- Email: [email protected]
Feasibility of slice width reduction for spiral cranial computed tomography using iterative image reconstruction
European Journal of Radiology 83 (2014) 964–969
Contents lists available at ScienceDirect
European Journal of Radiology journal homepage: www.elsevier.com/locate/ejrad
Feasibility of slice width reduction for spiral cranial computed tomography using iterative image reconstruction Holger Haubenreisser a , Christian Fink a , John W. Nance Jr. a , Martin Sedlmair b , Bernhard Schmidt b , Stefan O. Schoenberg a , Thomas Henzler a,∗ a b
Institute of Clinical Radiology and Nuclear Medicine, University Medical Center Mannheim, Medical Faculty Mannheim, Heidelberg University, Germany Siemens Healthcare, Division CT, Forchheim, Germany
a r t i c l e
i n f o
Article history: Received 27 November 2013 Received in revised form 28 January 2014 Accepted 30 January 2014 Keywords: CT Iterative reconstruction Head Brain SAFIRE
a b s t r a c t Purpose: To prospectively compare image quality of cranial computed tomography (CCT) examinations with varying slice widths using traditional filtered back projection (FBP) versus sinogram-affirmed iterative image reconstruction (SAFIRE). Materials and methods: 29 consecutive patients (14 men, mean age: 72 ± 17 years) referred for a total of 40 CCT studies were prospectively included. Each CCT raw data set was reconstructed with FBP and SAFIRE at 5 slice widths (1–5 mm; 1 mm increments). Objective image quality was assessed in three predefined regions of the brain (white matter, thalamus, cerebellum) using identical regions of interest (ROIs). Subjective image quality was assessed by 2 experienced radiologists. Objective and subjective image quality parameters were statistically compared between FBP and SAFIRE reconstructions. Results: SAFIRE reconstructions resulted in mean noise reductions of 43.8% in the white matter, 45.6% in the thalamus and 42.0% in the cerebellum (p < 0.01) compared to FBP on non contrast-enhanced 1 mm slice width images. Corresponding mean noise reductions on 1 mm contrast-enhanced studies were 45.7%, 47.3%, and 45.0% in the white matter, thalamus, and cerebellum, respectively (p < 0.01). There was no significant difference in mean attenuation of any region or slice width between the two reconstruction methods (all p > 0.05). Subjective image quality of IR images was mostly rated higher than that of the FBP images. Conclusion: Compared to FBP, SAFIRE provides significant reductions in image noise while increasing subjective image in CCT, particularly when thinner slices are used. Therefore, SAFIRE may allow utilization of thinner slices in CCT, potentially reducing partial volume effects and improving diagnostic accuracy. © 2014 Elsevier Ireland Ltd. All rights reserved.
1. Introduction Cranial computed tomography (CCT) is the first-line imaging modality of choice to rule out intracranial hemorrhage [1] in patients with suspected stroke or trauma. Until recently, filtered back projection (FBP) was the clinical gold-standard for CT image reconstruction [2]. Traditionally, CCT examinations are reconstructed with relatively high image slice widths (approximately 3–5 mm) when using FBP reconstruction, as thinner slices are associated with linear increases in image noise. Increased noise is particularly detrimental when evaluating gray–white matter differentiation (due to the inherently low contrast between gray and
∗ Corresponding author at: Institute for Clinical Radiology and Nuclear Medicine University, Medical Center Mannheim, Medical Faculty Mannheim – Heidelberg University, Theodor-Kutzer-Ufer 1-3, D-68167 Mannheim, Germany. Tel.: +49 621 383 2067; fax: +49 621 383 1910. E-mail address: [email protected] (T. Henzler). 0720-048X/$ – see front matter © 2014 Elsevier Ireland Ltd. All rights reserved. http://dx.doi.org/10.1016/j.ejrad.2014.01.026
white matter) and the base of the skull (due to prominent beamhardening artifacts which can obscure cerebellar or brainstem pathology). Unfortunately, reconstruction slice widths thicker than 3 mm are prone to partial volume effects that can make assessment of brain hemorrhage challenging [3], possibly leading to follow-up CCT examinations in order to safely rule out hemorrhage. While it is technically possible to reduce image noise by increasing radiation dose, potentially allowing decreased slice width, CCT examinations are already associated with a high radiation dose [4], and further increases in ionizing radiation should be avoided. Iterative reconstruction (IR) was first used almost four decades ago [5]; however, computer power has only recently evolved enough to allow IR utilization within a clinically acceptable timeframe. The main benefit of most current IR methods is image noise reduction, which has implications for both improving image quality and decreasing radiation dose [6]. In addition, IR may allow the use of thinner slices while maintaining an acceptable level of image noise. For this study, we were able to use sinogram affirmed iterative reconstruction (SAFIRE), whereby the iterative loops are done
H. Haubenreisser et al. / European Journal of Radiology 83 (2014) 964–969
965
Fig. 1. ROI placement for the objective data analysis. ROIs were placed in the white matter, the thalamus and the cerebellum.
in the raw data space before being converted into the image space. This is in contrast to the previous generation iterative image reconstruction in image space (IRIS) where the data is first converted into the image space before the iterative loops for noise reduction are applied. Accordingly, the aim of this study was to prospectively compare objective and subjective measures of image quality in CCT examinations reconstructed at varying slice widths with FBP and SAFIRE. 2. Methods The study protocol was approved by our institutional review board and complies with the Declaration of Helsinki and the Health Insurance Portability and Accountability Act (HIPAA). Since we solely performed additional image reconstruction, the requirement for written informed consent was waived in all patients. 2.1. Patient cohort A total of 40 examinations between July 2011 and March 2012 were prospectively included in this study. 29 consecutive patients (14 men, mean age: 72 ± 17 years) that were referred to our institution for a CCT examination in this timeframe were enrolled in this study. All patients received a non contrast-enhanced CCT examination for the exclusion of hemorrhage and stroke. Additionally, 11 patients with known malignancy received an additional contrastenhanced examination to exclude brain metastases.
Germany). All examinations were performed using a spiral acquisition technique with the following scan parameters: 20 mm × 0.6 mm detector collimation (40 mm × 0.6 mm effective with z-shift), 280 ms gantry rotation time, and 420 mAs per rotation tube current time product. Tube voltage was fixed at 120 kV. Acquisition was cranio-caudal from the top of the cranium to the base of the cerebellum. Contrast-enhanced examinations were performed with 60–90 ml of iodinated contrast material (Imeron 400, Bracco Imaging S.p.A, Milan, Italy) followed by a 50 ml saline chaser, all injected at 2.5 ml/s through an 18-G intravenous antecubital catheter using a dual-syringe injector (Stellant D, Medrad, Indianola, PA). The scan delay was 5 min. 2.3. FBP and iterative reconstruction series CCT raw data was anonymized, exported to an external storage medium, and reconstructed using a prototype offline software application provided by the vendor. Each CCT examination was reconstructed using a conventional FBP algorithm and SAFIRE, a commercially available IR technique. A detailed description of the SAFIRE algorithm has been previously described [7]. Reconstruction parameters comprised of a standard cranial “H31s” FBP convolution kernel and a dedicated “J30s” IR kernel (SAFIRE strength 3). Each raw dataset was reconstructed at 5 slice widths (1–5 mm, 1 mm increments) with both FBP and SAFIRE. 2.4. Assessment of image noise, attenuation, and subjective image quality
2.2. CT image acquisition All patients were examined using a dual-source CT system (SOMATOM Definition, Siemens Healthcare Sector, Forchheim,
FBP and SAFIRE image datasets were transferred to an image viewing workstation (Aycan Osirix Pro [aycan Digitalsysteme GmbH, Wuerzburg, Germany]). In each data set, one observer ( )
Table 1 Subjective evaluation parameters and 4 point grading system. Rating
Subjective Noise
Sharpness
Diagnostic acceptability
Artifacts
1
Little to no noise
Fully acceptable
No artifacts
2
Optimum noise
Probably acceptable
Minor artifacts
3
Noisy, but permits evaluation
Noisy, degrades image so that no evaluation possible
Only acceptable under limited conditions Unacceptable
Major artifacts
4
Structures well defined with sharp contours Structures can be seen, contours are sharp enough for diagnostic reporting Structures can be seen, but not sufficiently for diagnostic reporting Structures cannot be defined
Major artifacts that make interpretation impossible
966
H. Haubenreisser et al. / European Journal of Radiology 83 (2014) 964–969
Table 2 Mean attenuation of FBP and IR images with different slice widths. Type of study
Slice width (mm)
Attenuation FBP
p IR
Non-contrast-enhanced
1 2 3 4 5
36.43 36.58 36.59 36.63 36.59
± ± ± ± ±
2.40 2.28 2.29 2.30 2.33
36.56 36.67 36.58 36.57 36.55
± ± ± ± ±
2.31 2.31 2.29 2.31 2.32
0.79 0.80 0.93 0.88 0.95
Contrast-enhanced
1 2 3 4 5
37.95 37.96 37.90 37.89 37.88
± ± ± ± ±
2.53 2.54 2.51 2.49 2.46
37.83 37.85 37.86 37.86 37.80
± ± ± ± ±
2.62 2.59 2.52 2.50 2.52
0.75 0.75 0.90 0.90 0.99
Fig. 2. Comparison of a non contrast-enhanced study reconstructed using 5 mm slice widths with both FBP (left) and IR (right) algorithms.
Table 3 Mean image noise of FBP and IR images with different slice widths. Type of study
Slice width (mm)
Noise
p
Non-contrast-enhanced
1 2 3 4 5
7.56 5.85 5.18 4.73 4.37
± ± ± ± ±
1.10 0.84 1.09 0.96 0.82
4.25 3.34 3.00 2.84 2.75
± ± ± ± ±
0.49 0.55 0.31 0.30 0.30
<0.01 <0.01 <0.01 <0.01 <0.01
Contrast-enhanced
1 2 3 4 5
7.97 6.13 5.10 4.66 4.36
± ± ± ± ±
1.43 1.06 0.28 0.26 0.24
4.30 3.45 3.03 2.85 2.72
± ± ± ± ±
0.26 0.25 0.25 0.27 0.24
<0.01 <0.01 <0.01 <0.01 <0.01
FBP
IR
Note: FPB, filtered back projection; IR, iterative reconstruction.
Table 4 Mean SNR of FBP and IR images with different slice widths. Type of study
Slice width (mm)
SNR
p
Non-contrast-enhanced
1 2 3 4 5
4.90 6.35 7.31 7.99 8.61
± ± ± ± ±
0.66 0.82 1.26 1.33 1.42
8.71 11.33 12.32 13.03 13.49
± ± ± ± ±
1.11 2.36 1.40 1.56 1.80
<0.01 <0.01 <0.01 <0.01 <0.01
Contrast-enhanced
1 2 3 4 5
4.88 6.33 7.44 8.14 8.79
± ± ± ± ±
0.76 0.95 0.37 0.41 0.49
8.82 11.01 12.56 13.40 13.96
± ± ± ± ±
0.60 0.86 1.16 1.30 1.24
<0.01 <0.01 <0.01 <0.01 <0.01
FBP
Note: FPB, filtered back projection; IR, iterative reconstruction.
IR
H. Haubenreisser et al. / European Journal of Radiology 83 (2014) 964–969
967
Fig. 3. Comparison of a non contrast-enhanced study reconstructed using 1 mm slice width and FBP (top left), 5 mm slice width and FBP (top right), 1 mm slice width and IR (bottom left), 5 mm slice width and IR (bottom right).
measured image noise, defined as the standard deviation of the measured Hounsfield units (HU), and the mean attenuation (signal) within identical circular regions of interest (ROIs) measuring 1 cm2 on axial images. The signal-to-noise (SNR) ratio was calculated using these measurements. ROIs were drawn in the white matter the thalamus and the cerebellum (Fig. 1), excluding pathology that could affect results, such as foreign bodies, blood products, and encephalomalacia. Subjective image quality was independently rated by two experienced radiologists ( ; ) with 10 years and 5 years experience in CCT, working independently, who were blinded to the reconstruction method and slice width. The subjective image criteria and 4-points grading system used are summarized in Table 1. 2.5. Statistical analysis Statistical analyses were performed using JMP 10.0 (SAS Institute Inc., Cary, NC, USA). Normally distributed data were identified using the Shapiro–Wilk W test. Continuous variables are presented as mean ± standard deviation and compared using an independent t-test for normally distributed data or a Mann–Whitney U-test for non-normally distributed data. Ordinal variables (image quality) are presented as median with 25% to 75% interquartile ranges and are compared using the Kruskal–Wallis analysis of variance. P-values < 0.05 were considered statistically significant.
Inter-observer agreements for subjective image quality were quantified using Cohen’s kappa.
3. Results All 40 studies were successfully completed and demonstrated diagnostic image quality at 5 mm. None of the studies suffered from a loss of image quality due to motion artifacts.
3.1. Objective image quality FBP and SAFIRE reconstructions did not demonstrate a significant difference in mean attenuation at any ROI or slice thickness on both contrast and non contrast-enhanced studies (all p > 0.05; Table 2). SAFIRE resulted in significantly decreased image noise compared to FBP reconstructions in all evaluated ROIs and at each slice width (p < 0.01; Table 3), with the highest relative noise reduction observed at 1 mm slice widths. The mean noise reduction achieved with the use of SAFIRE in the non contrast-enhanced studies was 43.8% in the white matter, 45.6% in the thalamus and 42.0% in the cerebellum (Figs. 2 and 3). Contrast-enhanced studies reconstructed with SAFIRE showed mean noise reductions of 45.7% in the white matter, 47.3% in the thalamus and 45.0% in the cerebellum (Fig. 4) at 1 mm slice width. SNR was significantly higher
Fig. 4. Comparison of a contrast-enhanced study reconstructed using 1 mm slice widths and both FBP (left) and IR (right) algorithms.
0.60 0.15 1.00 0.15 1.00 0.40 0.30 0.00 0.00 0.00 ± ± ± ± ± 1.82 1.09 1.00 1.00 1.00 0.65 0.46 0.00 0.20 0.00 ± ± ± ± ± 1.73 1.32 1.00 1.09 1.00 <0.01 0.06 0.62 0.07 1.00 0.54 0.40 0.20 0.00 0.00 ± ± ± ± ± 2.09 1.18 1.09 1.00 1.00 0.16 0.79 0.30 0.47 0.00
0.37 0.18 0.00 0.00 0.00 ± ± ± ± ± 1.38 1.07 1.00 1.00 1.00
IR
0.75 0.72 0.63 0.59 0.56 ± ± ± ± ± 2.38 1.72 1.33 1.24 1.17
FBP
Diagnostic acceptability
in SAFIRE images compared to FBP images (all p < 0.01). Detailed results for each slice width are summarized in Table 4. When comparing SAFIRE reconstructions performed at 1 mm to FBP reconstructions at 5 mm, mean attenuation, image noise and SNR did not differ statistically between the 2 slice widths (p = 0.413–0.996) (see Fig. 5).
± ± ± ± ±
0.38 0.15 0.00 0.00 0.00 ± ± ± ± ± 1.36 1.05 1.00 1.00 1.00
IR
0.80 0.64 0.42 0.49 0.30 ± ± ± ± ± 1.71 1.36 1.16 1.17 1.09
FBP
<0.01 <0.01 <0.01 0.01 0.08
Artifacts p
Fig. 5. Box and whisker plots demonstrate a significant lower image noise of iterative reconstructions compared to the corresponding FBP reconstructions in all the regions where measurements were done.
3.18 1.73 1.09 1.27 1.00
0.16 0.03 0.02 0.04 0.08
H. Haubenreisser et al. / European Journal of Radiology 83 (2014) 964–969
p
968
<0.01 <0.01 0.01 <0.01 0.03 0.50 0.41 0.00 0.00 0.00 ± ± ± ± ± 1.64 1.23 1.00 1.00 1.00 0.59 0.54 0.52 0.49 0.50 ± ± ± ± ± 2.5 2.09 1.45 1.59 1.36 <0.01 <0.01 <0.01 <0.01 <0.01 0.63 0.30 0.00 0.00 0.00 ± ± ± ± ± Note: FPB, filtered back projection; IR, iterative reconstruction.
0.32 0.58 0.00 0.35 0.00 ± ± ± ± ± 3.14 2.59 2.00 2.05 2.00 1 2 3 4 5 Contrast-enhanced
2.00 1.09 1.00 1.00 1.00
1.16 1.17 1.00 1.03 1.09 0.47 0.53 0.46 0.48 0.37 ± ± ± ± ± 1.79 1.12 1.03 1.02 1.03 0.45 0.50 0.51 0.53 0.57 ± ± ± ± ± 3.02 2.50 2.10 1.98 1.84 1 2 3 4 5 Non-contrast-enhanced
FBP
Slice width (mm) Type of study
Table 5 Subjective evaluation scores.
Subjective noise
IR
± ± ± ± ±
4. Discussion Our study demonstrated that SAFIRE provides increased subjective and objective image quality of both contrast- and non contrast-enhanced CCT examinations compared to FBP, especially in reconstructions using thinner slices. Our results suggest that improved image quality is primarily driven by decreased image noise, which is most apparent in inherently noisy thin slices. To maximize the noise reduction, we chose to use the newer SAFIRE IR algorithm as opposed to the IRIS IR algorithm from the same vendor. The former further reduces noise by incorporating iterative reconstruction loops in the raw data space, while the latter only utilizes the image data space for these reconstruction loops. Our findings are in accordance with multiple previously published studies that have demonstrated improved image quality of IR images in comparison to traditional FBP for various CT applications, including high resolution chest CT, coronary CT angiography and abdominal CT [7–12]. To the best of our knowledge, however, few studies have evaluated IR for CCT examinations [13–15]. Becker et al. [13] compared subjective and objective image quality between IR and FBP non contrast-enhanced CCT examinations using a dedicated head phantom as well as patient CCT examinations at a slice with of 5 mm. These were performed with standard
2.29 1.86 1.43 1.38 1.24
FBP
<0.01 <0.01 <0.01 <0.01 <0.01 0.39 0.29 0.13 0.09 0.13
p
Sharpness
IR
± ± ± ± ±
0.27 0.33 0.00 0.13 0.19
p
Both observers rated subjective noise, sharpness, and diagnostic acceptability of images higher when using SAFIRE compared to FBP at slice widths of 1–3 mm for non contrast-enhanced studies and 1–2 mm for contrast-enhanced studies (all p < 0.01). The reconstruction algorithm did not have a significant impact on the rating of artifacts at these slice widths (p = 0.03–0.6). At thicker slice widths of 4 mm and 5 mm, subjective image noise and sharpness was still rated better in the SAFIRE images compared to the FBP images (p < 0.05), while the ratings for diagnostic acceptability and the presence of artifacts did not differ significantly (p = 0.01–1.00). The visual scores for both readers are summarized in Table 5. Interobserver agreement was moderate for native studies (Cohen’s Ä = 0.551) and excellent for contrast-enhanced studies (Cohen’s Ä = 0.939).
<0.01 <0.01 <0.01 <0.01 0.07
3.2. Subjective image quality
H. Haubenreisser et al. / European Journal of Radiology 83 (2014) 964–969
and reduced doses and reconstructed with FBP and IR. The authors were able to demonstrate that IR allows a dose reduction of 20.2% (DLP) while significantly improving subjective image quality using the standard slice width of 5 mm. Rapalino et al. [15] also recently demonstrated potential radiation dose reductions with the use of IR in CCT examinations. In this study, CCT studies were reconstructed using a similar algorithm from another vendor (Adaptive Statistical Iterative Reconstruction (ASIR), GE Healthcare, Milwaukee, Wisconsin) resulted in an effective radiation dose of 1.95 mSv, compared to 2.66 mSv using a standard FBP protocol. Importantly, both SNR and contrast-to-noise ratio were significantly higher in the reduced dose IR datasets compared to the full dose FBP CCT examinations. Similarly, Kilic et al. [14] demonstrated a dose reduction from 2.3 mSv to 1.6 mSv utilizing a similar IR technique to Rapalino et al. However, in this study CNR was similar in both groups while the higher-dose FBP group displayed better SNR values. Subjective image quality was rated higher in the higher-dose FBP group, but diagnostic acceptability did not differ significantly. In contrast to the above-mentioned studies, our primary goal was to evaluate the feasibility IR for slice width reduction rather than radiation dose reduction. CCT has a very high sensitivity (91% in one study) to rule out subarachnoid hemorrhage but frequently leads to false positive results due to partial volume and beam hardening artifacts, especially at the skull base [16,17]. Slice width reduction is desirable to reduce partial volume effects; however, linear increases in image noise preclude diagnostic image quality with thin slices reconstructed using traditional FBP, especially given the marginal differences of densities of intracranial structures, most notable between gray and white matter. Like prior studies, we demonstrated improvements in image noise and subjective image quality using IR compared to FBP. Importantly, images reconstructed with SAFIRE and 1 mm slice thickness maintained similar to improved noise and image quality compared to images reconstructed with FBP and 5 mm slices, suggesting that SAFIRE may allow routine use of thinner image slice reconstructions without loss of diagnostic accuracy or increasing radiation dose. 4.1. Study limitations Our study has several limitations that must be considered. First, the patient cohort investigated in this feasibility study was relatively small. Second, we looked only at objective and subjective image quality rather than the effect of reconstruction method and slice thickness on diagnostic accuracy. Larger prospective studies are indicated to further elucidate the effects of thin slice IR reconstructions on diagnostic accuracy and patient outcomes. Finally, several studies have noted that the subjective appearance of images reconstructed with IR techniques appear unfamiliar, sometimes described as “waxy” or “plastic” [7–15]. While we did not explicitly evaluate this parameter, we feel that our subjective evaluation indirectly addressed this phenomenon; our results suggest that this either is not a significant detriment on subjective image quality or is compensated for by other improvements, notably reduced noise.
969
5. Conclusion SAFIRE allows CCT reconstructions with slice widths down to 1 mm while improving or maintaining the objective and subjective image quality of thicker FBP reconstructions. Given that there is no change in radiation dose and small lesions may be better visualized with thinner slices, the creation and use of smaller slice width reconstructions should be considered when SAFIRE is available. Conflict of interest None declared. References [1] Valovich McLeod TC. The prediction of intracranial injury after minor head trauma in the pediatric population. J Athl Train 2005;40(2): 123–5. [2] Ziegler A, Köhler T, Proksa R. Noise and resolution in images reconstructed with FBP and OSC algorithms for CT. Med Phys 2007;34(2):585–98. [3] Held P, Breit A. Comparison of CT and MRI methods in diagnosis of tumors of the para- and retropharyngeal space and temporal bone. Bildgebung 1994;61(4):263–71. [4] Bundesamt für Strahlenschutz. Bekanntmachung der aktualisierten diagnostischen Referenzwerte für diagnostische und interventionelle Röntgenuntersuchungen. Bundesanzeiger 2010;111:2594–6. [5] Hounsfield GN. Computerized transverse axial scanning (tomography): 1. Description of system. Br J Radiol 1973;46(552):1016–22. [6] Beister M, Kolditz D, Kalender WA. Iterative reconstruction methods in X-ray CT. Phys Med 2012;28(2):94–108. [7] Moscariello A, Takx RAP, Schoepf UJ, et al. Coronary CT angiography: image quality, diagnostic accuracy, and potential for radiation dose reduction using a novel iterative image reconstruction technique-comparison with traditional filtered back projection. Eur Radiol 2011;21(10):2130–8. [8] Henzler T, Hanley M, Arnoldi E, Bastarrika G, Schoepf UJ, Becker H-C. Practical strategies for low radiation dose cardiac computed tomography. J Thorac Imaging 2010;25(3):213–20. [9] Leipsic J, LaBounty TM, Heilbron B, et al. Estimated radiation dose reduction using adaptive statistical iterative reconstruction in coronary CT angiography: the ERASIR study. Am J Roentgenol 2010;195(3):655–60. [10] Prakash P, Kalra MK, Digumarthy SR, et al. Radiation dose reduction with chest computed tomography using adaptive statistical iterative reconstruction technique: initial experience. J Comput Assist Tomogr 2010;34(1):40–5. [11] Singh S, Kalra MK, Hsieh J, et al. Abdominal CT: comparison of adaptive statistical iterative and filtered back projection reconstruction techniques. Radiology 2010;257(2):373–83. [12] Renker M, Nance JW, Schoepf UJ, et al. Evaluation of heavily calcified vessels with coronary ct angiography: comparison of iterative and filtered back projection image reconstruction. Radiology 2011;260(2):390–9. [13] Becker H-C, Augart D, Karpitschka M, et al. Radiation exposure and image quality of normal computed tomography brain images acquired with automated and organ-based tube current modulation multiband filtering and iterative reconstruction. Invest Radiol 2012;47(3):202–7. [14] Kilic K, Erbas G, Guryildirim M, Arac M, Ilgit E, Coskun B. Lowering the dose in head CT using adaptive statistical iterative reconstruction. AJNR Am J Neuroradiol 2011;32(9):1578–82. [15] Rapalino O, Kamalian S, Kamalian S, et al. Cranial CT with adaptive statistical iterative reconstruction: improved image quality with concomitant radiation dose reduction. AJNR Am J Neuroradiol 2012;33(4):609–15. [16] McCormack RF, Hutson A. Can computed tomography angiography of the brain replace lumbar puncture in the evaluation of acute-onset headache after a negative noncontrast cranial computed tomography scan? Acad Emerg Med 2010;17(4):444–51. [17] Greenberg SM, Vernooij MW, Cordonnier C, et al. Cerebral microbleeds: a field guide to their detection and interpretation. Lancet Neurol 2009;8(2):165–74.
Contents lists available at ScienceDirect
European Journal of Radiology journal homepage: www.elsevier.com/locate/ejrad
Feasibility of slice width reduction for spiral cranial computed tomography using iterative image reconstruction Holger Haubenreisser a , Christian Fink a , John W. Nance Jr. a , Martin Sedlmair b , Bernhard Schmidt b , Stefan O. Schoenberg a , Thomas Henzler a,∗ a b
Institute of Clinical Radiology and Nuclear Medicine, University Medical Center Mannheim, Medical Faculty Mannheim, Heidelberg University, Germany Siemens Healthcare, Division CT, Forchheim, Germany
a r t i c l e
i n f o
Article history: Received 27 November 2013 Received in revised form 28 January 2014 Accepted 30 January 2014 Keywords: CT Iterative reconstruction Head Brain SAFIRE
a b s t r a c t Purpose: To prospectively compare image quality of cranial computed tomography (CCT) examinations with varying slice widths using traditional filtered back projection (FBP) versus sinogram-affirmed iterative image reconstruction (SAFIRE). Materials and methods: 29 consecutive patients (14 men, mean age: 72 ± 17 years) referred for a total of 40 CCT studies were prospectively included. Each CCT raw data set was reconstructed with FBP and SAFIRE at 5 slice widths (1–5 mm; 1 mm increments). Objective image quality was assessed in three predefined regions of the brain (white matter, thalamus, cerebellum) using identical regions of interest (ROIs). Subjective image quality was assessed by 2 experienced radiologists. Objective and subjective image quality parameters were statistically compared between FBP and SAFIRE reconstructions. Results: SAFIRE reconstructions resulted in mean noise reductions of 43.8% in the white matter, 45.6% in the thalamus and 42.0% in the cerebellum (p < 0.01) compared to FBP on non contrast-enhanced 1 mm slice width images. Corresponding mean noise reductions on 1 mm contrast-enhanced studies were 45.7%, 47.3%, and 45.0% in the white matter, thalamus, and cerebellum, respectively (p < 0.01). There was no significant difference in mean attenuation of any region or slice width between the two reconstruction methods (all p > 0.05). Subjective image quality of IR images was mostly rated higher than that of the FBP images. Conclusion: Compared to FBP, SAFIRE provides significant reductions in image noise while increasing subjective image in CCT, particularly when thinner slices are used. Therefore, SAFIRE may allow utilization of thinner slices in CCT, potentially reducing partial volume effects and improving diagnostic accuracy. © 2014 Elsevier Ireland Ltd. All rights reserved.
1. Introduction Cranial computed tomography (CCT) is the first-line imaging modality of choice to rule out intracranial hemorrhage [1] in patients with suspected stroke or trauma. Until recently, filtered back projection (FBP) was the clinical gold-standard for CT image reconstruction [2]. Traditionally, CCT examinations are reconstructed with relatively high image slice widths (approximately 3–5 mm) when using FBP reconstruction, as thinner slices are associated with linear increases in image noise. Increased noise is particularly detrimental when evaluating gray–white matter differentiation (due to the inherently low contrast between gray and
∗ Corresponding author at: Institute for Clinical Radiology and Nuclear Medicine University, Medical Center Mannheim, Medical Faculty Mannheim – Heidelberg University, Theodor-Kutzer-Ufer 1-3, D-68167 Mannheim, Germany. Tel.: +49 621 383 2067; fax: +49 621 383 1910. E-mail address: [email protected] (T. Henzler). 0720-048X/$ – see front matter © 2014 Elsevier Ireland Ltd. All rights reserved. http://dx.doi.org/10.1016/j.ejrad.2014.01.026
white matter) and the base of the skull (due to prominent beamhardening artifacts which can obscure cerebellar or brainstem pathology). Unfortunately, reconstruction slice widths thicker than 3 mm are prone to partial volume effects that can make assessment of brain hemorrhage challenging [3], possibly leading to follow-up CCT examinations in order to safely rule out hemorrhage. While it is technically possible to reduce image noise by increasing radiation dose, potentially allowing decreased slice width, CCT examinations are already associated with a high radiation dose [4], and further increases in ionizing radiation should be avoided. Iterative reconstruction (IR) was first used almost four decades ago [5]; however, computer power has only recently evolved enough to allow IR utilization within a clinically acceptable timeframe. The main benefit of most current IR methods is image noise reduction, which has implications for both improving image quality and decreasing radiation dose [6]. In addition, IR may allow the use of thinner slices while maintaining an acceptable level of image noise. For this study, we were able to use sinogram affirmed iterative reconstruction (SAFIRE), whereby the iterative loops are done
H. Haubenreisser et al. / European Journal of Radiology 83 (2014) 964–969
965
Fig. 1. ROI placement for the objective data analysis. ROIs were placed in the white matter, the thalamus and the cerebellum.
in the raw data space before being converted into the image space. This is in contrast to the previous generation iterative image reconstruction in image space (IRIS) where the data is first converted into the image space before the iterative loops for noise reduction are applied. Accordingly, the aim of this study was to prospectively compare objective and subjective measures of image quality in CCT examinations reconstructed at varying slice widths with FBP and SAFIRE. 2. Methods The study protocol was approved by our institutional review board and complies with the Declaration of Helsinki and the Health Insurance Portability and Accountability Act (HIPAA). Since we solely performed additional image reconstruction, the requirement for written informed consent was waived in all patients. 2.1. Patient cohort A total of 40 examinations between July 2011 and March 2012 were prospectively included in this study. 29 consecutive patients (14 men, mean age: 72 ± 17 years) that were referred to our institution for a CCT examination in this timeframe were enrolled in this study. All patients received a non contrast-enhanced CCT examination for the exclusion of hemorrhage and stroke. Additionally, 11 patients with known malignancy received an additional contrastenhanced examination to exclude brain metastases.
Germany). All examinations were performed using a spiral acquisition technique with the following scan parameters: 20 mm × 0.6 mm detector collimation (40 mm × 0.6 mm effective with z-shift), 280 ms gantry rotation time, and 420 mAs per rotation tube current time product. Tube voltage was fixed at 120 kV. Acquisition was cranio-caudal from the top of the cranium to the base of the cerebellum. Contrast-enhanced examinations were performed with 60–90 ml of iodinated contrast material (Imeron 400, Bracco Imaging S.p.A, Milan, Italy) followed by a 50 ml saline chaser, all injected at 2.5 ml/s through an 18-G intravenous antecubital catheter using a dual-syringe injector (Stellant D, Medrad, Indianola, PA). The scan delay was 5 min. 2.3. FBP and iterative reconstruction series CCT raw data was anonymized, exported to an external storage medium, and reconstructed using a prototype offline software application provided by the vendor. Each CCT examination was reconstructed using a conventional FBP algorithm and SAFIRE, a commercially available IR technique. A detailed description of the SAFIRE algorithm has been previously described [7]. Reconstruction parameters comprised of a standard cranial “H31s” FBP convolution kernel and a dedicated “J30s” IR kernel (SAFIRE strength 3). Each raw dataset was reconstructed at 5 slice widths (1–5 mm, 1 mm increments) with both FBP and SAFIRE. 2.4. Assessment of image noise, attenuation, and subjective image quality
2.2. CT image acquisition All patients were examined using a dual-source CT system (SOMATOM Definition, Siemens Healthcare Sector, Forchheim,
FBP and SAFIRE image datasets were transferred to an image viewing workstation (Aycan Osirix Pro [aycan Digitalsysteme GmbH, Wuerzburg, Germany]). In each data set, one observer ( )
Table 1 Subjective evaluation parameters and 4 point grading system. Rating
Subjective Noise
Sharpness
Diagnostic acceptability
Artifacts
1
Little to no noise
Fully acceptable
No artifacts
2
Optimum noise
Probably acceptable
Minor artifacts
3
Noisy, but permits evaluation
Noisy, degrades image so that no evaluation possible
Only acceptable under limited conditions Unacceptable
Major artifacts
4
Structures well defined with sharp contours Structures can be seen, contours are sharp enough for diagnostic reporting Structures can be seen, but not sufficiently for diagnostic reporting Structures cannot be defined
Major artifacts that make interpretation impossible
966
H. Haubenreisser et al. / European Journal of Radiology 83 (2014) 964–969
Table 2 Mean attenuation of FBP and IR images with different slice widths. Type of study
Slice width (mm)
Attenuation FBP
p IR
Non-contrast-enhanced
1 2 3 4 5
36.43 36.58 36.59 36.63 36.59
± ± ± ± ±
2.40 2.28 2.29 2.30 2.33
36.56 36.67 36.58 36.57 36.55
± ± ± ± ±
2.31 2.31 2.29 2.31 2.32
0.79 0.80 0.93 0.88 0.95
Contrast-enhanced
1 2 3 4 5
37.95 37.96 37.90 37.89 37.88
± ± ± ± ±
2.53 2.54 2.51 2.49 2.46
37.83 37.85 37.86 37.86 37.80
± ± ± ± ±
2.62 2.59 2.52 2.50 2.52
0.75 0.75 0.90 0.90 0.99
Fig. 2. Comparison of a non contrast-enhanced study reconstructed using 5 mm slice widths with both FBP (left) and IR (right) algorithms.
Table 3 Mean image noise of FBP and IR images with different slice widths. Type of study
Slice width (mm)
Noise
p
Non-contrast-enhanced
1 2 3 4 5
7.56 5.85 5.18 4.73 4.37
± ± ± ± ±
1.10 0.84 1.09 0.96 0.82
4.25 3.34 3.00 2.84 2.75
± ± ± ± ±
0.49 0.55 0.31 0.30 0.30
<0.01 <0.01 <0.01 <0.01 <0.01
Contrast-enhanced
1 2 3 4 5
7.97 6.13 5.10 4.66 4.36
± ± ± ± ±
1.43 1.06 0.28 0.26 0.24
4.30 3.45 3.03 2.85 2.72
± ± ± ± ±
0.26 0.25 0.25 0.27 0.24
<0.01 <0.01 <0.01 <0.01 <0.01
FBP
IR
Note: FPB, filtered back projection; IR, iterative reconstruction.
Table 4 Mean SNR of FBP and IR images with different slice widths. Type of study
Slice width (mm)
SNR
p
Non-contrast-enhanced
1 2 3 4 5
4.90 6.35 7.31 7.99 8.61
± ± ± ± ±
0.66 0.82 1.26 1.33 1.42
8.71 11.33 12.32 13.03 13.49
± ± ± ± ±
1.11 2.36 1.40 1.56 1.80
<0.01 <0.01 <0.01 <0.01 <0.01
Contrast-enhanced
1 2 3 4 5
4.88 6.33 7.44 8.14 8.79
± ± ± ± ±
0.76 0.95 0.37 0.41 0.49
8.82 11.01 12.56 13.40 13.96
± ± ± ± ±
0.60 0.86 1.16 1.30 1.24
<0.01 <0.01 <0.01 <0.01 <0.01
FBP
Note: FPB, filtered back projection; IR, iterative reconstruction.
IR
H. Haubenreisser et al. / European Journal of Radiology 83 (2014) 964–969
967
Fig. 3. Comparison of a non contrast-enhanced study reconstructed using 1 mm slice width and FBP (top left), 5 mm slice width and FBP (top right), 1 mm slice width and IR (bottom left), 5 mm slice width and IR (bottom right).
measured image noise, defined as the standard deviation of the measured Hounsfield units (HU), and the mean attenuation (signal) within identical circular regions of interest (ROIs) measuring 1 cm2 on axial images. The signal-to-noise (SNR) ratio was calculated using these measurements. ROIs were drawn in the white matter the thalamus and the cerebellum (Fig. 1), excluding pathology that could affect results, such as foreign bodies, blood products, and encephalomalacia. Subjective image quality was independently rated by two experienced radiologists ( ; ) with 10 years and 5 years experience in CCT, working independently, who were blinded to the reconstruction method and slice width. The subjective image criteria and 4-points grading system used are summarized in Table 1. 2.5. Statistical analysis Statistical analyses were performed using JMP 10.0 (SAS Institute Inc., Cary, NC, USA). Normally distributed data were identified using the Shapiro–Wilk W test. Continuous variables are presented as mean ± standard deviation and compared using an independent t-test for normally distributed data or a Mann–Whitney U-test for non-normally distributed data. Ordinal variables (image quality) are presented as median with 25% to 75% interquartile ranges and are compared using the Kruskal–Wallis analysis of variance. P-values < 0.05 were considered statistically significant.
Inter-observer agreements for subjective image quality were quantified using Cohen’s kappa.
3. Results All 40 studies were successfully completed and demonstrated diagnostic image quality at 5 mm. None of the studies suffered from a loss of image quality due to motion artifacts.
3.1. Objective image quality FBP and SAFIRE reconstructions did not demonstrate a significant difference in mean attenuation at any ROI or slice thickness on both contrast and non contrast-enhanced studies (all p > 0.05; Table 2). SAFIRE resulted in significantly decreased image noise compared to FBP reconstructions in all evaluated ROIs and at each slice width (p < 0.01; Table 3), with the highest relative noise reduction observed at 1 mm slice widths. The mean noise reduction achieved with the use of SAFIRE in the non contrast-enhanced studies was 43.8% in the white matter, 45.6% in the thalamus and 42.0% in the cerebellum (Figs. 2 and 3). Contrast-enhanced studies reconstructed with SAFIRE showed mean noise reductions of 45.7% in the white matter, 47.3% in the thalamus and 45.0% in the cerebellum (Fig. 4) at 1 mm slice width. SNR was significantly higher
Fig. 4. Comparison of a contrast-enhanced study reconstructed using 1 mm slice widths and both FBP (left) and IR (right) algorithms.
0.60 0.15 1.00 0.15 1.00 0.40 0.30 0.00 0.00 0.00 ± ± ± ± ± 1.82 1.09 1.00 1.00 1.00 0.65 0.46 0.00 0.20 0.00 ± ± ± ± ± 1.73 1.32 1.00 1.09 1.00 <0.01 0.06 0.62 0.07 1.00 0.54 0.40 0.20 0.00 0.00 ± ± ± ± ± 2.09 1.18 1.09 1.00 1.00 0.16 0.79 0.30 0.47 0.00
0.37 0.18 0.00 0.00 0.00 ± ± ± ± ± 1.38 1.07 1.00 1.00 1.00
IR
0.75 0.72 0.63 0.59 0.56 ± ± ± ± ± 2.38 1.72 1.33 1.24 1.17
FBP
Diagnostic acceptability
in SAFIRE images compared to FBP images (all p < 0.01). Detailed results for each slice width are summarized in Table 4. When comparing SAFIRE reconstructions performed at 1 mm to FBP reconstructions at 5 mm, mean attenuation, image noise and SNR did not differ statistically between the 2 slice widths (p = 0.413–0.996) (see Fig. 5).
± ± ± ± ±
0.38 0.15 0.00 0.00 0.00 ± ± ± ± ± 1.36 1.05 1.00 1.00 1.00
IR
0.80 0.64 0.42 0.49 0.30 ± ± ± ± ± 1.71 1.36 1.16 1.17 1.09
FBP
<0.01 <0.01 <0.01 0.01 0.08
Artifacts p
Fig. 5. Box and whisker plots demonstrate a significant lower image noise of iterative reconstructions compared to the corresponding FBP reconstructions in all the regions where measurements were done.
3.18 1.73 1.09 1.27 1.00
0.16 0.03 0.02 0.04 0.08
H. Haubenreisser et al. / European Journal of Radiology 83 (2014) 964–969
p
968
<0.01 <0.01 0.01 <0.01 0.03 0.50 0.41 0.00 0.00 0.00 ± ± ± ± ± 1.64 1.23 1.00 1.00 1.00 0.59 0.54 0.52 0.49 0.50 ± ± ± ± ± 2.5 2.09 1.45 1.59 1.36 <0.01 <0.01 <0.01 <0.01 <0.01 0.63 0.30 0.00 0.00 0.00 ± ± ± ± ± Note: FPB, filtered back projection; IR, iterative reconstruction.
0.32 0.58 0.00 0.35 0.00 ± ± ± ± ± 3.14 2.59 2.00 2.05 2.00 1 2 3 4 5 Contrast-enhanced
2.00 1.09 1.00 1.00 1.00
1.16 1.17 1.00 1.03 1.09 0.47 0.53 0.46 0.48 0.37 ± ± ± ± ± 1.79 1.12 1.03 1.02 1.03 0.45 0.50 0.51 0.53 0.57 ± ± ± ± ± 3.02 2.50 2.10 1.98 1.84 1 2 3 4 5 Non-contrast-enhanced
FBP
Slice width (mm) Type of study
Table 5 Subjective evaluation scores.
Subjective noise
IR
± ± ± ± ±
4. Discussion Our study demonstrated that SAFIRE provides increased subjective and objective image quality of both contrast- and non contrast-enhanced CCT examinations compared to FBP, especially in reconstructions using thinner slices. Our results suggest that improved image quality is primarily driven by decreased image noise, which is most apparent in inherently noisy thin slices. To maximize the noise reduction, we chose to use the newer SAFIRE IR algorithm as opposed to the IRIS IR algorithm from the same vendor. The former further reduces noise by incorporating iterative reconstruction loops in the raw data space, while the latter only utilizes the image data space for these reconstruction loops. Our findings are in accordance with multiple previously published studies that have demonstrated improved image quality of IR images in comparison to traditional FBP for various CT applications, including high resolution chest CT, coronary CT angiography and abdominal CT [7–12]. To the best of our knowledge, however, few studies have evaluated IR for CCT examinations [13–15]. Becker et al. [13] compared subjective and objective image quality between IR and FBP non contrast-enhanced CCT examinations using a dedicated head phantom as well as patient CCT examinations at a slice with of 5 mm. These were performed with standard
2.29 1.86 1.43 1.38 1.24
FBP
<0.01 <0.01 <0.01 <0.01 <0.01 0.39 0.29 0.13 0.09 0.13
p
Sharpness
IR
± ± ± ± ±
0.27 0.33 0.00 0.13 0.19
p
Both observers rated subjective noise, sharpness, and diagnostic acceptability of images higher when using SAFIRE compared to FBP at slice widths of 1–3 mm for non contrast-enhanced studies and 1–2 mm for contrast-enhanced studies (all p < 0.01). The reconstruction algorithm did not have a significant impact on the rating of artifacts at these slice widths (p = 0.03–0.6). At thicker slice widths of 4 mm and 5 mm, subjective image noise and sharpness was still rated better in the SAFIRE images compared to the FBP images (p < 0.05), while the ratings for diagnostic acceptability and the presence of artifacts did not differ significantly (p = 0.01–1.00). The visual scores for both readers are summarized in Table 5. Interobserver agreement was moderate for native studies (Cohen’s Ä = 0.551) and excellent for contrast-enhanced studies (Cohen’s Ä = 0.939).
<0.01 <0.01 <0.01 <0.01 0.07
3.2. Subjective image quality
H. Haubenreisser et al. / European Journal of Radiology 83 (2014) 964–969
and reduced doses and reconstructed with FBP and IR. The authors were able to demonstrate that IR allows a dose reduction of 20.2% (DLP) while significantly improving subjective image quality using the standard slice width of 5 mm. Rapalino et al. [15] also recently demonstrated potential radiation dose reductions with the use of IR in CCT examinations. In this study, CCT studies were reconstructed using a similar algorithm from another vendor (Adaptive Statistical Iterative Reconstruction (ASIR), GE Healthcare, Milwaukee, Wisconsin) resulted in an effective radiation dose of 1.95 mSv, compared to 2.66 mSv using a standard FBP protocol. Importantly, both SNR and contrast-to-noise ratio were significantly higher in the reduced dose IR datasets compared to the full dose FBP CCT examinations. Similarly, Kilic et al. [14] demonstrated a dose reduction from 2.3 mSv to 1.6 mSv utilizing a similar IR technique to Rapalino et al. However, in this study CNR was similar in both groups while the higher-dose FBP group displayed better SNR values. Subjective image quality was rated higher in the higher-dose FBP group, but diagnostic acceptability did not differ significantly. In contrast to the above-mentioned studies, our primary goal was to evaluate the feasibility IR for slice width reduction rather than radiation dose reduction. CCT has a very high sensitivity (91% in one study) to rule out subarachnoid hemorrhage but frequently leads to false positive results due to partial volume and beam hardening artifacts, especially at the skull base [16,17]. Slice width reduction is desirable to reduce partial volume effects; however, linear increases in image noise preclude diagnostic image quality with thin slices reconstructed using traditional FBP, especially given the marginal differences of densities of intracranial structures, most notable between gray and white matter. Like prior studies, we demonstrated improvements in image noise and subjective image quality using IR compared to FBP. Importantly, images reconstructed with SAFIRE and 1 mm slice thickness maintained similar to improved noise and image quality compared to images reconstructed with FBP and 5 mm slices, suggesting that SAFIRE may allow routine use of thinner image slice reconstructions without loss of diagnostic accuracy or increasing radiation dose. 4.1. Study limitations Our study has several limitations that must be considered. First, the patient cohort investigated in this feasibility study was relatively small. Second, we looked only at objective and subjective image quality rather than the effect of reconstruction method and slice thickness on diagnostic accuracy. Larger prospective studies are indicated to further elucidate the effects of thin slice IR reconstructions on diagnostic accuracy and patient outcomes. Finally, several studies have noted that the subjective appearance of images reconstructed with IR techniques appear unfamiliar, sometimes described as “waxy” or “plastic” [7–15]. While we did not explicitly evaluate this parameter, we feel that our subjective evaluation indirectly addressed this phenomenon; our results suggest that this either is not a significant detriment on subjective image quality or is compensated for by other improvements, notably reduced noise.
969
5. Conclusion SAFIRE allows CCT reconstructions with slice widths down to 1 mm while improving or maintaining the objective and subjective image quality of thicker FBP reconstructions. Given that there is no change in radiation dose and small lesions may be better visualized with thinner slices, the creation and use of smaller slice width reconstructions should be considered when SAFIRE is available. Conflict of interest None declared. References [1] Valovich McLeod TC. The prediction of intracranial injury after minor head trauma in the pediatric population. J Athl Train 2005;40(2): 123–5. [2] Ziegler A, Köhler T, Proksa R. Noise and resolution in images reconstructed with FBP and OSC algorithms for CT. Med Phys 2007;34(2):585–98. [3] Held P, Breit A. Comparison of CT and MRI methods in diagnosis of tumors of the para- and retropharyngeal space and temporal bone. Bildgebung 1994;61(4):263–71. [4] Bundesamt für Strahlenschutz. Bekanntmachung der aktualisierten diagnostischen Referenzwerte für diagnostische und interventionelle Röntgenuntersuchungen. Bundesanzeiger 2010;111:2594–6. [5] Hounsfield GN. Computerized transverse axial scanning (tomography): 1. Description of system. Br J Radiol 1973;46(552):1016–22. [6] Beister M, Kolditz D, Kalender WA. Iterative reconstruction methods in X-ray CT. Phys Med 2012;28(2):94–108. [7] Moscariello A, Takx RAP, Schoepf UJ, et al. Coronary CT angiography: image quality, diagnostic accuracy, and potential for radiation dose reduction using a novel iterative image reconstruction technique-comparison with traditional filtered back projection. Eur Radiol 2011;21(10):2130–8. [8] Henzler T, Hanley M, Arnoldi E, Bastarrika G, Schoepf UJ, Becker H-C. Practical strategies for low radiation dose cardiac computed tomography. J Thorac Imaging 2010;25(3):213–20. [9] Leipsic J, LaBounty TM, Heilbron B, et al. Estimated radiation dose reduction using adaptive statistical iterative reconstruction in coronary CT angiography: the ERASIR study. Am J Roentgenol 2010;195(3):655–60. [10] Prakash P, Kalra MK, Digumarthy SR, et al. Radiation dose reduction with chest computed tomography using adaptive statistical iterative reconstruction technique: initial experience. J Comput Assist Tomogr 2010;34(1):40–5. [11] Singh S, Kalra MK, Hsieh J, et al. Abdominal CT: comparison of adaptive statistical iterative and filtered back projection reconstruction techniques. Radiology 2010;257(2):373–83. [12] Renker M, Nance JW, Schoepf UJ, et al. Evaluation of heavily calcified vessels with coronary ct angiography: comparison of iterative and filtered back projection image reconstruction. Radiology 2011;260(2):390–9. [13] Becker H-C, Augart D, Karpitschka M, et al. Radiation exposure and image quality of normal computed tomography brain images acquired with automated and organ-based tube current modulation multiband filtering and iterative reconstruction. Invest Radiol 2012;47(3):202–7. [14] Kilic K, Erbas G, Guryildirim M, Arac M, Ilgit E, Coskun B. Lowering the dose in head CT using adaptive statistical iterative reconstruction. AJNR Am J Neuroradiol 2011;32(9):1578–82. [15] Rapalino O, Kamalian S, Kamalian S, et al. Cranial CT with adaptive statistical iterative reconstruction: improved image quality with concomitant radiation dose reduction. AJNR Am J Neuroradiol 2012;33(4):609–15. [16] McCormack RF, Hutson A. Can computed tomography angiography of the brain replace lumbar puncture in the evaluation of acute-onset headache after a negative noncontrast cranial computed tomography scan? Acad Emerg Med 2010;17(4):444–51. [17] Greenberg SM, Vernooij MW, Cordonnier C, et al. Cerebral microbleeds: a field guide to their detection and interpretation. Lancet Neurol 2009;8(2):165–74.