Abstract
Background and purpose
To evaluate the added value of a model-based reconstruction algorithm in the assessment of acute traumatic brain lesions in emergency non-enhanced computed tomography, in comparison with a standard hybrid iterative reconstruction approach.
Materials and methods
We retrospectively evaluated a total of 350 patients who underwent a 256-row non-enhanced computed tomography scan at the emergency department for brain trauma. Images were reconstructed both with hybrid and model-based iterative algorithm. Two radiologists, blinded to clinical data, recorded the presence, nature, number, and location of acute findings. Subjective image quality was performed using a 4-point scale. Objective image quality was determined by computing the signal-to-noise ratio and contrast-to-noise ratio. The agreement between the two readers was evaluated using k-statistics.
Results
A subjective image quality analysis using model-based iterative reconstruction gave a higher detection rate of acute trauma-related lesions in comparison to hybrid iterative reconstruction (extradural haematomas 116 vs. 68, subdural haemorrhages 162 vs. 98, subarachnoid haemorrhages 118 vs. 78, parenchymal haemorrhages 94 vs. 64, contusive lesions 36 vs. 28, diffuse axonal injuries 75 vs. 31; all P<0.001). Inter-observer agreement was moderate to excellent in evaluating all injuries (extradural haematomas k=0.79, subdural haemorrhages k=0.82, subarachnoid haemorrhages k=0.91, parenchymal haemorrhages k=0.98, contusive lesions k=0.88, diffuse axonal injuries k=0.70). Quantitatively, the mean standard deviation of the thalamus on model-based iterative reconstruction images was lower in comparison to hybrid iterative one (2.12 ± 0.92 vsa 3.52 ± 1.10; P=0.030) while the contrast-to-noise ratio and signal-to-noise ratio were significantly higher (contrast-to-noise ratio 3.06 ± 0.55 vs. 1.55 ± 0.68, signal-to-noise ratio 14.51 ± 1.78 vs. 8.62 ± 1.88; P<0.0001). Median subjective image quality values for model-based iterative reconstruction were significantly higher (P=0.003).
Conclusion
Model-based iterative reconstruction, offering a higher image quality at a thinner slice, allowed the identification of a higher number of acute traumatic lesions than hybrid iterative reconstruction, with a significant reduction of noise.
Keywords: Multidetector computed tomography, knowledge bases, traumatic brain injuries
Introduction
Non-enhanced computed tomography (NECT) is recommended in the emergency setting to investigate patients affected by traumatic brain injury (TBI).1–3 It can determine the presence and extent of the damage, guide clinical management and eventually surgical planning and, together with magnetic resonance imaging (MRI), has a main role in assessing prognosis, in identifying chronic sequelae of TBI and in guiding rehabilitation. 4
Even if radiation dose reduction is not a primary concern in the management of acute trauma patients, the possibility of reducing the radiation dose is still an important issue.
In the past few years, several strategies have been used to reduce radiation dose exposure, related both to acquisition parameters or to reconstruction algorithms applied to raw data. Moreover, the combination of both of those strategies has allowed us to achieve a correct balance between a reduction of radiation dose exposure, linked to the increase in image noise, and consequently to the final image quality.5–7
The image noise represents a particularly significant issue in brain, computed tomography (CT), as the attenuation difference between grey and white matter regions is as low as 5–10 Hounsfield units (HU), and the noise level is also increased by artefacts of beam-hardening due to the skull base interface between bone and brain matter, particularly in the posterior fossa as well. 8
The most diffused brain CT protocols, reconstructed either with filtered back-projection (FBP) or hybrid iterative reconstruction (HIR), are employed with a reconstruction thickness of 5 mm or 4.5 mm, representing a good compromise between noise and thickness. As the thinner is the slice the higher is the noise; indeed, if those algorithms are applied to a thinner slice (e.g. 2 mm), the final image noise would be too high, resulting in low diagnostic quality.9–11
In the past few years, different vendors have developed new purely iterative reconstruction algorithms, the so-called model-based iterative reconstruction (MBIR) models, characterised by a more powerful noise reduction skill, that can be employed to optimise image quality and simultaneously reduce the radiation dose exposure. Thanks to a more powerful reduction of the image noise, these models can lead to a reduction in the noise of thin slice brain images, thus increasing the volume of the effectively visible brain as well as the anatomical definition of brain structures, also reducing beam-hardening artefacts in the bone–brain interface, with a potential increase of sensitivity in the detection of parenchymal lesions.8,10,11
Different studies have recently compared FBP with HIR and MBIR in phantoms and patients, in the oncology setting, in chest CT, abdomen CT and brain CT for acute stroke, demonstrating that MBIR can increase diagnostic capability while reducing radiation dose exposure.12–17
Moreover, as a previous study has demonstrated, the use of the MBIR algorithm on a brain CT scan at 2 mm thickness improved the diagnostic quality of images, in terms of subjective evaluation and also significantly increased the sensitivity in detecting a subtle sign such as the hyperdense artery sign in stroke patients. 18
Our purpose was to evaluate the possible added value of the MBIR algorithm in the assessment of acute traumatic brain lesions in emergency brain NECT, in comparison with the HIR approach.
Materials and methods
Patients
All patients referred to the emergency department for trauma, who underwent NECT were retrospectively evaluated from January 2019 to February 2020.
Inclusion criteria were: (a) age greater than 18 years; (b) different kinds of brain trauma, with or without loss of consciousness; (c) at least one further radiological follow-up examination during clinical hospitalisation, including CT scans and/or MRI. Patients with neither radiological signs of brain trauma on initial NECT nor follow-up imaging were excluded.
CT scanning parameters
All CT examinations were performed using a 256-slice single-source CT scanner (iCT Elite; Philips Healthcare, Eindhoven, The Netherlands) with the following acquisition protocol: 120 kV, 269 mAs, pitch 0.280, rotation time of 0.5 ms, collimation of 32 × 0.625 mm, dose right index 35, scan time 15 s (including scout). A single volumetric acquisition was performed, then the raw data were reconstructed using both the HIR technique (iDose level 4; Philips Healthcare, slice thickness 4 mm, increment 4 mm) and with the MBIR approach (IMR level 1; Philips Healthcare, slice thickness 2 mm, increment 1 mm).
All images were reconstructed with brain standard kernel and brain routine kernel for the HIR and the MBIR approach, respectively.
Qualitative image analysis and lesion detection
To assess the image quality of the two different reconstruction protocols, the images were evaluated using a standard brain window setting (window level 40 HU and a width of 100 HU) on a picture archiving and communication system (PACS) viewer. Two radiologists, blinded to clinical data and reconstruction algorithm, independently evaluated the presence, the nature, the number and the location of acute findings as follows: haemorrhages (extradural, subdural, subarachnoid, parenchymal, intraventricular), contusive lesions, and haemorrhagic diffuse axonal injuries (DAIs). The readers evaluated all images, previously anonymised and randomly presented, and were able to modify window settings and, if necessary, use multiplanar reconstruction (MPR), as appropriate.
A qualitative image quality evaluation was also performed using a 4-point Likert scale as follows: 1 = unacceptable; 2 = acceptable; 3 = good; 4 = more than average.
Contusive lesions, extradural, subdural, subarachnoid, parenchymal and intraventricular haemorrhages were confirmed during follow-up CT or MRI examinations while DAIs were confirmed with MRI.
Quantitative image analysis
For each scan, and both protocols (HIR and MBIR) the mean attenuation value of the thalamus (CTthalamus) and of the internal capsule (CTinternal capsule) were evaluated by drawing a circular region of interest (ROI) of 10 mm and 4 mm, respectively.
Besides, to evaluate the image noise, the standard deviation (SD) of the thalamus attenuation value (SDthalamus) was recorded.
The contrast-to-noise ratio (CNR) was calculated as follows: CNR = (CTthalamus–CTinternal capsule)/(SDthalamus). The signal-to-noise ratio (SNR) was calculated using the above-mentioned ROIs as follows: SNR = (HUthalamus/SDthalamus). SNR and CNR were both calculated following the methods suggested by Nakaura et al. 10
Radiation dose exposure
The dose-length product (DLP) and CT dose index volume (CTDIvol), automatically displayed on the dose report of the CT scanner, were reported. The effective dose (ED) was derived from the product of the DLP and a conversion coefficient for the brain (k=0.0021 mSv/mGy × cm). 19
Statistical analysis
Categorical variables are expressed as numbers and percentages and compared by using the chi-squared test. Continuous variables are expressed as mean ± SD and compared by using the Student’s t-test. Subjective image quality, expressed as a number and interquartile range (IQR), was compared by using the Mann–Whitney U-test. Cohen’s kappa (k) coefficient was calculated to examine the inter-observer agreement for image quality assessment. Five levels of kappa values are defined as follows: very poor reliability (kappa value <0.20); poor reliability (0.21–0.40); fair reliability (0.41–0.60); moderate reliability (0.61–0.80); and good reliability (0.81–1). 20
All tests were two-sided, and the P value of 0.05 or less was considered statistically significant. All statistical analyses were performed using IBM SPSS 26.0 (SPSS Inc., Chicago, IL, USA).
Results
From a total of 420 patients, 70 were excluded due to the absence of signs of brain trauma at the first NECT (n=50) or due to the lack of follow-up imaging, both MRI and CT (n=20). Finally, a total of 350 patients (188 men/162 women) with a mean age of 47.4 years (±16.5, range 33–88) were enrolled.
Qualitative image analysis and lesion detection
The median score assessed as image quality for CT images with the MBIR algorithm was 4 (IQR 3–4), while the median score assessed for CT images with the HIR algorithm was 3 (IQR 2–3), with significant differences between the two reconstruction algorithms (P=0.003). According to image quality, the inter-observer agreement was moderate (k=0.79) (Figure 1).
Figure 1.
CT scan of a 75-year-old man who underwent brain trauma 2 days before admittance to the emergency room. The patient was asymptomatic and under anticoagulant therapy. Axial CT images with brain window, acquired with a 256 MDCT and reconstructed with HIR (axial native images in (a) and (e), and zoom in (c) and (g)) and MBIR (axial native images in (b) and (f), and zoom in (d) and (h)), with 4 mm and 2 mm slice thickness, respectively. MBIR offered high diagnostic quality images, less noise, and high anatomical details. CT: computed tomography; HIR: hybrid iterative reconstruction; MBIR: model-based iterative reconstruction; MDCT: multiple detector computed tomography.
Per patient lesion evaluation showed an overall increased value of detecting acute trauma-related brain lesions; in particular, 78 patients (22.28) were judged negative with the HIR protocol while only 11 (3.14) were judged negative with the MBIR protocol.
Images reconstructed with MBIR offered a higher detection rate of acute lesions in comparison to HIR for all types of lesions (P<0.001). MBIR allowed the identification of 116 extradural haematomas while HIR allowed the identification of 68 (P<0.001), 162 versus 98 subdural haemorrhages (P<0.001) and 118 versus 78 subarachnoid haemorrhages (P<0.001) (Figure 2). Respectively, 94 parenchymal haemorrhages were detected with MBIR, while 64 were detected with HIR (P<0.001). Contusive lesions identified with MBIR were 36 while 28 were identified with HIR (P<0.001). DAIs detected with MBIR were 75 while 31 were detected with HIR (P<0.001). HIR and MBIR allowed the identification of all the intraventricular haemorrhages (P=1.0).
Figure 2.
CT scan of a 42-year-old woman involved in a car accident. Axial CT images set with width and level to evaluate brain parenchyma. (a) Images reconstructed with HIR shows a focal but not well-defined hyperdensity in the right frontal lobe. (b) Images reconstructed with the MBIR algorithm, confirmed the presence of subarachnoid bleeding in the right frontal lobe. Another focus of bleeding is appreciable near the falx, in the right frontal anterior lobe. HIR reconstruction (c), shows a punctiform hyperdensity, but MBIR (d), thanks to the thinner image reconstruction, also gives information about the real extent of the subarachnoid bleeding. CT: computed tomography; HIR: hybrid iterative reconstruction; MBIR: model-based iterative reconstruction.
Agreement between readers
Inter-observer agreement was almost excellent in evaluating each reconstruction protocol for subdural haemorrhages (k=0.82), subarachnoid haemorrhages (k=0.91), parenchymal haemorrhages (k=0.98), and contusive lesions (k=0.88), while it was moderate for extradural haematoma (k=0.79) and DAIs (k=0.79) (Figure 3). A comparison of image quality and lesion detection is summarised in Table 1.
Figure 3.
CT scan of a 26-year-old woman who underwent brain trauma. Axial CT images, set with width and level to evaluate brain parenchyma show a punctiform hyperdensity in the frontal lobe, rounded with focal hypodensity. HIR reconstruction does not clearly define the real extent of the suffering parenchyma due to the contusion (a), indeed, oedema surrounding the bleeding is not well defined, while in the image reconstructed with MBIR (c) all pathological findings are better appreciable. Moreover, HIR images, due to the beam-hardening artefacts near the bone–brain interface, do not allow a detailed evaluation of brain parenchyma, as shown in (b). CT: computed tomography; HIR: hybrid iterative reconstruction; MBIR: model-based iterative reconstruction.
Table 1.
Lesion detection using the two algorithms.
| N=350 | HIR algorithm | MBIR algorithm | P value | k value |
|---|---|---|---|---|
| Extradural haematomas (n, %) | 68 (19.42) | 116 (33.14) | <0.001 | 0.79 |
| Subdural haemorrhages (n, %) | 98 (28.0) | 162 (46.28) | <0.001 | 0.82 |
| Subarachnoid haemorrhages (n, %) | 78 (22.28) | 118 (33.71) | <0.05 | 0.91 |
| Parenchymal haemorrhages (n, %) | 64 (18.28) | 94 (26.85) | <0.001 | 0.98 |
| Contusive lesions (n, %) | 28 (0.08) | 36 (10.28) | <0.05 | 0.88 |
| DAIs (n, %) | 31 (8.85) | 75 (21.42) | <0.001 | 0.70 |
| Intraventricular haemorrhages (n, %) | 15 (0.04) | 15 (0.04) | 1.0 | 1.0 |
DAIs: diffuse axonal injuries; HIR: hybrid iterative reconstruction; MBIR: model-based iterative reconstruction.
Quantitative image analysis
The mean HU of the thalamus and the internal capsule on CT imaging reconstructed with MBIR were slightly higher in comparison to the HIR approach (34.71 ± 1.60 vs. 33.82 ± 2.14 and 26.12 ± 2.01 vs. 25.24 ± 2.61), without reaching a significant difference (P=0.650 and P=0.140, respectively).
The mean SD of the thalamus on CT imaging reconstructed with MBIR was lower in comparison to the HIR approach (2.12 ± 0.92 vs. 3.52 ± 1.10), with a statistically significant difference (P=0.030).
The CNR was significantly higher in MBIR (3.06 ± 0.55) than HIR (1.55 ± 0.68), reaching a significant difference (P<0.0001). The same results were found regarding the SNR, which was significantly higher in MBIR (14.51 ± 1.78) than HIR (8.62 ± 1.88), with a significant difference (P<0.0001). Quantitative image results are summarized in Table 2.
Table 2.
Quantitative image evaluation using the two algorithms.
| N=350 | HIR algorithm | MBIR algorithm | P value |
|---|---|---|---|
| HU thalamus (HU ± SD) | 33.82 ± 2.14 | 34.71 ± 1.60 | 0.650 |
| HU internal capsule (HU ± SD) | 25.24 ± 2.61 | 26.12 ± 2.01 | 0.140 |
| SD thalamus (HU ± SD) | 3.52 ± 1.10 | 2.12 ± 0.92 | 0.030 |
| CNR (± SD) | 1.55 ± 0.68 | 3.06 ± 0.55 | <0.0001 |
| SNR (± SD) | 8.62 ± 1.88 | 14.51 ± 1.78 | <0.0001 |
P values in bold are considered statistically significant.
CNR: contrast-to-noise ratio; HIR: hybrid iterative reconstruction; HU: Hounsfield units; MBIR: model-based iterative reconstruction; SD: standard deviation; SNR: signal-to-noise ratio.
Radiation dose exposure
The DLP mean value was 763.8 ± 74.02 mGy × cm, CTDIvol was 40.04 ± 3.26 mGy and ED was 1.60 ± 0.16 mSv.
Discussion
The MBIR approach using a low-dose protocol compared to other algorithms helped to identify a greater number of discernible objects, as demonstrated in a phantom study by Löve et al. 8 If we relate ‘discernible objects’ to the acute traumatic brain lesions, our study confirms that MBIR, compared to the HIR algorithm, improves image quality, reducing noise and improving the detection rate. Another fundamental consequence of the application of MBIR, thanks to thinner slices, is the possibility to obtain multiplanar reconstructed images, thus evaluating with more detail the brain and bone structures, decreasing the artefacts in the bone–brain interface, without loss of image quality, especially in the posterior fossa. The effective reduction of artefacts with MBIR is helpful for radiologists to avoid missing small lesions close to the skull or mistaking artefacts for pathological findings. 21 , 22
Our results showed that the overall image quality for CT images reconstructed with the MBIR algorithm was significantly higher in comparison with the HIR algorithm, confirmed by a moderate inter-observer agreement (k=0.79). In particular, we found that thanks to noise reduction and slice thickness, MBIR offered a higher detection rate of all acute traumatic brain lesions in comparison to HIR, allowing us to identify 50% more extradural haematomas, 40% more subdural haemorrhages, 34% more subarachnoid haemorrhages, 32% more parenchymal haemorrhages and 22% more contusive lesions. Moreover, the thinner slice thickness allows the depiction of 40% more DAIs. For all lesions, the agreement was good to almost perfect.
We also found that MBIR objective image quality, expressed both with SNR and CNR, was slightly higher in comparison with the HIR approach, reaching a significant difference (P<0.0001).
The high capability of MBIR in the detection of brain injuries for all the above-mentioned properties leads to a more accurate clinical interpretation of lesions. Our study demonstrates that all TBIs are more easily detectable in images reconstructed with a model-based algorithm, compared to the iterative algorithm. The MBIR algorithm is capable of identifying very small acute brain lesions, such as DAIs, mainly due to the use of thin-slice images and the reduction of image noise. We found no statistically significant differences in detecting intraventricular haemorrhages because of high intrinsic contrast resolution between blood and cerebrospinal fluid.
Our results are in line with those of Nakaura et al., 10 who compared MBIR, HIR and FBP techniques in brain CT, determining that MBIR provides a higher image noise reduction especially using thin-slice imaging. The authors also demonstrated that thin-slice images can reduce image noise in multiplanar reconstruction, useful in evaluating brain structures, in line with our study. Our study is also consistent with that of Lombardi et al., 18 which demonstrated that the combination of a thin slice (2 mm) with a noise reduction approach from MBIR increases the sensitivity in detecting the hyperdense artery sign in ischaemic stroke and also offers better overall image quality of the whole brain.
When MBIR was first introduced, the reconstruction protocol was very long-lasting as compared to the hybrid algorithm, and therefore this approach was not reliable in emergency settings. Notohamiprodjo et al. 16 demonstrated that MBIR, in comparison with HIR, was superior in terms of the reduction of image noise and consequently artefacts but reported a very long-lasting reconstruction time for the MBIR protocol, about 32 minutes. Conversely in our series, the duration of the reconstruction algorithm of MBIR is about 10 images per second with a total time of about 60 seconds for brain CT, also allowing the application of MBIR in the emergency settings as a feasible and reliable algorithm. For the aforementioned reasons, images reconstructed with MBIR can generally be used in the emergency setting, when a patient’s condition requires immediate and precise diagnosis to establish the best treatment.
Some limitations of our study should be taken into account. First, it was a retrospective non-randomised study population. Moreover, MBIR images were compared with 4 mm HIR images, which represented, at that time, the standard of care at our institution.
In conclusion, MBIR, offering higher image quality with a thinner slice, allowed us to identify a higher number of acute traumatic lesions than HIR, with a significant reduction in noise and beam-hardening artefacts, along with a reduction of radiation dose exposure, representing a very useful tool in assessing the brain findings of trauma patients.
Footnotes
Availability of data and materials: All data generated or analysed during this study are included in this published article.
Conflict of interest: The author(s) declared no potential conflicts of interest with respect to the research, authorship, and/or publication of this article.
Consent: All patients signed the informed consent form to be eligible for this study, and gave consent for information about themselves to be published in scientific journals.
Ethics approval: Local ethical committee review of the protocol deemed that formal approval was not required because of the retrospective, observational and anonymous nature of this study.
Funding: The author(s) received no financial support for the research, authorship, and/or publication of this article.
ORCID iD: Davide Ippolito https://orcid.org/0000-0002-2696-7047
References
- 1.Dublin AB, French BN, Rennick JM. Computed tomography in head trauma. Radiology 1977; 122: 365–369. 10.1148/122.2.365 [DOI] [PubMed] [Google Scholar]
- 2.Kaups KL, Davis JW, Parks SN. Routinely repeated computed tomography after blunt head trauma: does it benefit patients? J Trauma: Injury, Infect Crit Care 2004; 56: 475–481. 10.1097/01.TA.0000114304.56006.D4 [DOI] [PubMed] [Google Scholar]
- 3.Nelson DW, Nyström H, MacCallum RM, et al. Extended analysis of early computed tomography scans of traumatic brain injured patients and relations to outcome. J Neurotrauma 2010; 27: 51–64. 10.1089/neu.2009.0986 [DOI] [PubMed] [Google Scholar]
- 4.Diaconis JN, Rao KC. CT in head trauma: a review. J Comput Tomogr 1980; 4: 261–270. 10.1016/0149-936x(80)90018-1 [DOI] [PubMed] [Google Scholar]
- 5.Gunn MLD, Kohr JR. State of the art: technologies for computed tomography dose reduction. Emerg Radiol 2010; 17: 209–218. 10.1007/s10140-009-0850-6 [DOI] [PubMed] [Google Scholar]
- 6.Hidajat N, Mäurer J, Schröder R-J, et al. Radiation exposure in spiral computed tomography: dose distribution and dose reduction. Invest Radiol 1999; 34: 51–57. 10.1097/00004424-199901000-00008 [DOI] [PubMed] [Google Scholar]
- 7.Yu L, Liu X, Leng S, et al. Radiation dose reduction in computed tomography: techniques and future perspective. Imaging Med 2009; 1: 65–84. 10.2217/iim.09.5 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 8.Löve A, Olsson M-L, Siemund R, et al. Six iterative reconstruction algorithms in brain CT: a phantom study on image quality at different radiation dose levels. Br J Radiol 2013; 86: 20130388. 10.1259/bjr.20130388 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 9.Ren Q, Dewan SK, Li M, et al. Comparison of adaptive statistical iterative and filtered back projection reconstruction techniques in brain CT. Eur J Radiol 2012; 81: 2597–2601. 10.1016/j.ejrad.2011.12.041 [DOI] [PubMed] [Google Scholar]
- 10.Nakaura T, Iyama Y, Kidoh M, et al. Comparison of iterative model, hybrid iterative, and filtered back projection reconstruction techniques in low-dose brain CT: impact of thin-slice imaging. Neuroradiology 2016; 58: 245–251. 10.1007/s00234-015-1631-4 [DOI] [PubMed] [Google Scholar]
- 11.Iyama Y, Nakaura T, Oda S, et al. Iterative reconstruction designed for brain CT: a correlative study with filtered back projection for the diagnosis of acute ischemic stroke. J Comput Assist Tomogr 2017; 41: 884–890. 10.1097/RCT.0000000000000626 [DOI] [PubMed] [Google Scholar]
- 12.Pickhardt PJ, Lubner MG, Kim DH, et al. Abdominal CT with model-based iterative reconstruction (MBIR): initial results of a prospective trial comparing ultra low-dose with standard-dose imaging. Am J Roentgenol 2012; 199: 1266–1274. 10.2214/AJR.12.9382 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 13.Chen B, Barnhart H, Richard S, et al. Volumetric quantification of lung nodules in CT with iterative reconstruction (ASiR and MBIR): nodule volume quantification in CT with iterative reconstructions. Med Phys 2013; 40: 111902. 10.1118/1.4823463 [DOI] [PubMed] [Google Scholar]
- 14.Katsura M, Matsuda I, Akahane M, et al. Model-based iterative reconstruction technique for radiation dose reduction in chest CT: comparison with the adaptive statistical iterative reconstruction technique. Eur Radiol 2012; 22: 1613–1623. 10.1007/s00330-012-2452-z [DOI] [PubMed] [Google Scholar]
- 15.Chang W, Lee JM, Lee K, et al. Assessment of a model-based, iterative reconstruction algorithm (MBIR) regarding image quality and dose reduction in liver computed tomography. Invest Radiol 2013; 48: 598–606. 10.1097/RLI.0b013e3182899104 [DOI] [PubMed] [Google Scholar]
- 16.Notohamiprodjo S, Deak Z, Meurer F, et al. Image quality of iterative reconstruction in cranial CT imaging: comparison of model-based iterative reconstruction (MBIR) and adaptive statistical iterative reconstruction (ASiR). Eur Radiol 2015; 25: 140–146. 10.1007/s00330-014-3374-8 [DOI] [PubMed] [Google Scholar]
- 17.Hérin E, Gardavaud F, Chiaradia M, et al. Use of model-based iterative reconstruction (MBIR) in reduced-dose CT for routine follow-up of patients with malignant lymphoma: dose savings, image quality and phantom study. Eur Radiol 2015; 25: 2362–2370. 10.1007/s00330-015-3656-9 [DOI] [PubMed] [Google Scholar]
- 18.Lombardi S, Riva L, Patassini M, et al. “Hyperdense artery sign” in early ischemic stroke: diagnostic value of model-based reconstruction approach in comparison with standard hybrid iterative reconstruction algorithm. Neuroradiology 2018; 60: 1273–1280. 10.1007/s00234-018-2092-3 [DOI] [PubMed] [Google Scholar]
- 19.Schlattl H, Zankl M, Becker J, et al. Dose conversion coefficients for CT examinations of adults with automatic tube current modulation. Phys Med Biol 2010; 55: 6243–6261. 10.1088/0031-9155/55/20/013 [DOI] [PubMed] [Google Scholar]
- 20.McHugh ML. Interrater reliability: the kappa statistic. Biochem Med (Zagreb) 2012; 22: 276–282. [PMC free article] [PubMed] [Google Scholar]
- 21.Machida H, Takeuchi H, Tanaka I, et al. Improved delineation of arteries in the posterior fossa of the brain by model-based iterative reconstruction in volume-rendered 3D CT angiography. AJNR Am J Neuroradiol 2013; 34: 971–975. 10.3174/ajnr.A3320 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 22.Nishizawa M, Tanaka H, Watanabe Y, et al. Model-based iterative reconstruction for detection of subtle hypoattenuation in early cerebral infarction: a phantom study. Jpn J Radiol 2015; 33: 26–32. 10.1007/s11604-014-0376-z [DOI] [PubMed] [Google Scholar]