ª Springer Science+Business Media New York 2014

Abdominal Imaging

Abdom Imaging (2014) DOI: 10.1007/s00261-014-0081-5

CT of the pancreas: comparison of image quality and pancreatic duct depiction among model-based iterative, adaptive statistical iterative, and filtered back projection reconstruction techniques Xiao-Zhu Lin,1 Haruhiko Machida,2 Isao Tanaka RT,2 Rika Fukui RT,2 Eiko Ueno,2 Ke-Min Chen,1 Fu-Hua Yan1 1

Department of Radiology Ruijin Hospital, Shanghai Jiaotong University School of Medicine, 197 2nd Ruijin Road, Shanghai 200025, China 2 Department of Radiology, Tokyo Women’s Medical University Medical Center East, 2-1-10 Nishiogu, Arakawa-ku, Tokyo 116-8567, Japan

Abstract The purpose of this study is to compare CT images of the pancreas reconstructed with model-based iterative reconstruction (MBIR), adaptive statistical iterative reconstruction (ASiR), and filtered back projection (FBP) techniques for image quality and pancreatic duct (PD) depiction. Data from 40 patients with contrastenhanced abdominal CT [CTDIvol: 10.3 ± 3.0 (mGy)] during the late arterial phase were reconstructed with FBP, 40% ASiR–FBP blending, and MBIR. Two radiologists assessed the depiction of the main PD, image noise, and overall image quality using 5-point scale independently. Objective CT value and noise were measured in the pancreatic parenchyma, and the contrast-tonoise ratio (CNR) of the PD was calculated. The Friedman test and post-hoc multiple comparisons with Bonferroni test following one-way ANOVA were used for qualitative and quantitative assessment, respectively. For the subjective assessment, scores for MBIR were significantly higher than those for FBP and 40% ASiR (all P < 0.001). No significant differences in CT values of the pancreatic parenchyma were noted among FBP, 40% ASiR, and MBIR images (P > 0.05). Objective image noise was significantly lower and CNR of the PD was higher with MBIR than with FBP and 40% ASiR (all P < 0.05). Our results suggest that pancreatic CT ima-

Correspondence to: Fu-Hua Yan; email: [email protected]

ges reconstructed with MBIR have lower image noise, better image quality, and higher conspicuity and CNR of the PD compared with FBP and ASiR. Key words: Pancreas—Computed tomography— Iterative reconstruction—Image quality

Multidetector row computed tomography (MDCT) of the pancreas is a widely accepted technique that is used in the detection and preoperative staging of pancreatic neoplasms [1, 2]. Patients with suspected pancreatic pathology often have abnormalities of the pancreatic duct (PD). Previous studies showed that slight dilatation of the main PD and the presence of pancreatic cysts was the strong independent predictors of the subsequent development of pancreatic cancer [3–5]. The evaluation of intraductal papillary mucinous neoplasms of the pancreas (IPMNs) with MDCT for the differentiation of malignant from benign disease has been attempted [6–9]. Improvements in MDCT technology have ensured optimal scan timing and permitted thin image reconstruction for the pancreas imaging. PDs were better delineated on the thin slice images, which may help assessing of the IPMNs and discriminating of benign and malignant PD strictures. A normal PD is tortuous and the size of its lumen is very small especially in the body and tail. In spite of advances in MDCT technology, the evaluation of normal or minimally dilated PDs at CT still remains crucial because of the small diameter and low contrast of the duct.

X.-Z. Lin et al.: CT of the pancreas

Image reconstruction algorithms play a critical role in the quality and appearance of tomographic images [10, 11]. Filtered back projection (FBP) algorithms were used for CT image reconstruction owing to their faster image reconstruction and ease of implementation [12]. Over the past decade, the desire for finer resolution, greater volume coverage, and faster scan times and the desire to concurrently lower radiation dose have pushed the performance of FBP reconstruction to its limits. Iterative reconstruction of image data, an imaging processing technique that was previously used in nuclear medicine, has been recently introduced for MDCT scanners with the goal of reducing image noise [13]. Studies [13–16] on abdominal CT showed the benefit of the adaptive statistical iterative reconstruction (ASiR; GE Healthcare, Milwaukee, WI) technique in terms of lower image noise and radiation dose. Recently, modelbased iterative reconstruction (MBIR; Veo for commercial name; GE Healthcare, Milwaukee, WI) has been introduced after ASiR and has become clinically available as a novel CT reconstruction algorithm. It includes an algorithm that accurately models the entire optical chain (real size of focal spots and detectors) and takes into account the noise of the system (photons statistics and electronic noise). From a raw dataset and without an initial FBP reconstruction, MBIR uses backward and forward projections to match the reconstructed image to the acquired data iteratively according to a statistical metric. Several recent studies [17–20] indicated that MBIR technique, by modeling these optical effects, may improve the image quality and spatial resolution and reduce streaking artifacts better than the widespread CT reconstruction algorithms such as ASiR and FBP do. Thus, MBIR is expected to improve image quality and reduce radiation and contrast-medium dose in CT study of various clinical fields. To our knowledge, however, the effect of MBIR algorithm on pancreas imaging especially for the evaluation of the PDs has not been fully investigated. The purpose of this study was to determine if the use of MBIR can improve visualization of the PD and image quality of the pancreas on MDCT images when compared with the use of ASiR and FBP images reconstructed from the same raw dataset.

Materials and methods Subjects This study protocol was approved by the institutional review committee of Tokyo Women’s Medical University. Standard-of-care datasets were used to generate the three series of reconstructions for retrospective analysis and no patient’s informed consent was required. ASiR and MBIR became available at Tokyo Women’s Medical University Medical Center East for clinical CT imaging in March 2011. During the following 2-month period (March through April 2011), a total of 45 (40 assessed and 5 training) consecutive patients who underwent a standard contrast-enhanced dual-phase abdominal CT were

included in this study in order to evaluate the level of noise reduction and overall image quality improvement with ASiR and MBIR technique. For the 40 patients assessed, there were 24 men and 16 women. The mean age was 63 ± 13 years ranging from 37 to 91 years. The average body weight was 58.8 ± 10.9 (kg). These patients were referred for various abdominal disorders such as gastric cancer, colon cancer, hepatic cyst, hepatic hemangioma, hepatic cancer, adrenal gland tumor, renal cancer, ovarian caner, abdominal pain, and hematuria. Patients with pancreatic disease and bile duct disease were excluded from this study.

Data acquisition and scanning protocols All examinations were performed by using a 64-row MDCT system (Discovery CT750 HD; GE Healthcare, Milwaukee, WI). Patients were placed in the supine position, feet first, on the CT table. The following parameters for the contrast-enhanced helical CT scan were applied: tube voltage, 120 kV; collimation, 64 9 0.625 mm; gantry rotation time, 0.5 s; and pitch, 1.375:1 for a 110-mm/s table speed per gantry rotation. Automated tube current modulation in the z-axis (AutomA; GE Healthcare) with a preset noise index (standard deviation [SD] of the regional CT number) of 10 HU was used. This noise index was relative to the reconstructed FBP images with a slice thickness of 5 mm and standard reconstruction kernel. The average CTDIvol was 10.3 ± 3.0 (mGy). All patients received non-ionic iodinated contrast material (Iopamiron; Bayer HealthCare, Osaka, Japan) at a concentration of 600 mg I per kilogram of body weight followed by 30 mL saline via a commercially available power injector (Dual Shot-Type GX; Nemotokyorindo, Tokyo, Japan) with a fixed injection duration of 30 s. Each patient had a 20-gage plastic intravenous catheter placed in an upper extremity vein (typically an antecubital vein). In all patients, the late arterial and portal venous phase images were obtained. The dual-phase scans were initiated 40 and 70 s after the start of the contrast media injection. Pre-contrast images were also obtained in all patients; however, pre-contrast images and portal venous phase images were not evaluated in this study for comparing different reconstruction algorithms. In order to perform focused assessments of objective image noise and overall subjective image quality (as opposed to lesion identification), we analyzed only the late arterial phase images, according to the method used in previous studies [21, 22].

Image reconstruction FBP reconstructions with 5-mm slice thickness and standard reconstruction kernel were obtained for routine clinical evaluation. In addition, images with 1.25-mm slice thickness were obtained from FBP (group 1), ASiR (group 2), and MBIR (group 3) reconstructions using the

X.-Z. Lin et al.: CT of the pancreas

late arterial phase CT scan data. ASiR image reconstruction was selected for 40% of blending [15, 23–25]. A total of 120 image sets (3 image datasets in each patient) were obtained. For MBIR reconstructions, raw data were transferred to an independent server. The reconstructed images were sent back to the scanner and independent workstation (AW4.5 Advantage Workstation; GE Healthcare, Milwaukee, WI) for image view, reformat, and measurement. Datasets were reviewed on a commercially available workstation AW4.5 Advantage Workstation. One radiology technician (I.T., 15-year experience in CT imaging and three-dimensional reconstruction) generated multiplanar reformation (MPR) images for the PD at AW4.5 workstation. Averaged MPR images of the late arterial phase were reconstructed in the oblique coronal and axial plane with a 1.25-mm thickness for the main PD in the pancreatic head and body section (two images for each case, total 240 images for 40 patients). We used the same parameters in all patients. Each image dataset was coded, patient information was removed, and the sets were randomized by a study coauthor (R.F.) to enable doubleblinded evaluation. The routine 5-mm-thick diagnostic images with FBP reconstruction algorithm were not assessed for algorithm comparison purpose in our study.

Qualitative image assessment of PD depiction and image quality The 1.25-mm-thick transverse and MPR CT images from the three reconstruction algorithms were used for assessing PD depiction and image quality. Two radiologists (X.Z.L., H.M.; 11 and 15 year-experience in body CT diagnosis, respectively) independently reviewed three CT reconstruction groups of the late arterial phase MPR images randomly on the advantage workstation. The readers were blinded to the reconstruction techniques. The radiologists used late arterial phase MPR images to evaluate the degrees of depiction of the PD. Before starting the assessment, both radiologists were given the criteria for image grading, and trained on five image datasets for the grading of different aspects of subjective image quality and PD depiction assessment so that they would understand the evaluation system, as well as to improve interobserver agreement. These five image datasets belonged to the first five patients recruited in our study and were not used for the subsequent statistical analysis. Both radiologists assessed five image datasets just before the evaluation of the rest of the CT studies that were used in the statistical analysis. The depiction of the PD including conspicuity of the wall and the clarity of the inner lumen of the PD, image noise, and overall image quality was all evaluated with a single grade on a 5-point scale ranging from 1 (worst) to 5 (best) (grade 1, non-diagnostic, almost not visible; grade 2, poor; grade 3, acceptable; grade 4, good; grade 5, perfect). The readers were asked to give the grade of

conspicuity of the PD. They also rated the clarity of the inside of PD, image noise, and overall image quality for the entire CT dataset. Overall image quality was subjectively assessed for noise, sharpness, and contrast. Images were initially presented with a preset abdominal window (window width, 260 HU; window level, 60 HU), the readers were allowed to modify the window width and level at their own discretion.

Quantitative image analysis Objective image noise (i.e., SD) and CT numbers (in Hounsfield units, HU) were measured for the 240 CT image series. Mean CT numbers and SD of the pancreatic parenchyma and PD were measured on the late arterial phase images by the same trained technician (I.T., 15-years experience of CT imaging and threedimensional reconstruction experience) at the independent workstation by using a circular region of interest (ROI) cursor. For the three reconstruction groups, we use copy and paste method to keep the consistency for ROI. The attenuation value of the pancreatic parenchyma adjacent to the ducts was measured. Blood vessels, calcifications, and artifacts were carefully excluded from all measurement areas. To avoid inaccuracies in a single measurement, two anatomic areas (pancreatic head and body) were measured for each patient. To ensure homogeneity, all measurements were performed three times, and mean values were calculated. The same measurement process was conducted for the 40 patients. Image noise was defined as the SD of the attenuation value measured in the pancreatic parenchyma. The ductto-pancreas contrast-to-noise ratio (CNR) was calculated as CNR = (ROI_p - ROI_d)/ N, where ROI_p is the mean attenuation value of the pancreatic parenchyma, ROI_d is the mean attenuation value of the PD, and N is image noise.

Statistical analysis Statistical analyses were performed by using statistical software (SPSS for Windows, version 13.0; SPSS Chicago, IL). Results were expressed as mean ± SD for continuous variables, frequencies, and percentages for categorical variables. Intraobserver variability was not estimated, as each radiologist assessed the images only once. Interobserver variability was estimated by using both kappa statistics and percentage agreement between the two radiologists for each of the assessed subjective image quality and PD depiction parameters. A kappa value of less than 0.20 was considered to indicate slight agreement; 0.21–0.40, fair agreement; 0.41–0.60, moderate agreement; 0.61–0.80, substantial agreement; and 0.81 or greater, almost perfect agreement [26]. The Friedman test was used to evaluate intergroup for differences in qualitative degree of depiction of the

X.-Z. Lin et al.: CT of the pancreas

main PD, image noise, and image quality. The post-hoc multiple comparisons with Bonferroni test following oneway ANOVA were used to evaluate the difference for the objective measurements of CT numbers of the pancreatic parenchyma, SD of CT numbers of the pancreatic parenchyma and the PD, and CNR of the PD. The P values of less than 0.05 were considered to indicate a significant difference.

Results Qualitative image analysis The percentage agreement between the two radiologists ranged from 53.8% to 83.8% with a fair to moderate agreement (kappa = 0.22–0.59), as summarized in Table 1. The distribution of subjective scores for pancreatic depiction and image quality is summarized in Tables 2. Image quality as assessed by the five-point scale showed a significant difference between groups, with the highest values noted for reconstructions using MBIR. In comparison with FBP, both 40% ASiR and MBIR were associated with a statistically significant increase in relative image quality as defined by the scale (P < 0.001 for each) (Tables 2, 3). The degree of depiction for the main PD on MPR images for the MBIR group ranked higher than the ASiR group, and the ASiR group higher than the FBP group (Fig. 1). Subjective image noise rated by the two independent readers to be the worst for the FBP group compared with the ASiR group and MBIR group. Overall image quality was classified to be good or perfect in 14 (8.8%) and 0/160 images (0.0%) using FBP, 83 (51.9%) and 0/160 images (0%) using ASiR, and 82 (51.3%) and 69/160 (43.1%) using MBIR (Fig. 2). The overall image quality was significantly improved on ASiR images compared with FBP images (P < 0.001) and on MBIR images compared with ASiR images (P < 0.001) (Tables 2, 3).

Quantitative image analysis Detailed objective image quality values are summarized in Table 4. No significant intergroup difference was

Table 2. Subjective image assessment (median/modal value) of pancreatic CT images with FBP, ASiR, and MBIR techniques derived from two readers

FBP Viewer Viewer 40%ASiR Viewer Viewer MBIR Viewer Viewer

Wall of PD

Inside of PD

Noise

IQ

1 2

3/3 3/3

3/3 3/3

3/3 3/3

3/3 3/3

1 2

3/3 4/4

3/3 3/3

4/4 4/4

3.5/4 4/4

1 2

4/4 4/4

4/4 4/4

5/5 5/5

4/4 4/4

PD, pancreatic duct; IQ, image quality; FBP, filtered back projection; ASiR, adaptive statistical iterative reconstruction; MBIR, model-based iterative reconstruction Table 3. Comparison of subjective image assessments for pancreatic CT among FBP, 40%ASiR, and MBIR techniques from two independent readers

Wall of PD (P) Inside of PD (P) Noise (P) IQ (P)

Viewer 1

Viewer 2