed with 100% FBP from our low dose protocol (NI 40) were significantly noisier, were scored lower for subjective image quality, and were scored within the diagnostically unacceptable range compared with images reconstructed with 100% FBP from our routine dose protocol (NI 30). Images from our low dose protocol when reconstructed with 50/50% ASIR/FBP had significantly lower noise than images from our routine dose protocol reconstructed with 100% FBP, were scored statistically similarly with regard to subjective image quality, and were both scored within the diagnostically acceptable range. There was a tendency for the subjective ASIR image quality scores to be slightly lower than the routine dose FBP images despite having less measured noise, likely due to the unfamiliar appearance of smoothing. In addition, we found that our low dose protocol resulted in a significantly lower patient radiation dose (41% less) than our routine dose protocol when the two protocols were performed at different times in the same patient. As there were no significant differences between the routine dose protocol and the low dose protocol with regard to Z axis image length and patient weight, the significantly lower level of patient radiation is most likely explained by the higher noise index (NI 40 versus NI 30).

Other investigators have found similar results for abdominal CT incorporating ASIR. Prakash et al. reported an average 25% dose reduction and significantly lower image noise with the use of ASIR reconstruction compared with FBP in a consecutive series of patients who had abdominal CT [2]. When they divided patients by weight categories, their results were similar to our findings. However, the compared examinations were not in the same patient and were not matched for Z-axis image length. Marin et al. also compared ASIR and FBP reconstruction and reported a twofold reduction in image noise when effective dose was normalized in abdominal CT imaging; when noise was normalized, ASIR resulted in a 70% dose reduction [4]. However, that study involved ten patients and utilized the low-kVp portion of a dual energy examination. Hara et al. looked at ASIR related dose reduction in an ACR phantom and in twelve patients who had previously undergone routine dose abdominal CT; they found dose reductions within the range of 32–65% [5]. This same group later reported on 53 patients who had both low dose ASIR and routine dose FBP abdominal CT with dose reductions of 23–66%. However, many of these compared examinations differed with regard to the CT manufacturer, slice thickness, BMI, and kVp [9]. In our series, the CT, slice thickness, and kVp were constant between the two CT examinations in the same patient and there was no statistical difference in patient weight.

This investigation has some limitations. First, this series is relatively small and is retrospective. These results should be tested in a larger multi-center prospective trial. Second, image review was performed in consensus between two reviewers. While this is a true study design limitation, we chose the consensus approach in part to encourage review focused on defined image quality parameters rather than on individual reaction to the unfamiliar appearance of the ASIR images. This consensus read approach precluded assessment of interobserver variability. Third, while reviewers were blinded to protocol, reconstruction technique, and patient characteristics, this blinding was not perfect because each reconstruction produced different image appearances. Fourth, this investigation did not attempt to assess lesion conspicuity with the different protocols and image sets. Rather, we chose to assess objective and subjective measurements of image quality and to compare patient radiation dose. Finally, these results only apply to the specified protocols, CT manufacturer, and patient population and may not apply to other protocols, vendors, or patient populations.

In conclusion, in the same patient, hepatic arterial phase CT images processed with a low dose protocol (NI = 40) reconstructed with ASIR may have lower measured noise, similar although slightly lower diagnostically acceptable image quality, and a 41% lower patient radiation dose compared with CT images processed with a routine dose protocol (NI = 30) reconstructed with FBP. These results suggest that low dose liver CT with ASIR may be a viable technique and is worthy of further study (Fig. 1).
https://static-content.springer.com/image/art%3A10.1007%2Fs00330-011-2186-3/MediaObjects/330_2011_2186_Fig1_HTML.gif
Fig. 1

63-year-old, 96 kg man with cirrhosis, 2.5 mm axial late hepatic arterial phase computed tomography images A Routine Dose Protocol: noise index of 30 reconstructed with FBP, Quality score 4, B Low Dose Protocol: noise index of 40 reconstructed with FBP, Quality score 3, C Low Dose Protocol: noise index of 40 reconstructed with a blend of 50/50% ASIR/FBP, Quality score 4

Acknowledgement

This work was supported in part by an unrestricted grant for clinical research from GE Healthcare.

Copyright information

© European Society of Radiology 2011