Gastrointestinal Imaging Original Research

Transcription

1 Gastrointestinal Imaging Original Research Schindera et al. Abdominal CT for Obese Patients Gastrointestinal Imaging Original Research Sebastian T. Schindera 1,2 Devang Odedra 1 Diego Mercer 1 Seng Thipphavong 1 Paul Chou 1 Zsolt Szucs-Farkas 3 Patrik Rogalla 1 Schindera ST, Odedra D, Mercer D, et al. Keywords: CT, image quality, iterative reconstruction technique, obese patients, radiation dose DOI: /AJR Received December 27, 2012; accepted after revision May 30, P. Rogalla received a research grant from Toshiba Medical Systems. 1 Department of Medical Imaging, University Health Network, University of Toronto, Toronto, ON, Canada. 2 Department of Radiology and Nuclear Medicine, University Hospital Basel, Petersgraben 4, CH-4031 Basel, Switzerland. Address correspondence to S. T. Schindera (sschindera@aol.com). 3 Department of Diagnostic Radiology, Hospital Center of Biel, Biel, Switzerland. WEB This is a web exclusive article. AJR 2014; 202:W146 W X/14/2022 W146 American Roentgen Ray Society Hybrid Iterative Reconstruction Technique for Abdominal CT Protocols in Obese Patients: Assessment of Image Quality, Radiation Dose, and Low-Contrast Detectability in a Phantom OBJECTIVE. The purpose of this study was to assess the impact of a noise reduction technique on image quality, radiation dose, and low-contrast detectability in abdominal CT for obese patients. MATERIALS AND METHODS. A liver phantom with 12 different tumors was designed, and fat rings were added to mimic intermediately sized and large patients. The intermediate and large phantoms were scanned with our standard abdominal CT protocol (image noise level of 15 HU and filtered back projection [FBP]). The large phantom was scanned with five different noise levels (10, 12.5, 15, 17.5, and 20 HU). All datasets for the large phantom were reconstructed with FBP and the noise reduction technique. The image noise and the contrast-to-noise ratio (CNR) were assessed. Tumor detection was independently performed by three radiologists in a blinded fashion. RESULTS. The application of the noise reduction method to the large phantom decreased the measured image noise (range, 14.5% to 37.0%) and increased the CNR (range, %) compared with FBP at the same noise level (p < 0.001). However, noise reduction was unable to improve the sensitivity for tumor detection in the large phantom compared with FBP at the same noise level (p > 0.05). Applying a noise level of 15 HU, the overall sensitivity for tumor detection in the intermediate and large phantoms with FBP measured 75.5% and 87.7% and the radiation doses measured 42.0 and 23.7 mgy, respectively. CONCLUSION. Although noise reduction significantly improved the quantitative image quality in simulated large patients undergoing abdominal CT compared with FBP, no improvement was observed for low-contrast detectability. O besity has reached epidemic proportions in many developed countries around the globe. Because obesity is associated with various medical conditions, including cardiovascular disease, diabetes mellitus, cholelithiasis, abdominal hernias, and different types of cancers (e.g., breast and colorectal cancer), many of these countries may find themselves facing a major public health problem in the future [1]. CT plays an important role in the diagnostic workup and therapeutic follow-up for many of these conditions. However, CT examinations of obese patients generally pose two significant challenges for radiologists: poor image quality and the high radiation dose required. In obese patients, the poor image quality is caused by a greater absorption of the x-ray beam by the subcutaneous and visceral fat. Fewer incident photons contribute to image formation, which results in increased image noise. For abdominal CT examinations, high image noise levels are a critical issue because the noise may obscure subtle low-contrast lesions in parenchymal organs, such as the liver, pancreas, and spleen [2, 3]. To obtain diagnostic-quality images in large patients, the CT protocols need to be modified. The widely applied automatic tube current modulation technique helps radiologists maintain image quality over a range of patient sizes by automatically adjusting the tube current to the x- ray attenuation of the patient section being scanned. However, the modified CT protocols for large patients can result in a substantial increase in the patient radiation dose [4]. CT manufacturers have introduced multiple noise reduction methods, including iterative reconstruction algorithms for image noise improvement, to allow a radiation dose reduction while maintaining image quality [5 7]. Two recent studies have reported a loss of low-contrast detectability for low-dose abdominal CT W146 AJR:202, February 2014

2 Abdominal CT for Obese Patients despite the application of iterative reconstruction algorithms [8, 9]. However, to date, there is limited knowledge of the impact of noise reduction or iterative reconstruction algorithms on CT examinations of large patients [10]. To optimize abdominal CT protocols for large patients with a new noise reduction method, the impact of the iterative reconstruction algorithm on image quality and low-contrast detectability needs to be addressed. This prompted us to design a phantom that allows multiple repeatable CT studies with different technical parameters. The objective of our phantom study was to assess the impact of a new noise reduction method on image quality, radiation dose, and lowcontrast detectability in obese patients undergoing abdominal CT. Materials and Methods Liver Phantom A custom liver phantom (QRM) was manufactured to mimic the liver parenchyma during the portal venous phase. The liver phantom had a cylindric shape (length, 29.2 cm; diameter, 15.0 cm) and contained multiple spherical lesions that simulated hypodense hepatic metastases (e.g., metastases from colorectal, lung, or breast cancer) (Fig. 1). We decided to simulate hepatic lesions because they represent one of the most challenging pathologies to detect using abdominal CT in obese patients. In addition, the portal venous phase is by far the most frequently used enhancement phase of abdominal CT. The material of the liver parenchyma and the tumors was composed of a homogeneous mixture of resin, including additives such as calcium carbonate and organic iodine. The mean attenuation of the phantom liver parenchyma was measured to be ± (SD) 2.7 HU at 120 kvp. The simulated hepatic lesions had three different tumorto-liver contrast values ( 10, 20, and 40 HU) and four different diameters, measuring 5, 10, 15, and 20 mm (Fig. 2). The liver phantom contained a total of 12 different lesions. There was only one lesion per slice. To simulate patients with intermediate and large body sizes, a fat ring with a diameter of 30 or 40 cm, respectively, was added to the phantom (Fig. 1). The size of the fat ring was selected to match the abdominal cross-sectional dimension of an intermediately sized patient with an estimated body weight ranging between 72 and 85 kg and a large patient with an estimated body weight ranging between 118 and 142 kg [11]. The mean attenuation of the fat ring measured 95.7 ± 0.8 HU at 120 kvp. Fig. 1 Photograph shows liver phantom (black star) with added fat ring (single white star) that mimics intermediately sized patient (total diameter, 30 cm) and larger fat ring (double white stars), without liver phantom, that mimics larger patient (total diameter, 40 cm). Fig. 2 CT images of hyperdense round structure in center of phantom shows simulated liver (black star) surrounded by hypodense fat ring (white star), mimicking large patient. Each of four images contains one simulated hypodense tumor (arrows) with different diameter (5, 10, 15, and 20 mm, from top right to top left in clockwise direction, respectively) in liver. All tumors have same tumor-to-liver contrast of 40 HU. Same window settings are applied (window width, 350 HU; window level, 40 HU). Images were acquired with noise level of 10 HU and were reconstructed with filtered back projection algorithm. CT Scanning A 320-MDCT scanner (Aquilion ONE, Toshiba Medical Systems) was used to scan the liver phantom with the added fat rings. First, the intermediate and large phantoms were scanned with our standard abdominal CT protocol using an image noise level of 15 HU (Sure-Exposure 3D, Toshiba Medical Systems). Then, the large phantom was scanned with four additional protocols using different image noise levels (10, 12.5, 17.5, and 20 HU). The image noise level is the operator-selected image quality setting of the automatic tube current modulation. Automatic tube current modulation maintains the image noise for different-sized patients by adjusting the tube current. Because the phantom had a round shape and the attenuation was maintained along the z-axis, the applied tube current was constant. The applied tube current time product settings for the different protocols are given in Table 1. The tube voltage (120 kvp), collimation ( mm), and pitch (0.83) were maintained for all CT studies. The intermediate phantom was only reconstructed with the filtered back projection (FBP) algorithm, and the large phantom was reconstructed with the FBP and novel noise reduction method (AIDR-3D, Toshiba Medical Systems). In the applied noise reduction method, the raw data are first reconstructed with FBP before statistical and scanner models combined with projection noise estimation are applied for photon and electronic noise reduction. This step is followed by an iterative technique to optimize the reconstructions by detecting and preserving sharp details while simultaneously smoothing the image. Finally, weighted blending is applied to the original reconstruction and the output of this iterative process to maintain the noise granularity. The standard noise reduction settings of three levels (weak, standard, and strong) were applied for all scans reconstructed with noise reduction. The use of the standard AIDR-3D setting is recommended by the CT manufacturer for abdominal CT. The level of the reconstruction refers to the strength of noise reduction. The reconstruction speed of AIDR-3D is the same as for the FBP (at least 30 frames per second). The phantom was scanned eight times with each protocol, and for each scan the phantom was rotated 45 clockwise. The rotation of the phantom provided the same lesion at eight different locations. The phantom was positioned within the isocenter of the CT scanner with its cross section perpendicular to the scanner s z-axis (Fig. 1). The FBP and AIDR-3D AJR:202, February 2014 W147

3 Schindera et al. datasets were reconstructed with different reconstruction filters (FC04 and FC19, respectively), as suggested by the CT manufacturer. The reconstruction filter FC04 is slightly softer and uses a beamhardening correction compared with the filter FC19. Assessment of Objective Image Quality and Radiation Dose Assessment of quantitative image quality was performed by one of the authors who is a radiologist with subspecialty training in abdominal imaging. CT attenuation values of the liver parenchyma and the tumors were obtained for the 11 CT datasets by manually placing circular regions of interest (ROIs). To avoid misregistration due to the small tumor size, the lesion with a 20-mm diameter and a tumor-toliver contrast of 40 HU was used for attenuation measurements. The average of the ROIs was 1.4 cm 2 (range, cm 2 ) for the tumor and 3.9 cm 2 (range, cm 2 ) for the liver parenchyma. The attenuation value of the liver parenchyma was measured adjacent to the tumor. To ensure consistency, all measurements were performed three times, and mean values were calculated. Image noise was defined as the SD of the attenuation value measured in the liver parenchyma. The tumor-to-liver contrastto-noise ratio (CNR) was calculated as CNR = (ROI L ROI T ) / N, where ROI L is the mean attenuation value of the liver parenchyma, ROI T is the mean attenuation value of the simulated tumor, and N is noise. The volume CT dose index, provided by the CT scanner, was recorded for the different CT protocols. Assessment of Low-Contrast Detectability To assess the low-contrast detectability, we created dedicated datasets containing contiguous 5-mmthick transverse CT images (total number of images, 352). There were a total of 11 dedicated datasets available, and each dataset included 60 simulated lesions. The minimal number of lesions needed for the appropriate statistical power was calculated (see Statistical Analysis section). Each of the 12 different lesions appeared five times in the same dataset. To minimize the recall bias, the location of the same lesion was always different, and the succession of the lesions was randomized. The datasets were created using in-house software that was written in MATLAB 7.8 software (MathWorks). The development of this program was customized and tailored to the requirements of the current study. Three radiologists with subspecialty training in abdominal imaging evaluated the 11 CT datasets independently on the same high-definition liquid crystal display monitor (RadiForce GX 320, Eizo). The reading of the 11 datasets was divided in two sessions to avoid fatigue. The readers were blinded to the number, location, and size of the simulated tumors and to the scanning protocol. Before starting the assessment, each reader was given the criteria for image grading, and various test cases were assessed together. The readers were asked to mark the location of the tumors on evaluation sheets, along with the grade of conspicuity (grade 1, may be present; grade 2, most likely present; grade 3, definitely present). The 11 CT datasets were reviewed in the following order: the FBP dataset of the large phantom with a noise level of 20 HU, the AIDR-3D dataset of the large phantom with a noise level of 20 HU, the FBP dataset of the large phantom with a noise level of 17.5 HU, the AIDR-3D dataset of the large phantom with a noise level of 17.5 HU, the FBP dataset of the large phantom with a noise level of 15 HU, the AIDR-3D dataset of the large phantom with a noise level of 15 HU, the FBP dataset of the large phantom TABLE 1: Assessment of Image Noise, Contrast-to-Noise Ratio (CNR), and Radiation Dose CT Dataset Phantom Size Noise Level (HU) Applied mas Image Reconstruction with a noise level of 12.5 HU, the AIDR-3D dataset of the large phantom with a noise level of 12.5 HU, the FBP dataset of the large phantom with a noise level of 10 HU, the AIDR-3D dataset of the large phantom with a noise level of 10 HU, and the FBP dataset of the intermediate phantom with a noise level of 15 HU. Although the images were initially presented with a preset soft-tissue window (window width, 350 HU; window level, 40 HU), the readers were allowed to modify the window width and level at their own discretion. Ambient room lighting was maintained at a low and constant level for the period of review. Subjective Lesion Conspicuity A side-by-side analysis was chosen to assess the subjective lesion conspicuity. For this assessment, the original scanned datasets of the phantom (at 0 ) were used. The 11 datasets were presented simultaneously on two high-definition liquid crystal display monitors (RadiForce GX 320, Eizo), and the order of the datasets was randomized. The side-byside analysis had the advantage that the three readers were able to compare the exact lesion of the 11 datasets next to each other. The three readers rated image noise (grade 1, major, unacceptable; grade 2, substantial, above average; grade 3, moderate, average; grade 4, minor, below average; grade 5, absent) and overall image quality (grade 1, bad, no diagnosis possible; grade 2, poor, diagnostic confidence substantially reduced; grade 3, moderate, but sufficient for diagnosis; grade 4, good; grade 5, excellent). The readers were blinded to the scanning protocols. Before the subjective reading session, the three readers took part in a short introductory session. Again, the readers were allowed to modify the window width and level at their own discretion. Image Noise (HU) Difference (%) CNR Difference (%) CTDI vol (mgy) Intermediate FBP 14.8 ± 0.99 NA 2.8 ± 0.03 NA 23.7 Large FBP 10.1 ± ± AIDR-3D 8.2 ± ± FBP 11.7 ± ± AIDR-3D 10.0 ± ± FBP 15.0 ± ± AIDR-3D 11.3 ± ± FBP 17.1 ± ± AIDR-3D 12.0 ± ± FBP 20.5 ± ± AIDR-3D 12.9 ± ± 0.10 Note Difference data are between AIDR-3D (Toshiba Medical Systems) and filtered back projection (FBP). The CNR values were calculated for the 20-mm, 40 HU tumor. CTDI vol = volume CT dose index, NA = not applicable. W148 AJR:202, February 2014

4 Abdominal CT for Obese Patients Statistical Analysis In the preliminary power analysis, we assumed a mean sensitivity of 0.8 for lesion detection, which was based on former reader studies [12]. We calculated that 60 lesions per CT protocol would confirm a difference of 8.6% in sensitivity between FBP and the noise reduction method (AIDR-3D) at a power of 0.8 and an α-level of The readers marks were classified as true-positive and false-positive. The construction plan of the phantom was used as the reference standard. Because reader data were collected in a free-response manner, we planned to use an TABLE 2: Detection Data for 60 Simulated Hepatic Tumors CT Dataset No of Findings Phantom Size Noise Level (HU) Image Reconstruction CTDI Conspicuity a vol (mgy) True-Positive False-Positive False-Negative Overall Sensitivity (%) Intermediate 15 FBP (2.0, 3.0) Large 10 FBP (1.3, 3.0) AIDR-3D 2.2 (1.5, 3.0) FBP (0.3, 3.0) AIDR-3D 2.1 (1.3, 3.0) FBP (0.3, 3.0) AIDR-3D 2.0 (0.8, 3.0) FBP (0.0, 3.0) AIDR-3D 1.9 (0.3, 3.0) FBP (0.3, 3.0) AIDR-3D 1.8 (0.3, 3.0) Note Data are the means of the results from three independent readers. CTDI vol = volume CT dose index, FBP = filtered back projection, AIDR-3D (Toshiba Medical Systems) = applied noise reduction method. a The conspicuity of the tumors was rated on a 3-point scale: grade 1, may be present; grade 2, most likely present; grade 3, definitely present. The data in parentheses are the 25th and 75th percentiles. Fig. 3 CT images show simulated hepatic tumor (at 12-o clock position) with tumor-to-liver contrast of 10 HU and diameter of 10 mm in phantoms of intermediately sized and large patients scanned with different CT protocols and reconstructed with filtered back projection (FBP) and iterative noise reduction method (AIDR-3D, Toshiba Medical Systems). Top row shows intermediately sized phantom, and second and third rows show large phantom. Same window settings are applied (window width, 350 HU; window level, 40 HU). Tumor is easily detected in intermediate phantom (arrow in top row). In large phantom, scanned with noise levels of 10 and 12.5 HU, tumor is poorly visualized (arrows). With noise levels of 15, 17.5, and 20 HU, tumor cannot be detected in large phantom. CT noise reduction using AIDR-3D decreases image noise in large phantom at same noise level, but no improvement in tumor detection is observed compared with FBP. AJR:202, February 2014 W149

5 Schindera et al. appropriate method (jackknife alternative free-response receiver operating characteristic [JAFROC]) for the statistical analysis. However, the total number of false-positive findings marked by the readers was only with each protocol. At a low false-positive rate, the JAFROC method is unreliable. Thus, an analysis of variance for repeated measurements with a post hoc test was used to assess the readers data. We calculated the weighted kappa value to evaluate the interobserver agreement. Agreement between the readers was graded as follows: poor, less than 0.20; fair, ; moderate, ; good, ; and very good, [13]. Statistical tests were performed with Statistica (StatSoft) and MedCalc (MedCalc) software, and p values of less than 0.05 were considered statistically significant. Results Objective Image Quality and Radiation Dose Similar image noise and CNR values were achieved in the large compared with the intermediate phantoms with the FBP algorithm by increasing the radiation dose from 23.7 to 42.0 mgy, respectively (Table 1). Conversely, at similar radiation dose levels in the large and intermediate phantoms, both reconstructed using the FBP algorithm, the image noise was increased from 14.8 to 20.5 HU and the CNR was decreased from 2.8 to 1.7 in the large phantom compared with the intermediate phantom. Compared with the FBP algorithm, AIDR-3D yielded an image noise reduction ranging between 14.5% and 37.0% at the five radiation dose levels in the large phantom (p < 0.001). Additionally, the noise reduction method improved the CNR of the simulated tumors, ranging between 26.7% and 70.6% (p < 0.001). In general, as the radiation dose decreased and the image noise increased, the impact of the noise reduction method on the image noise reduction and CNR improvement increased. The large phantom reconstructed with the novel noise reduction method showed lower image noise and higher CNR values at all five radiation dose levels compared with the intermediate phantom reconstructed with the FBP algorithm. Low-Contrast Detectability At the noise level of 15 HU, the three readers detected 14.0% fewer tumors in the large phantom compared with the intermediate phantom when both were reconstructed with the FBP algorithm (overall sensitivity of 75.5% vs 87.7%, respectively) (Table 2). This difference failed to reach statistical significance (p = 0.84). In the large phantom, fewer tumors with a smaller diameter and lower contrast value were detected compared with the intermediate phantom (Table 3). Despite dose increases of 135.0% and 252.0%, the CT protocols with noise levels of 12.5 and 10 HU, respectively, in the large phantom still yielded smaller detection rates than the CT protocol with a noise level of 15 HU in the intermediate phantom, although the differences were not significant (p = 0.88 and 0.95, respectively). Image reconstruction with AIDR-3D did not improve the sensitivity of low-contrast detectability compared with FBP at the same noise levels (p > 0.05) (Fig. 3). Because the noise level increased and the radiation dose of the CT protocols in the large phantom decreased, the three readers detected fewer tumors with both reconstruction methods (Fig. 3). The overall sensitivity decreased from 80.0% (noise level of 10 HU with FBP) to 65.0% (noise level of 20 HU with FBP). The comparison for the true-positive findings between the different CT protocols in the large phantom reconstructed with FBP and the AIDR-3D did not show any significant difference (p between and 1.0). The application of this iterative noise reduction method to the large phantom did not result in a significant improvement for the detection of low-contrast or small tumors in comparison with the FBP algorithm (p between and 1.0) (Table 3). The interobserver agreement among the three readers for the true-positive findings was good (mean weighted κ = 0.75; range, ). No difference was found between the image reconstruction methods for the number of false-positive marks at any dose level (p between and 1.0) (Table 2). Subjective Lesion Conspicuity At the same noise level of 15 HU, the three independent readers rated the overall subjec- TABLE 3: Number of True-Positive Findings by Contrast Value and Diameter of the Simulated Tumors CT Dataset Tumor-to-Liver Contrast (HU) Tumor Diameter (mm) Phantom Size Noise Level (HU) Image Reconstruction Intermediate 15 FBP Large 10 FBP AIDR-3D FBP AIDR-3D FBP AIDR-3D FBP AIDR-3D FBP AIDR-3D Large All noise levels FBP AIDR-3D Note Data were derived from reports of three independent and blinded readers. The numbers in the last row represents pooled data for the large phantom across all of noise levels. The total number of simulated tumors for each tumor-to-liver contrast value was 20 for each noise level and 100 for all noise levels together. The total number of simulated tumors for each tumor diameter was 15 for each noise level and 75 for all noise levels together. FBP = filtered back projection, AIDR-3D (Toshiba Medical Systems) = applied noise reduction method. W150 AJR:202, February 2014

6 tive image noise and image quality of the intermediate phantom substantially better compared with the large phantom (Table 4). In the large phantom, AIDR-3D was unable to substantially improve the overall subjective image noise and image quality compared with FBP. The grading of overall subjective image noise and image quality revealed that readers rankings were either slightly better or worse for the AIDR-3D datasets than they were for the FBP datasets at the same noise level. Abdominal CT for Obese Patients TABLE 4: Subjective Assessment of the Image Noise and Overall Image Quality Phantom Size Noise Level (HU) Image Reconstruction Image Noise Overall Image Quality Intermediate 15 FBP 4.0 ± ± 1.5 Large 10 FBP 3.7 ± ± AIDR-3D 3.3 ± ± FBP 3.3 ± ± AIDR-3D 4.0 ± ± FBP 2.3 ± ± AIDR-3D 2.7 ± ± FBP 2.3 ± ± AIDR-3D 2.0 ± ± FBP 2.3 ± ± AIDR-3D 2.3 ± ± 0 Note Values are mean ± SD as derived from single scores per series given by three readers. Image noise was rated on a 5-point scale: grade 1, major, unacceptable; grade 2, substantial, above average; grade 3, moderate, average; grade 4, minor, below average; and grade 5, absent. Overall image quality was also rated on a 5-point scale: grade 1, bad, no diagnosis possible; grade 2, poor, diagnostic confidence substantially reduced; grade 3, moderate but sufficient for diagnosis; grade 4, good; and grade 5, excellent). FBP = filtered back projection, AIDR-3D (Toshiba Medical Systems) = applied noise reduction method. Discussion Obese patients undergoing abdominal CT receive substantially higher radiation doses and have a substantially decreased low-contrast detectability compared with intermediate patients [4, 14]. Thus, optimization of abdominal CT protocols for large patients in regard to radiation dose and low-contrast detectability is urgently required. Unfortunately, the two factors are inversely correlated, which makes the task challenging. In our phantom study, we attempted to optimize the CT protocols for obese patients by applying an iterative noise reduction method (AIDR-3D). With this technique, we assessed the potential for radiation dose reduction while maintaining image quality and low-contrast detectability. Our phantom study showed a significant increase in quantitative image quality assessed by image noise and CNR using the noise reduction method compared with FBP in the simulated large patient. A recent clinical study by Desai et al. [10] also showed a significant improvement in image quality using another noise reduction method (adaptive statistical iterative reconstruction [ASIR, GE Healthcare]) in patients weighing between 91 and 182 kg and undergoing abdominal CT. The radiation dose was lowered by 31.5% with their noise reduction method compared with the FBP algorithm, and the authors concluded that their noise reduction method provided diagnostic-quality images in obese patients at a substantially lower radiation dose. The shortcoming of this study was that the diagnostic effectiveness, which is the second level for the health technology assessment of diagnostic tools, was not assessed [15, 16]. The recommendation for dose reduction using CT noise reduction was based only on the assessment of the technical efficacy, which is the first level for a health technology assessment. The low-contrast detectability in our liver phantom did not improve as much as the image quality using CT noise reduction method. The AIDR-3D and FBP datasets showed similar sensitivities for tumor detection at the same radiation dose, and both reconstruction methods showed a similar trend for lower low-contrast detectability in large patients as the radiation dose decreased. Similar findings have been reported for other noise reduction techniques from other CT manufacturers (ASIR, GE Healthcare, and idose 4, Philips Healthcare) [17]. The authors performed a detection experiment of low-contrast spherical objects in a phantom, which was not modified to simulate large patients and showed no significant improvement in object detectability for noise reduction compared with the FBP algorithm as the radiation dose decreased. Because CT noise reduction primarily improves the image quality and not lesion detection, we recommend applying CT noise reduction with caution for radiation dose reduction in obese patients to avoid a negative influence on the low-contrast detectability. Furthermore, it is important to point out that the CT Accreditation Program of the American College of Radiology requires low-contrast assessment in a phantom, which is primarily based on CNR values. Unfortunately, our study shows that CNR is not a meaningful parameter for evaluating the potential for radiation dose reduction with CT noise reduction methods in comparison with a FBP. Another interesting result of our study was that the image noise level in the large phantom must be substantially lower than in the intermediate phantom to achieve similar sensitivities for tumor detection. At the same time, the radiation dose increased at least 3.5-fold in the large phantom compared with the intermediate phantom. Despite having the same image quality in the large phantom and the intermediate phantom, tumor detection was decreased by 14.0% in the large phantom. We do not have a plausible explanation for why the same image quality in simulated patients with different body habitus does not result in the same sensitivity. Further investigations of the various physical characteristics of CT images in differently sized patients are required. Previous studies have shown that subjective image quality increases with patient size on abdominal CT images with constant image noise [18, 19] that is, radiologists tend to accept greater noise for the same subjective image quality for larger patients. The increased subjective image quality for larger patients may be explained by the increase in fat deposition around the abdominal organs that adds to the inherent tissue contrast between various organs. Thus, some authorities recommend greater acceptance of higher noise for large patients to lower the delivered radiation dose [20]. However, higher image noise levels result in a substantial decrease in low-contrast detectability in large patients compared with intermediate-sized patients, as shown by our study results. When we applied a noise level of 20 HU to the large phantom, the low-contrast detectability decreased by 25.9% compared with a noise level of 15 HU in the intermediate phantom. Thus, the image quality for abdominal CT studies in large patients AJR:202, February 2014 W151

7 Schindera et al. must be adjusted to the clinical indication to simultaneously maximize the dose efficiency and diagnostic performance. Several potential limitations of our study merit consideration. First, we assessed the diagnostic effectiveness and quantitative image quality of using the noise reduction method of only one CT manufacturer. Because the technical approaches of CT noise reduction differ among methods and CT manufacturers, it is not possible to automatically translate our findings to other major CT manufacturers. Second, our liver phantom was simplified to mimic the enhancement condition of the hepatic parenchyma and hypovascular liver tumors existing momentarily during the portal venous phase. However, our liver phantom did not model dynamic contrast enhancement, heterogeneous parenchymal enhancement, distorted hepatic anatomy, hepatic steatosis, or different enhancement patterns of hepatic tumors. All these factors can influence the detection of hepatic tumors, particularly if they are smaller than 1 cm. Third, there was a possible recall bias of the three readers in regard to the detection of the simulated tumors, although the succession and location of the different tumors was randomized for the different datasets. Fourth, the phantom s shape was round and not elliptical, as typically observed in patients. Because the studied noise reduction method was developed for patients, the novel noise reduction method we studied might show a greater impact on noise reduction and low-contrast detectability in an elliptical than in a round shape. Fifth, we did not investigate the impact of different tube voltages on the diagnostic accuracy of CT noise reduction. For our investigation, we chose a tube voltage of 120 kvp because it is by far the most frequently used voltage for abdominal CT. In the current study, we decided to lower the tube current to a fixed tube voltage of 120 kvp. Changing both technical parameters simultaneously would create data that would be difficult to interpret. The impact of lower tube voltages (e.g., 100 kvp) on low-contrast detectability was assessed by our research group in a previous investigation [21]. In conclusion, our phantom data show that a novel noise reduction method significantly improved the quantitative image quality but not the low-contrast detectability in obese patients undergoing abdominal CT. Therefore, radiologists should be cautious when applying CT noise reduction methods for dose reduction in obese patients. The results of the current investigations need to be validated in patients. References 1. Mun EC, Blackburn GL, Matthews JB. Current status of medical and surgical therapy for obesity. Gastroenterology 2001; 120: 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: Funama Y, Awai K, Miyazaki O, et al. Improvement of low-contrast detectability in low-dose hepatic multidetector computed tomography using a novel adaptive filter: evaluation with a computersimulated liver including tumors. Invest Radiol 2006; 41: Schindera ST, Nelson RC, Toth TL, et al. Effect of patient size on radiation dose for abdominal MDCT with automatic tube current modulation: phantom study. AJR 2008; 190:[web]W100 W Marin D, Nelson RC, Schindera ST, et al. Lowtube-voltage, high-tube-current multidetector abdominal CT: improved image quality and decreased radiation dose with adaptive statistical iterative reconstruction algorithm initial clinical experience. Radiology 2010; 254: 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: 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: Baker ME, Dong F, Primak A, et al. Contrast-tonoise ratio and low-contrast object resolution on full- and low-dose MDCT: SAFIRE versus filtered back projection in a low-contrast object phantom and in the liver. AJR 2012; 199: Pickhardt PJ, Lubner MG, Kim DH, et al. Abdominal CT with model-based iterative reconstruction (MBIR): initial results of a prospective trial comparing ultralow-dose with standard-dose imaging. AJR 2012; 199: Desai GS, Uppot RN, Yu EW, Kambadakone AR, Sahani DV. Impact of iterative reconstruction on image quality and radiation dose in multidetector CT of large body size adults. Eur Radiol 2012; 22: Menke J. Comparison of different body size parameters for individual dose adaptation in body CT of adults. Radiology 2005; 236: Schindera ST, Diedrichsen L, Muller HC, et al. Iterative reconstruction algorithm for abdominal multidetector CT at different tube voltages: assessment of diagnostic accuracy, image quality, and radiation dose in a phantom study. Radiology 2011; 260: Landis JR, Koch GG. The measurement of observer agreement for categorical data. Biometrics 1977; 33: Schindera ST, Torrente JC, Ruder TD, et al. Decreased detection of hypovascular liver tumors with MDCT in obese patients: a phantom study. AJR 2011; 196:[web]W772 W Sardanelli F, Hunink MG, Gilbert FJ, Di Leo G, Krestin GP. Evidence-based radiology: why and how? Eur Radiol 2010; 20: Fryback DG, Thornbury JR. The efficacy of diagnostic imaging. Med Decis Making 1991; 11: Miéville FA, Gudinchet F, Brunelle F, Bochud FO, Verdun FR. Iterative reconstruction methods in two different MDCT scanners: physical metrics and 4-alternative forced-choice detectability experiments a phantom approach. Phys Med 2013; 29: Kalra MK, Maher MM, Kamath RS, et al. Sixteen-detector row CT of abdomen and pelvis: study for optimization of Z-axis modulation technique performed in 153 patients. Radiology 2004; 233: Wilting JE, Zwartkruis A, van Leeuwen MS, Timmer J, Kamphuis AG, Feldberg M. A rational approach to dose reduction in CT: individualized scan protocols. Eur Radiol 2001; 11: Li J, Udayasankar UK, Tang X, Toth TL, Small WC, Kalra MK. Patient size compensated automatic tube current modulation in multi-detector row CT of the abdomen and pelvis. Acad Radiol 2011; 18: Schindera ST, Diedrichsen L, Müller HC, et al. Iterative reconstruction algorithm for abdominal multidetector CT at different tube voltages: assessment of diagnostic accuracy, image quality, and radiation dose in a phantom study. Radiology 2011; 260: W152 AJR:202, February 2014