Submillisievert Chest CT With Filtered Back Projection and Iterative Reconstruction Techniques

Transcription

1 Cardiopulmonary Imaging Original Research Padole et al. Submillisievert Chest CT Cardiopulmonary Imaging Original Research Atul Padole 1 Sarabjeet Singh Jeanne B. Ackman Carol Wu Synho Do Sarvenaz Pourjabbar Ranish Deedar Ali Khawaja Alexi Otrakji Subba Digumarthy Jo-Anne Shepard Mannudeep Kalra Padole A, Singh S, Ackman JB, et al. Keywords: chest CT, CT radiation dose reduction, filtered back projection, iterative reconstruction techniques DOI: /AJR Received November 18, 2013; accepted after revision February 15, M. Kalra and S. Singh have received Educational Scholar grants from the Radiological Society of North America Research and Education Foundation. S. Singh has received research grants from GE Healthcare, Siemens Healthcare, Philips Healthcare, and the Society of Thoracic Radiology. None of the other coauthors have any pertinent financial disclosures and had complete unrestricted access to the data at all times. 1 All authors: Department of Radiology, Massachusetts General Hospital, Harvard Medical School, 55 Fruit St, Founders-202, Boston, MA Address correspondence to A. Padole (apadole@partners.org). AJR 2014; 203: X/14/ American Roentgen Ray Society Submillisievert Chest CT With Filtered Back Projection and Iterative Reconstruction Techniques OBJECTIVE. The purpose of this study was to compare submillisievert chest CT images reconstructed with filtered back projection (FBP), SafeCT, adaptive statistical iterative reconstruction (ASIR), and model-based iterative reconstruction (MBIR) with standard of care FBP images. SUBJECTS AND METHODS. Fifty patients (33 men and 17 women; mean age [± SD], 62 ± 10 years) undergoing routine chest CT gave written informed consent for acquisition of an additional submillisievert chest CT series with reduced tube current but identical scanning length as standard of care chest CT. Sinogram data of the submillisievert series were reconstructed with FBP, SafeCT, ASIR, and MBIR and compared with FBP images at standarddose chest CT (n = 8 50 = 400 series). Two thoracic radiologists performed independent comparison for visualization of lesion margin, visibility of small structures, and diagnostic acceptability. Objective noise measurements and noise spectral density were obtained. RESULTS. Of 287 detected lesions, 162 were less than 1-cm noncalcified nodules. Lesion margins were well seen on all submillisievert reconstruction images except MBIR, on which they were poorly visualized. Likewise, only submillisievert MBIR images were suboptimal for visibility of normal structures, such as pulmonary vessels in the outer 2 cm of the lung, interlobular fissures, and subsegmental bronchial walls. MBIR had the lowest image noise compared with other techniques. CONCLUSION. FBP, SafeCT, ASIR, and MBIR can enable optimal lesion evaluation on chest CT acquired at a volume CT dose index of 2 mgy. However, all submillisievert reconstruction techniques were suboptimal for visualization of mediastinal structures. Submillisievert MBIR images were suboptimal for visibility of normal lung structures despite showing lower image noise. T he increased use of CT has raised concerns over the risk of patients developing radiation-induced cancer later in life. The estimated annual effective dose of medically related radiation dose per individual in the United States population has increased from 3.6 msv in 1980 to 6.2 msv in 2006, with the largest contribution of increase due to CT [1]. Therefore, dose reduction in CT without compromising diagnostic accuracy can play an important role in mitigating some of the concerns over radiation-induced carcinogenesis. Several strategies have been used to reduce the CT radiation dose to patients, such as lower tube current or voltage; iterative reconstruction techniques; noise reduction filters; and automatic tube current modulation according to patient size, body region, and protocol type [2]. However, low-dose CT has higher image noise and can affect the diagnostic information, especially with conventional filtered back projection (FBP). Initial studies with iterative reconstruction techniques have shown promising results for reducing radiation dose of CT [3 14]. Iterative reconstruction techniques accurately model the CT data to generate a set of synthesized projections to reduce image noise associated with low-dose CT. Adaptive statistical iterative reconstruction (ASIR) technique (GE Healthcare) was the first iterative reconstruction technique released for clinical use [1]. A more complex iterative reconstruction technique, the model-based iterative reconstruction (MBIR or Veo) technique (GE Healthcare) has recently become available for clinical use [4]. SafeCT (MedicVision) is a third-party vendor neutral imagingbased iterative reconstruction approach that 772 AJR:203, October 2014

2 Submillisievert Chest CT has been approved by the United States Food and Drug Administration [5]. The purpose of our prospective study was to compare submillisievert chest CT images reconstructed with FBP, SafeCT, ASIR, and MBIR with standard of care FBP images. Subject and Methods Patient Selection The human research committee of our institutional review board approved our prospective clinical study. Before undergoing standard of care clinically indicated chest CT, 50 adult patients (33 men and 17 women) gave written informed consent A D G B E H Fig. 1 Transverse chest CT images in 77-year-old man (weight, 64 kg) acquired at standard of care and submillisievert doses and reconstructed with filtered back projection (FBP) and various iterative reconstruction techniques. Emphysematous areas (arrows) were depicted optimally on standard of care FBP and submillisievert FBP, submillisievert adaptive statistical iterative reconstruction (ASIR), and submillisievert SafeCT (MedicVision) images. Lesion margin for emphysematous areas was suboptimal with submillisievert model-based iterative reconstruction (MBIR) images. A H, Standard of care FBP (A), submillisievert FBP (B), ASIR with 50% blending (C), ASIR with 70% blending (D), SafeCT Chest4 setting (E), SafeCT Lung1 setting (F), SafeCT Lung2 setting (G), and MBIR (H). for acquisition of an additional imaging series at a submillisievert dose. The submillisievert series were acquired with reduced tube current but identical scanning length compared with the standard of care chest CT. Hemodynamically stable patients scheduled for standard of care routine chest CT and capable of providing written informed consent for a submillisievert CT series were included in the study. Patients who were unable to hold their breath or lie still for 10 seconds or with history of contrast agent allergy were excluded. Patients undergoing emergency CT and women who were pregnant or trying to become pregnant were also excluded. The weight and height of all participating patients were recorded. Body mass index (BMI) was calculated for all patients. Scanning Techniques All patients were scanned on a 64-MDCT scanner (Discovery CT750 HD, GE Healthcare). Patients were centered in the gantry isocenter for acquisition of chest localizer radiographs at 80 kv and 20 mas. Patients were instructed to avoid any voluntary movements during CT. Within 10 seconds of acquisition of the standard of care chest CT, the submillisievert series was acquired over the same body region and scanning length. For the submillisievert C F AJR:203, October

3 Padole et al. series (with estimated effective dose just under 1 msv), the dose-length product (DLP) was targeted at just less than 70 mgy cm over an identical scanning length, with a conversion coefficient of [15]. Dose reduction for the submillisievert series was achieved by using a lower fixed tube current, to obtain the target DLP. All scanning parameters with the exception of tube current were held constant, including tube potential of 120 kv, pitch of 0.984:1, 0.5-second gantry rotation time, helical acquisition mode, mm table feed per gantry rotation, mm detector configuration, 2.5-mm reconstructed section thickness, and 2.5- mm reconstruction section interval. No additional A B C D E F G H Fig. 2 Transverse chest CT images in 65-year-old man (weight 82 kg) acquired at standard of care and submillisievert doses and reconstructed with filtered back projection (FBP) and various iterative reconstruction techniques. Interlobular fissure (arrows) was depicted optimally on standard of care FBP and submillisievert FBP, submillisievert adaptive statistical iterative reconstruction (ASIR), and submillisievert SafeCT (MedicVision) images and suboptimally on submillisievert model-based iterative reconstruction (MBIR) images. A H, Standard of care FBP (A), submillisievert FBP (B), ASIR with 50% blending (C), ASIR with 70% blending (D), SafeCT Chest4 setting (E), SafeCT Lung1 setting (F), SafeCT Lung2 setting (G), and MBIR (H). IV contrast material was administered for the acquisition of the submillisievert series. Image Reconstruction The conventional FBP image reconstruction technique has greater image noise and artifacts because it does not accurately take into account the x-ray photon statistics and system hardware details, such as focal spot size and detectors. ASIR is a hybrid iterative reconstruction technique which involves blending with FBP images and takes into account the photon and electronic noise that primarily affect the image noise; this technique has been described in prior publications. MBIR is a pure iterative reconstruction technique that does not blend with FBP images. In addition to the modeling of photon and noise statistics, as with ASIR, MBIR incorporates more accurate information on x-ray focal spot size, active detector size, and image voxel shape and size. Several other image reconstruction techniques such as IRIS and SAFIRE (Siemens Healthcare), idose (Philips Healthcare), and AIDR-3D (Toshiba Medical Systems), have recently been introduced but are not included in this study [6, 16]. MBIR is a full iterative reconstruction technique and requires a longer reconstruction time of approximately minutes for a single dataset compared with 774 AJR:203, October 2014

4 Submillisievert Chest CT other iterative reconstruction techniques, including ASIR, which are nearly real time. The SafeCT technique processes FBP images for denoising. SafeCT image processing is based on Generic Iterative Retro Reconstruction in 3D (GiRR3D, Medic Vision) technology. GiRR3D enables volumetric retrospective reconstruction of low-dose CT images, which reduces the noise while maintaining fine structural details. It uses knowledge of statistical priors of the CT noise distribution and signal-to-noise ratio enhancement to generate images with lower noise and preserve image details. SafeCT is a vendor-neutral technique that can process FBP images from any CT vendor. Because it is a primarily DICOM image based A B C D E F G H Fig. 3 Transverse chest CT images in 77-year-old man (weight 64 kg) acquired at standard of care and submillisievert doses and reconstructed with FBP and various iterative reconstruction techniques. Subsegmental bronchial wall (arrows) and peripheral small blood vessels (arrowheads) were depicted optimally on standard of care FBP, submillisievert FBP, submillisievert adaptive statistical iterative reconstruction (ASIR), and submillisievert SafeCT (MedicVision images and suboptimally on submillisievert model-based iterative reconstruction (MBIR) images. A H, Standard of care FBP (A), submillisievert FBP (B), ASIR with 50% blending (C), ASIR with 70% blending (D), SafeCT Chest4 setting (E), SafeCT Lung1 setting (F), SafeCT Lung2 setting (G), and MBIR (H). technique, SafeCT processes images in almost real time [5]. FBP technique was used to reconstruct standard of care chest CT series. Sinography data of the submillisievert series were reconstructed with FBP, ASIR (at a blending percentage of 50% and 70% with FBP), and MBIR. The FBP, ASIR, and MBIR images were generated from the scanner at a 2.5-mm section thickness with a 2.5-mm section interval. Thin FBP images (0.6 mm) were used to generate SafeCT images (three settings: Chest4, Lung1, Lung2), on an offline processing workstation. The processed 0.6-mm images were reconstructed back to 2.5-mm section thickness on the image processing workstation. All 400 series (standard of care dose FBP images, submillisievert dose FBP, ASIR (50% and 70%), MBIR, and SafeCT images, n = 8 series per patient 50 patients = 400 total image series) were assessed for diagnostic image quality. Subjective Image Evaluation Subjective evaluation of the images was performed on a DICOM-compliant autocalibrated monitor with a two-megapixel resolution with the help of a standard DICOM image viewer (ClearCanvas workstation, ClearCanvas). All image datasets were independently evaluated in a blinded and randomized manner by two experienced thoracic radiologists with 32 and 16 years of experi- AJR:203, October

5 Padole et al. A D G B E H Fig. 4 Transverse chest CT images of 58-year-old man (weight 113 kg) acquired at standard of care (SOC) and submillisievert doses and reconstructed with FBP and various iterative reconstruction techniques. Pericardium (arrows) was depicted optimally on standard of care FBP and suboptimally on submillisievert filtered back projection (FBP), submillisievert adaptive statistical iterative reconstruction (ASIR), submillisievert SafeCT (MedicVision), and submillisievert model-based iterative reconstruction (MBIR) images. A H, Standard of care FBP (A), submillisievert FBP (B), ASIR with 50% blending (C), ASIR with 70% blending (D), SafeCT Chest4 setting (E), SafeCT Lung1 setting (F), SafeCT Lung2 setting (G), and MBIR (H). C F ence. Both radiologists were trained for image evaluation on CT images of two patients in one session. These two chest CT examinations were not included in the study analysis. The submillisievert FBP images were displayed first for evaluation of lesion detection. Then, all eight imaging series including standard of care chest CT and submillisievert series were displayed simultaneously for evaluation. All eight series were displayed at the same window width and level settings (lung: width, 1500 HU; level, 600 HU; mediastinal: width, 350 HU; level, 50 HU). In addition to the default preselected lung and mediastinal settings, radiologists were also allowed to change the window width and level according to their comfort level of assessment. 776 AJR:203, October 2014

6 Submillisievert Chest CT Lesion detection was performed separately in lung and mediastinal windows, and all clinically important lesions were recorded. The location, size, and attenuation of the smallest or most subtle lesions were recorded. Subjective image quality was assessed on a 3-point scale (1, substantially worse; 2, almost equal; and 3, better) in terms of lesion margin, diagnostic confidence, and visibility of ground-glass opacities (GGO), if present. Optimal visibility of small structures (0, unacceptable; 1, acceptable) including blood vessels in the outer 2 cm of the lung, subsegmental bronchial walls, and interlobular fissures were evaluated on the lung window. The pericardium, small lymph nodes, and internal mammary vessels were evaluated with a similar scale on the mediastinal window. In patients with no definite lesion, only visibility of normal structures was evaluated. Objective Image Evaluation Objective image noise and CT numbers were measured on a DICOM image viewer for all imaging series. Three circular ROIs were placed per imaging series in the tracheal air column and descending thoracic aorta at the level of the carina (without touching the wall to cover at least two thirds of its lumen). In the descending thoracic aorta, calcification was also avoided. Noise spectral density plots were created to assess the pattern and trend of noise across different image reconstruction techniques. Noise spectral density is defined as the noise power per unit of bandwidth or frequency and assesses the pattern and trend of noise in different reconstruction techniques by measuring the variance and spatial frequency of noise. Noise spectral density provides better strength and pattern of noise distribution than the CT number [3]. On the other hand, modulation transfer function (MTF) describes the spatial resolution of CT systems. In addition, MTF helps in describing the resolution of linear- and spatial-invariant CT systems [5, 17]. CT radiation dose descriptors, such as volume CT dose index (CTDI vol ) and DLP were recorded from the dose information pages of all CT examinations. Estimated effective dose was derived by multiplying the DLP by msv/mgy cm [15]. Anteroposterior and lateral diameters at midslice were measured in all patients. Effective diameter was calculated by the square root of the product of the anteroposterior and lateral diameters. Statistical Analysis The data were analyzed using standard statistical software (SPSS, version 21, IBM). Paired Student t tests were performed to compare the CT numbers and quantitative objective noise. The Wilcoxon signed rank test was used to compare the ordinal scores for subjective image quality and lesion assessment. The p value of 0.05 with 95% CI was considered significant. Interobserver variability was assessed using kappa statistics. On the basis of the kappa values, the strength of agreement was categorized as poor, < 0.2; fair, 0.2 to < 0.4; moderate, 0.4 to < 0.6; good, 0.6 to < 0.8; and very good, 0.8 to 1. TABLE 1: Subjective Image Quality Scores for Visibility and Diagnostic Acceptability for Submillisievert Filtered Back Projection (FBP) and Iterative Reconstruction Techniques Reader and Score Filtered Back Projection ASIR SafeCT Setting 50% Blending 70% Blending MBIR Chest4 Lung1 Lung2 Reader 1 Lesion margin 1 1/ /50 50/50 50/50 49/50 50/50 50/50 50/50 Diagnostic acceptability 2 50/50 50/50 50/50 50/50 50/50 50/50 50/50 Small vessels in lungs 0 1/50 5/ /50 50/50 49/50 45/50 50/50 50/50 50/50 Major fissures 0 1/50 1/50 2/50 22/50 1/50 1/50 1/ /50 49/50 48/50 28/50 49/50 49/50 49/50 Subsegmental bronchial wall 0 2/50 9/ /50 50/50 48/50 41/50 50/50 50/50 50/50 Subcentimeter lymph node 0 1/50 1/50 1/50 1/50 1/50 1/50 1/ /50 49/50 49/50 49/50 49/50 49/50 49/50 Pericardium 0 15/50 16/50 16/50 15/50 19/50 17/50 17/ /50 34/50 34/50 35/50 31/50 33/50 33/50 Internal mammary vessels 0 1/50 1/50 1/50 2/50 3/50 2/50 2/ /50 49/50 49/50 48/50 47/50 48/50 48/50 (Table 1 continues on next page) AJR:203, October

7 Padole et al. TABLE 1: Subjective Image Quality Scores for Visibility and Diagnostic Acceptability for Submillisievert Filtered Back Projection (FBP) and Iterative Reconstruction Techniques (continued) Reader and Score Results The mean (± SD) age, weight, and BMI of patients included in our study were 62 ± 10 years, 79 ± 16 kg, and 26.3 ± 4.7 kg/m 2, respectively. The mean measured (± SD) effective diameter was 31 ± 3 cm. Radiation Dose The mean (± SD) CTDI vol, DLP, and effective dose for standard of care chest CT were 8.0 ± 4.4 mgy, 299 ± 161 mgy cm, and 4.2 ± 2.3 msv, respectively. Corresponding values for submillisievert chest CT examination for standard of care and submillisievert chest CT were 1.8 ± 0.2 mgy, 66 ± 3.3 mgy cm, and 0.9 ± 0.1 msv, respectively. Filtered Back Projection ASIR 50% Blending 70% Blending MBIR Chest4 Lung1 Lung2 Subjective Image Quality The image quality scores of the two radiologists are summarized in Table 1. A total of 287 lesions were detected, including noncalcified lung nodules (n = 162) less than 1 cm, GGO (n = 25); emphysema (n = 13); calcified granuloma (n = 10); bronchial wall thickening (n = 9); lung cysts (n = 12); mediastinal lymph nodes (n = 5); pleural effusions (n = 3); atelectasis (n = 10); coronary artery calcification (n = 14); thyroid nodules (n = 4); liver cysts (n = 7); and other abnormalities (n = 13), including mosaic attenuation, bone lesions, hiatus hernia, and parenchymal scarring. Detailed analysis of lesions less than 1 cm (93/287) in size was performed. SafeCT Setting Reader 2 Lesion margin 1 6/50 1/50 1/ /50 43/50 43/50 41/50 42/50 34/50 42/50 3 6/50 7/50 7/50 3/50 7/50 15/50 8/50 Diagnostic acceptability 1 1/50 1/50 1/50 4/50 2/50 1/ /50 47/50 47/50 44/50 46/50 47/50 48/50 3 2/50 2/50 2/50 2/50 2/50 2/50 2/50 Small vessels in lungs 0 1/50 1/ /50 50/50 50/50 49/50 49/50 50/50 50/50 Major fissures 0 4/50 4/50 2/50 12/50 5/50 3/50 4/ /50 46/50 48/50 38/50 45/50 47/50 46/50 Subsegmental bronchial wall 0 6/50 7/50 10/50 14/50 5/50 1/ /50 43/50 40/50 36/50 45/50 49/50 50/50 Subcentimeter lymph node 0 3/50 1/50 5/ /50 49/50 50/50 50/50 50/50 45/50 50/50 Pericardium 0 3/49 3/49 4/49 3/49 4/49 3/49 4/ /49 46/49 45/49 46/49 45/49 46/49 45/49 Internal mammary vessels 0 1/50 1/50 1/50 1/50 1/50 1/50 1/ /50 49/50 49/50 49/50 49/50 49/50 49/50 Note Data are number/total patients. ASIR = adaptive statistical iterative reconstruction, MBIR = model-based iterative reconstruction. SafeCT manufactured by MedicVision. No additional lesions were found on standard of care chest CT images compared with submillisievert images regardless of reconstruction techniques. There was no significant difference in diagnostic confidence between standard of care CT and submillisievert imaging regardless of reconstruction techniques. Lesion size and attenuation on submillisievert chest CT were deemed identical to standard of care chest CT regardless of reconstruction techniques used for submillisievert CT. The visibility of the lesion margin was unacceptable in one patient (1/50, p = 0.3) for reader 1 and six patients (6/50, p = 0.014) for reader 2 on submillisievert MBIR images compared with 778 AJR:203, October 2014

8 Submillisievert Chest CT standard of care chest CT images (Fig. 1). The effective diameter in patients with unacceptable lesion margins on submillisievert MBIR images was greater than that in patients with acceptable lesion margins (p = 0.7). However, there was no significant difference in patient weight or BMI in these two groups (p = 0.4). Submillisievert FBP, ASIR, and SafeCT images were deemed acceptable for visualization of lung structures, including small vessels in the outer 2 cm of the lung, subsegmental bronchial walls, and interlobular fissures. However, both radiologists found the submillisievert MBIR images to be unacceptable for visibility of normal structures on lung window images. The visibility of interlobular fissures for submillisievert MBIR images was significantly lower when compared with standard of care in several patients (12 22/50, p < 0.002) (Fig. 2). The effective diameter in patients with unacceptable visibility of interlobular fissures on submillisievert MBIR images was greater than in patients with acceptable visibility of interlobular fissures (p = 0.01). Similarly, visibility of the subsegmental bronchial walls was suboptimal for submillisievert MBIR images (9 14/50, p < 0.002) compared with standard of care chest CT images. In addition, for submillisievert MBIR images, visibility of lung vessels was unacceptable in some patients (5/50, p = 0.024) compared with standard of care chest CT images (Fig. 3). Visibility of the pericardium (n = 3 19/50 patients) was suboptimal on all submillisievert images regardless of reconstruction technique compared with the standard of care chest CT images (p < 0.001) (Fig. 4). The effective diameter in patients with suboptimal visibility of the pericardium was greater than that of patients with optimal visibility of pericardium. Interobserver agreement was fair to moderate (k = ). However, visibility of small lymph nodes and internal mammary vessels was optimal on all submillisievert images regardless of reconstruction technique compared with the standard of care chest CT images. Objective Image Quality Detailed objective measurements are summarized in Table 2. There was no significant difference in CT numbers on submillisievert images reconstructed with different iterative techniques (p = 0.9). Mean objective noise in the descending thoracic TABLE 2: Objective Image Noise Reconstruction Techniques aorta and tracheal air column was significantly lower for the submillisievert MBIR images compared with submillisievert FBP, ASIR, and SafeCT images (p < ). In the descending thoracic aorta, submillisievert MBIR images had 28% (13/18), 38% (13/21), and 59% (13/32) lower noise compared with submillisievert SafeCT, ASIR, and FBP images, respectively. Submillisievert FBP, ASIR, and SafeCT images showed a similar noise spectral density pattern (Fig. 5), although SafeCT had a more consistent decrease in noise spectral density over frequency. On the noise spectral density graph, submillisievert MBIR had the lowest image noise with a distinct and more uniform noise spectrum compared with other techniques assessed in our study. Descending Thoracic Aorta Value p Trachea Air Standard of care FBP Mean attenuation (HU) 203 ± ± 18 SD of attenuation (HU) 16 ± 3 15 ± 4 Submillisievert FBP Mean attenuation (HU) 131 ± ± 12 SD of attenuation (HU) 32 ± 6 < ± 5 Submillisievert ASIR 50% blending Mean attenuation (HU) 131 ± ± 12 SD of attenuation (HU) 22 ± 4 < ± 8 Submillisievert ASIR 70% blending Mean attenuation (HU) 131 ± ± 14 SD of attenuation (HU) 19 ± 3 < ± 9 Submillisievert SafeCT Chest4 setting Mean attenuation (HU) 131 ± ± 12 SD of attenuation (HU) 14 ± 3 < ± 7 Submillisievert SafeCT Lung1 setting Mean attenuation (HU) 131 ± ± 12 SD of attenuation (HU) 24 ± 4 < ± 5 Submillisievert SafeCT Lung2 setting Mean attenuation (HU) 131 ± ± 14 SD of attenuation (HU) 16 ± 3 < ± 5 Submillisievert MBIR Mean attenuation (HU) 131 ± ± 13 SD of attenuation (HU) 13 ± 2 < ± 3 Note FBP = filtered back projection, ASIR = adaptive statistical iterative reconstruction, MBIR = model-based iterative reconstruction. SafeCT manufactured by MedicVision. Submillisievert MBIR images had lower image noise compared with submillisievert FBP, ASIR (50% and 70% blending), and SafeCT (Chest4, Lung1, and Lung2 settings). Discussion Because of inefficient computation power, clinical use of iterative reconstruction techniques had been limited until recently. One reason for longer reconstruction time with conventional iterative reconstruction techniques was the requirement of several iterations to generate satisfactory images on slow computer processing units. Compared with iterative reconstruction techniques, an analytical reconstruction-based FBP requires considerably less computational power. Although FBP is still widely used today in clinical settings, it does not provide optimal results at low radiation doses because it makes many incorrect assumptions about CT projection data. As a result, FBP images have higher noise and artifacts, particularly at low radiation dose. AJR:203, October

9 Padole et al. Growing demands for reducing radiation dose in patients have led to development of efficient computational hardware and more efficient iterative reconstruction techniques, and several iterative reconstruction techniques have become available on commercial CT systems [7]. Prior studies with iterative reconstruction techniques have shown the possibility of scanning patients at a lower radiation dose and obtaining clinically acceptable images with lower image noise [3 14]. For the ASIR technique, Singh et al. [1] reported 38% noise reduction in chest CT compared with FBP with CTDI vol down to 3.5 mgy while maintaining diagnostic confidence. For the MBIR technique, Vardhanabhuti et al. [4] reported that MBIR reduced image noise and improved image quality for chest CT at radiation dose of 0.9 msv (approximately 2mGy). In addition, ultralow-radiation-dose chest CT images ( msv) reconstructed with MBIR have been found to be acceptable for pulmonary nodule detection [6, 7]. To the best of our knowledge, there are no patient studies comparing multiple iterative reconstruction techniques from different CT vendors. We found that FBP, ASIR, MBIR, and SafeCT allow optimal lung lesion evaluation for chest CT acquired at a CTDI vol of 2 mgy or at a submillisievert radiation dose level (0.9 msv). However, none of these reconstruction techniques were optimal for the evaluation of subtle mediastinal structures, such as the pericardium. Despite having the lowest image noise compared with other reconstruction techniques (SafeCT, ASIR, and FBP), submillisievert MBIR images were suboptimal for evaluation of visibility of normal anatomic lung structures, such as interlobular fissures, subsegmental bronchial walls, and small peripheral blood vessels. This was particularly notable for patients with greater cross-sectional diameter (effective diameter > 32 cm). This loss of conspicuity of normal pulmonary structures on submillisievert MBIR images may have been due to excessive denoising of images or lack of edge enhancement, which may be responsible for removal or blurring of these details. Incidentally, prior studies have not reported any difference between the MTF of ASIR, MBIR, and FBP [18]. We also noted an overall blotchy and pixilated appearance on the submillisievert MBIR images. This appearance was not seen with the submillisievert FBP, ASIR, and SafeCT images in our study. Previous studies have reported this appearance with ASIR and MBIR [6, 7]. Because of advancement in the Fig. 5 Noise spectral density graph shows difference in noise patterns for standard of care filtered back projection (FBP) (black solid line); submillisievert FBP (solid blue line); adaptive statistical iterative reconstruction with 30% (solid red line), 50% (solid dark green line), 70% (solid purple line), and 90% (solid light green line) blending; SafeCT (MedicVision) with Chest4 (dashed black line), Lung1 (dashed blue line), and Lung 2 (dashed green line); and model-based iterative reconstruction (Veo, GE Healthcare, dashed red line). Power/Frequency (db/hz) ASIR algorithm, such an appearance is now seen less frequently with ASIR techniques. The pixelated appearance with aggressive noise reduction may explain a distinct noise reduction pattern on noise spectral density estimation of MBIR images compared with other techniques. We also noted that blurring of subtle lung structures may have been due to increased or pronounced blotchy and pixelated appearance of MBIR images. Although our study supports the reported uncompromised lung nodule detection in the lungs at low radiation doses (down to 0.3 mgy CTDI vol ) [6], to our knowledge, loss of optimal visibility of subtle anatomic pulmonary structures has not been assessed. Our findings at least raise the possibility that subtle but important abnormalities, such as mild fissural thickening, nodularity, and interlobular septal thickening, can be missed on MBIR images at submillisievert dose levels, particularly in larger patients (effective diameter > 32 cm). Unfortunately, we did not find any weight cutoff for patients at which submillisievert MBIR may not allow optimal delineation of these structures. Another implication of our study is that at 2 mgy FBP had similar performance compared with ASIR and SafeCT for visibility of normal and abnormal findings. Although, submillisievert ASIR and SafeCT images had significantly lower objective image noise compared with submillisievert FBP images, this finding did not translate to increased or better visualization of normal or abnormal findings on chest CT with ASIR techniques SD-FBP LD-FBP ASIR30 ASIR50 ASIR ASIR90 SafeCT-Chest4 SafeCT-Lung1 SafeCT-Lung2 Veo Frequency (Hz) Our study has some limitations. First, the sample size of this study was limited to 50 patients. We did not perform power analysis to determine the required sample size for our study. Second, submillisievert chest CT was performed with some delay after the standard of care CT and therefore has lower image contrast. This difference in contrast, can affect the appearance of lung and mediastinal structures and image quality evaluation. Third, we did not evaluate the effect of chest CT examinations at radiation doses lower than 2 mgy, which may be possible with these iterative reconstruction techniques, especially for small patients and assessment of lung nodules. We also acquired submillisievert chest CT at a fixed tube current. The effect of automatic tube current modulation was not assessed in this study because the intent of our study was to assess the feasibility of a constant submillisievert dose on chest CT in patients regardless of size. Furthermore, given up to 10% variation between before scanning and after scanning CTDI vol with automatic tube current modulation, we would not have been able to accurately predict the target CTDI vol needed for submillisievert chest CT. Also, the effect of lower kilovoltage for dose reduction was not assessed in our study. Because of the overall blotchy and pixilated appearance, the submillisievert MBIR images have a distinct image appearance that can make it difficult to perform a truly blinded and randomized image evaluation. The effect of MBIR was not assessed on the standard of care dose CT 780 AJR:203, October 2014

10 Submillisievert Chest CT because the intent of our study was primarily to assess the effect of MBIR (which is an expensive technique) on CT acquired at a considerably reduced radiation dose. In conclusion, submillisievert images at a CTDI vol of 2 mgy were diagnostically acceptable even with FBP for evaluation of lung findings. However, submillisievert FBP and all iterative reconstruction techniques were suboptimal for visibility of mediastinal structures. Submillisievert MBIR images are suboptimal for visibility of normal lung structures despite showing the lowest image noise compared with other reconstruction techniques. References 1. Singh S, Kalra MK, Gilman MD, et al. Adaptive statistical iterative reconstruction technique for radiation dose reduction in chest CT: a pilot study. Radiology 2011; 259: Kalra MK, Maher MM, Toth TL, et al. Strategies for CT radiation dose optimization. Radiology 2004; 230: Singh S, Kalra MK, Do S, et al. Comparison of hybrid and pure iterative reconstruction techniques with conventional filtered back projection: dose reduction potential in the abdomen. J Comput Assist Tomogr 2012; 36: Vardhanabhuti V, Loader RJ, Mitchell GR, Riordan RD, Roobottom CA. Image quality assessment of standard- and low-dose chest CT using filtered back projection, adaptive statistical iterative reconstruction, and novel model-based iterative reconstruction algorithms. AJR 2013; 200: Pourjabbar S, Singh S, Singh AK, et al. Preliminary results: prospective clinical study to assess imagebased iterative reconstruction for abdominal computed tomography acquired at 2 radiation dose levels. J Comput Assist Tomogr 2014; 38: Yamada Y, Jinzaki M, Tanami Y, et al. Modelbased iterative reconstruction technique for ultralow-dose computed tomography of the lung: a pilot study. Invest Radiol 2012; 47: Katsura M, Matsuda I, Akahane M, et al. Modelbased iterative reconstruction technique for radiation dose reduction in chest CT: comparison with the adaptive statistical iterative reconstruction technique. Eur Radiol 2012; 22: Kalra KK, Woisetschlager M, Dahlstrom N, et al. Sinogram affirmed iterative reconstruction of low dose chest CT: effect on image quality and radiation dose. AJR 2013; 201:[web]W236 W Khawaja RD, Singh S, Gilman M, et al. Computed tomography (CT) of the chest at less than 1 msv: an ongoing prospective clinical trial of chest CT at submillisievert radiation doses with iterative model image reconstruction and idose4 technique. J Comput Assist Tomogr [Epub 2014 Mar 19] 10. Prakash P, Kalra MK, Ackman JB, et al. Diffuse lung disease: CT of the chest with adaptive statistical iterative reconstruction technique. Radiology 2010; 256: Kalra MK, Maher MM, Blake MA, et al. Detection and characterization of lesions on low-radiation-dose abdominal CT images postprocessed with noise reduction filters. Radiology 2004; 232: Leipsic J, Nguyen G, Brown J, Sin D, Mayo JR. A prospective evaluation of dose reduction and image quality in chest CT using adaptive statistical iterative reconstruction. AJR 2010; 195: Hara AK, Paden RG, Silva AC, Kujak JL, Lawder HJ, Pavlicek W. Iterative reconstruction technique for reducing body radiation dose at CT: feasibility study. AJR 2009; 193: Hu XH, Ding XF, Wu RZ, Zhang MM. Radiation dose of non-enhanced chest CT can be reduced 40% by using iterative reconstruction in image space. Clin Radiol 2011; 66: Christner JA, Kofler JM, McCollough CH. Estimating effective dose for CT using dose-length product compared with using organ doses: consequences of adopting International Commission on Radiological Protection publication 103 or dualenergy scanning. AJR 2010; 194: Pontana F, Pagniez J, Flohr T, et al. Chest computed tomography using iterative reconstruction vs filtered back projection. Part 1. Evaluation of image noise reduction in 32 patients. Eur Radiol 2011; 21: Friedman SN, Fung GS, Siewerdsen JH, Tsui BM. A simple approach to measure computed tomography (CT) modulation transfer function (MTF) and noise-power spectrum (NPS) using the American College of Radiology (ACR) accreditation phantom. Med Phys 2013; 40: Singh S, Kalra MK, Hsieh J, et al. Abdominal CT: comparison of adaptive statistical iterative and filtered back projection reconstruction techniques. Radiology 2010; 257: AJR:203, October