Ultra-low dose abdominal MDCT: Using a knowledge-based Iterative Model Reconstruction technique for substantial dose reduction in a prospective clinical study

European Journal of Radiology 84 (2015) 2–10

Contents lists available at ScienceDirect

European Journal of Radiology journal homepage: www.elsevier.com/locate/ejrad

Ultra-low dose abdominal MDCT: Using a knowledge-based Iterative Model Reconstruction technique for substantial dose reduction in a prospective clinical study Ranish Deedar Ali Khawaja a,∗ , Sarabjeet Singh a , Michael Blake a , Mukesh Harisinghani a , Gary Choy a , Ali Karosmangulu a , Atul Padole a , Synho Do a , Kevin Brown b , Richard Thompson b , Thomas Morton b , Nilgoun Raihani b , Thomas Koehler c , Mannudeep K. Kalra a a b c

MGH Imaging, Massachusetts General Hospital and Harvard Medical School, Boston, MA, USA CT Research and Advanced Development, Philips Healthcare, Cleveland, OH, USA Philips Technologie GmbH, Innovative Technologies, Hamburg, Germany

a r t i c l e

i n f o

Article history: Received 1 July 2014 Received in revised form 8 September 2014 Accepted 29 September 2014 Keywords: Radiation dose reduction Abdominal MDCT Iterative reconstruction technique

a b s t r a c t Purpose: To assess lesion detection and image quality parameters of a knowledge-based Iterative Model Reconstruction (IMR) in reduced dose (RD) abdominal CT examinations. Materials and methods: This IRB-approved prospective study included 82 abdominal CT examinations performed for 41 consecutive patients (mean age, 62 ± 12 years; F:M 28:13) who underwent a RD CT (SSDE, 1.5 mGy ± 0.4 [∼0.9 mSv] at 120 kV with 17–20 mAs/slice) immediately after their standard dose (SD) CT exam (10 mGy ± 3 [∼6 mSv] at 120 kV with automatic exposure control) on 256 MDCT (iCT, Philips Healthcare). SD data were reconstructed using filtered back projection (FBP). RD data were reconstructed with FBP and IMR. Four radiologists used a five-point scale (1 = image quality better than SD CT to 5 = image quality unacceptable) to assess both subjective image quality and artifacts. Lesions were first detected on RD FBP images. RD IMR and RD FBP images were then compared side-by-side to SD-FBP images in an independent, randomized and blinded fashion. Friedman’s test and intraclass correlation coefficient were used for data analysis. Objective measurements included image noise and attenuation as well as noise spectral density (NSD) curves to assess the noise in frequency domain were obtained. In addition, a low-contrast phantom study was performed. Results: All true lesions (ranging from 32 to 55) on SD FBP images were detected on RD IMR images across all patients. RD FBP images were unacceptable for subjective image quality. Subjective ratings showed acceptable image quality for IMR for organ margins, soft-tissue structures, and retroperitoneal lymphadenopathy, compared to RD FBP in patients with a BMI ≤25 kg/m2 (median-range, 2–3). Irrespective of patient BMI, subjective ratings for hepatic/renal cysts, stones and colonic diverticula were significantly better with RD IMR images (P < 0.01). Objective image noise for RD FBP was 57–66% higher, and for RD IMR was 8–56% lower than that for SD-FBP (P < 0.01). NSD showed significantly lower noise in the frequency domain with IMR in all patients compared to FBP. Conclusion: IMR considerably improved both objective and subjective image quality parameters of RD abdominal CT images compared to FBP in patients with BMI less than or equal to 25 kg/m2 . © 2014 Elsevier Ireland Ltd. All rights reserved.

1. Introduction

Abbreviations: SD CT, standard dose CT; RD CT, reduced dose CT; IRT, iterative reconstruction technique; IMR, Iterative Model Reconstruction; FBP, filtered back projection; SSDE, size-specific dose estimate. ∗ Corresponding author at: 25 New Chardon Street, 4th Floor, Boston, MA 02114, USA. Tel.: +1 857 204 7450; fax: +1 617 643 0111. E-mail address: [email protected] (R.D.A. Khawaja). http://dx.doi.org/10.1016/j.ejrad.2014.09.022 0720-048X/© 2014 Elsevier Ireland Ltd. All rights reserved.

CT radiation is an important public safety concern [1–3]. Efforts have been made to decrease the necessary radiation dose with CT scanning. Iterative reconstruction techniques (IRT) have enabled dose reduction by reducing image noise while preserving image quality compared to traditional filtered back projection (FBP) based image reconstruction [3–7]. A goal of sub-millisievert radiation

R.D.A. Khawaja et al. / European Journal of Radiology 84 (2015) 2–10

3

Table 1 Baseline patient demographics. Parameters

Females (n = 28)a

Males (n = 13)a

P value

Mean age (y) Mean body weight (kg) Mean body mass index (kg/m2 ) Mean effective diameter (cm) Mean CTDIvol (mGy) Mean SSDE (mGy)

62 ± 13 73 ± 14 27 ± 5 30 ± 6 9±4 10 ± 4

60 ± 11 83 ± 18 28 ± 4 29 ± 6 8±3 11 ± 4

0.6 0.06 0.7 0.5 0.7 0.7

Indication for abdominal CT Hematuria Abdominal cancer (primary or metastatic) Abdominal pain Abdominal mass

4 (14) 10 (36) 12 (43) 2 (7)

4 (31) 6 (46) 3 (23) –

a

– – – –

Except where indicated, numbers in parentheses are percentages.

dose for CT has been advocated by several researchers as this would be below the average annual dose from background radiation [8–14]. Additionally, the desire to stay below this specific target dose comes from a special report from the Summit on Management of Radiation Dose in Computed Tomography in collaboration with National Institutes of Health [15,16]. Several different types of IRTs have recently become available to enable CT dose reduction. IRTs from major vendors such as Adaptive Iterative Dose Reduction 3D (AIDR 3D, Toshiba Healthcare), Adaptive Statistical Iterative Reconstruction (ASIR, GE Healthcare), iDose (Philips Healthcare), Sinogram-Affirmed Iterative Reconstruction (SAFIRE, Siemens Healthcare), and Model-Based Iterative Reconstruction (MBIR, GE Healthcare) have shown dose reduction of up to 66% [17–23]. They have been accepted as standard reconstruction algorithms in clinical practice for low dose protocols [17–23]. Iterative Model Reconstruction (IMR, Philips Healthcare) technique, a knowledge- and model-based iterative reconstruction algorithm, has been recently approved by the Food and Drug Agency (FDA). Role of IMR has been assessed for dose reduction in cardiac and chest CT [25–28]. To the best of our knowledge, there are no published clinical studies for IMR technique for dose reduction in abdominal CT at sub-millisievert radiation dose. Hence, the purpose of our study was to assess lesion detection and image quality parameters of a knowledge-based IMR in reduced dose (RD) abdominal CT examinations.

2.3. Image acquisition

2. Materials and methods

2.4. Estimation of radiation dose

2.1. Study design

Effective radiation doses were estimated by multiplying doselength product by a conversion factor of 0.015 mGy/mGy cm according to standard methodology. The volumetric CT dose index (CTDIvol ) was recorded for each study as reported by the scanner on

This prospective clinical study was conducted in compliance with Human Insurance Portability and Accountability Act (HIPAA) guidelines. The institutional review board approved the study, and all patients provided written informed consent.

2.2. Study cohort Inclusion criteria for this study consisted of adult patients (19 years or older in age). These patients were clinically referred for a routine outpatient abdominal CT examination. Other inclusion criteria included patients who were able to provide written informed consent, able to hold breath for duration of at least 10 s, able to follow verbal commands for breath-holding and remain still for the scanning duration, and hemodynamically stable. Patients were exluded if they were pregnant, or has a body-mass-index of 33 kg/m2 and above. All patients were consecutive and prospectively enrolled regardless of sex, race or clinical indication of abdominal CT. Baseline patient demographics are given in Table 1.

All patients underwent a standard dose (SD) abdominal CT examination on a 256-slice MDCT (Philips Healthcare, Andover, Massachusetts) with (n, 37) or without (n, 4) intravenous contrast. No additional contrast was injected for the reduced dose (RD) series. After centering the patient in the gantry isocenter, two orthogonal localizer radiographs were acquired to plan the SD CT as a usual CT protocol. SD CT was planned on the localizer radiograph extending from the dome of the diaphragm to the pubic symphysis. Next, the SD CT series was duplicated to plan the RD CT image series over the exact same scan region and range. Hence, a total of 82 abdominal CT series were performed for 41 patients. All scan parameters, with exception of tube current, were kept constant between the two image series. For the RD CT image series, we used a low fixed tube current of 17–20 mAs/slice (mA × gantry rotation time/pitch) in order to obtain a targeted dose-length product of less than or equal to 65 mGy cm which corresponds to an estimated effective dose of just under 1 mSv (65 mGy cm × 0.015 = 0.98 mSv). Scan parameters for both SD and RD CT image series are given in Table 2. SD and RD CT image series were not performed in a single breath hold. Patient’s standard of care SD CT was always performed first (with or without contrast, based on the ordered protocol). RD CT was performed immediately after SD CT. There was a delay of an average 5 s between the acquisition of SD and RD CT image series.

Table 2 Abdominal CT parameters. Parameter

Standard dose CTa

Reduced CT

Tube potential (kV) Tube current (mA) Tube current modulation Pitch Collimation

120 119 ± 48 Automatic exposure control 0.985 40

120 17 ± 1.2 Fixed tube current 0.985 40

a Standard dose CT is performed with automatic exposure control (AEC) at our institution. However, in order to reach SubmSv radiation dose (equivalent to a DLP less than 65 mGy × cm) for research scans, AEC was not helpful since the final dose cannot be specified upfront. Hence, we used fixed-tube current for SubmSv CT. The remaining scan parameters were kept constant between the two CT exams (helical acquisition mode, 128 mm × 0.625 mm detector configuration, 0.5 s gantry rotation time, slice thickness 5 mm, increment 2.5 mm, reconstruction filter A). The time period between completion of the standard-of-care abdominal CT and acquiring SubmSv research images was 10 s or less.

4

R.D.A. Khawaja et al. / European Journal of Radiology 84 (2015) 2–10

throughout the entire study performed on a DICOM image work station (ClearCanvas, Toronto). The pattern of noise spectral density (NSD) was assessed in MATLAB program for SD FBP, RD FBP and RD IMR settings and was reported graphically. 2.7. Phantom study

Fig. 1. Flowchart shows CT image acquisitions at standard and SubmSv radiation doses with two different image reconstruction algorithms (filtered back projection, FBP; standard dose, SD; SubmSv, sub-millisievert).

a 32-cm body phantom. Size-specific dose estimates (SSDE) were obtained from the individual effective patient diameter (equal to √ AP × LAT diameter) as measured from the scout images by the co-author (RDAK, two years of research experience). 2.5. Image reconstruction Raw data for all patients undergoing abdominal CT examinations were reconstructed on an offline reconstruction workstation with FBP and IMR. Please see Appendix 1 for description of IMR technique and its different settings. Hence, raw data of SD CT were reconstructed with FBP (SD FBP). Raw data of RD CT were reconstructed with IMR (three settings, IMR1 , IMR2 , IMR3 , body-soft level 2, body-routine level 2, and bodysharpPlus level 2, respectively). Therefore, for each patient there were five image series (one SD FBP; one RD FBP and three RD IMR series). For all study patients there were 205 image series (41 patients × 5 series). Before the formal image quality session, data for two patients were used for training the radiologists. Hence, final image quality evaluation was performed for 39 patients (39 × 5 = 195 CT series). Fig. 1 presents the study flow chart. 2.6. Image quality Four experienced abdominal radiologists (M.H. with 15 years, M.B. with 12 years, G.C. with 5 years and A.K. with 4 years of experience) evaluated subjective image quality independently and in a blinded fashion using a five-point scale (score 1, image quality better than standard dose CT; score 2, image quality equal to the standard dose CT and sufficient for clinical diagnostic performance; score 3, image quality is limited but sufficient for clinical diagnostic performance; score 4, image quality mostly below the clinical need and suboptimal clinical diagnostic performance; and score 5, unacceptable clinical diagnostic performance). All the assessed structures are stated in Appendix 2. The reported image quality score is the median and interquartile ranges of scores across four readers. Circular regions of interest (ROI, 20–30 mm) were drawn in the homogenous liver parenchyma, anterior abdominal fat, and abdominal aorta at the level of porta hepatis to cover at least two third of its lumen. Mean attenuation values (HU) and image noise (standard deviation, ) were measured for each ROI in all 195 CT series. Signal-to-noise ratios (SNR) were calculated using the formula (SNR = mean HU/BN), where BN is the background noise. The size and position of each ROI were kept constant

Quantitative measurements of low contrast detectability (LCD) were performed using a human observer study and images from phantom scans. Scans were taken of the MITA CT Performance Phantom (Phantom Laboratory, CCT183), which consists of a 20-cm phantom consisting of a CATPHAN-like shell and a background plastic with CT number of approximately 45 HU at 120 kVp. Phantom contained four 20 mm long low-contrast pins with different diameters and contrasts (14 HU contrast with 3 mm diameter, 7 HU contrast with 5 mm diameter, 5 HU contrast with 7 mm diameter, and 5 HU contrast with 10 mm diameter, where HU refers to Hounsfield Units or CT numbers). Phantom was scanned at 120 kVp using axial mode. 100 scans were performed at each dose listed in Appendix 3, and image ROIs were assembled to perform a 4-Alternative Forced Choice (4-AFC) test as described by Burgess [29]. The task measured was a signal-known-exactly, location-known-exactly task. In such a test an observer is presented with four images, only one of which contains the object to be detected and the other three contain only noise. The observer chooses the image which he or she judges to most likely contain the object, and a score of correct or incorrect is recorded for this trial. Each observer performed 100 independent trials, and the portion correct (Pc) measurement was calculated. Pc was converted to a detectability index (d ) (following again the method in Burgess et al. [29]), and the median d for all observers (n = 36, all imaging scientists and engineers who did not require any special training for detection) is recorded in the Appendix 3 for the different test conditions. A d = 0 means that the observer does no better than random guessing in detecting the object, and a d = 2 means the observer makes the correct choice 82.5% of the time. 2.8. Reference standard for true lesion detection For presence or absence of lesions within organs, SD FBP images were considered as the “reference standard” where all detected lesions were “true lesions”. Any lesions seen in the RD images but not in SD FBP were considered as “false positive or pseudo lesions”. Any lesions that were not seen in RD images but seen in SD FBP images were considered as “missed lesions”. First the radiologists were asked to perform the task of lesion detection on RD FBP images alone. Next, all six-image series (SD FBP, RD FBP, RD IMR1 , RD IMR2 , and RD IMR3 ) were displayed on a DICOM compliant workstation (LED 2-megapixel large display screen, NED). Radiologists were blinded to the arrangement of RD image series on the screen. However, they were made aware of the SD FBP images that were always located on the top right of the screen. Lesions were assessed in all RD images and directly compared to reference standard “SD FBP” for any missed and/or false positive lesions. All true lesions were then compared across RD images with SD FBP on the aforementioned grading scale. For each patient, there was no time delay between the visualization of different series of images. Once the lesion detection was performed, the radiologist was asked to perform a comprehensive subjective image quality assessment. All radiologists performed this evaluation independently and in a blinded fashion. That is no consensus session was held and all reduced dose image series were blinded to the radiologist. It was not shown which image series belonged to IMR or FBP technique. In addition to avoid any bias, all reduced dose IMR and FBP images were constantly randomized for each subsequent patient so as to

R.D.A. Khawaja et al. / European Journal of Radiology 84 (2015) 2–10

5

Table 4 Lesion detection by radiologists.

Table 3 Summary of radiation dose data. Parameters

Standard dose CT

Mean SSDE (mGy) Mean CTDIvol (mGy) Mean dose-length product (mGy × cm) Mean estimated effective dose (mSv)

11.0 9.0 434.0 6.0

± ± ± ±

3.5 3.5 195.0 2.9

Reduced dose CT 1.5 1.3 61.0 0.98

± ± ± ±

0.4 0.1 2.4 0.0

avoid any pre-formed assessment that could have happened in the absence of randomization.

Organs

Lesions

Number of lesions per radiologist

• Liver

Lowattenuation lesions Nodule Stones Mass Lowattenuated lesions Adenopathy Colonic diverticula – –

(12, 14, 14, 12)

• Adrenal glands • Gall bladder • Pancreas • Kideny

• Retroperitoneum • Bowel

2.9. Statistical analysis All statistical analysis was performed on SPSS software (version 21.0, SPSS Chicago, IL) and spreadsheet software (Microsoft Excel 2010; Microsoft, Richmond, VA). We performed one-way ANOVA to evaluate the differences in mean patient age, mean weight, mean BMI, mean effective diameter and mean CT dose metrics and objective image noise. Differences between subjective image quality for image reconstruction algorithms (RD FBP, and RD IMR1 to IMR3 ) were assessed with the Friedman’s test (non-parametric for repeated measures of ANOVA). Post hoc analysis was performed with Dunn’s multiple comparison test. Intraclass correlations (ICC) in addition to 95% confidence intervals (CI) were used to determine interobserver agreement whose interpretation was based as follows, 0 = no agreement; 1 = full agreement; 0 to 1 = agreement declines as correlation moves toward 0. A P value of 0.05 or less was considered statistically significant. 3. Results 3.1. Radiation doses The mean (± standard deviation) radiation doses for SD and RD abdominal CT are summarized in Table 3. Compared to SD CT, mean dose reduction was 85% (SSDE 1.3/9.0 mGy).

Total lesions (n, SubmSv FBP) Total lesions (n, SD FBP) Total lesions (%)

(2, 2, 3, 3) (3, 5, 4, 1) (2, 0, 1, 1) (15, 13, 15, 09)

(3, 3, 4, 2) (14, 08, 14, 04) (51, 45, 55, 32) (51, 45, 55, 32) (100%, 100%, 100%, 100%)

3.3. Subjective image evaluation based on organ systems A separate pooled analysis for visualization of hepatobiliary system (including liver, gall bladder and low attenuation focal hepatic lesions) showed acceptable image quality in 31/39 patients with RD IMR1 images compared to only one patient with FBP images (P < 0.01). Eight patients who had unacceptable image quality for liver parenchyma had a BMI between 31 and 32 kg/m2 . Similarly, for visualization of urinary system (including renal margins, renal pelvis, stones, and low attenuated focal renal lesions) showed acceptable image quality in 38/39 patients with RD IMR1 images compared 11/39 patients with FBP images (P < 0.01). The only patient with suboptimal visualization with IMR had both a BMI of 32 kg/m2 and a non-contrast CT exam. Visualization of gastrointestinal system (including bowel wall and abnormalities) was optimally seen in 37/39 patients with RD IMR1 images compared 17/39 patients with FBP images (P < 0.01). Two patients with suboptimal appearance with IMR images had BMI above 30 kg/m2 and a non-contrast study.

3.2. Lesion detection (true, pseudo-, and missed lesions) 3.4. Subjective image quality based on organs All true lesions (ranging from 32 to 55 for different radiologists; seen in SD FBP images) were detected in RD image series using pooled analysis (Table 4). A variety of lesions were seen in our study (Table 4 lists all the detected lesions), which covers the breadth of diagnoses routinely seen. No pesudo-lesions were seen in RD image series. There was a significant difference in diagnostic confidence between their visualization across RD FBP and RD IMR images (P < 0.01).

Subjective image quality grading has been summarized in Table 5. There was a substantial inter-observer agreement between the four radiologists (ICC, 0.69; P < 0.001). All the following organs were assessed on RD IMR images and compared with both RD FBP and SD FBP for image quality. Visibility of liver margins was sub-optimally visualized in 38/39 patients on RD FBP images and ranged from acceptable (35/39) to

Table 5 Subjective image quality.a Structures

RD FBP

RD IMR1

RD IMR2

RD IMR3

• Liver margins • Liver parenchyma • Liver lesions • Gall bladder wall • Pancreatic contours • Adrenal contours • Renal margins • Renal pelvis • Renal cysts • Peritoneum • Retroperitoneal lymphadenopathy • Bowel wall • Colonic diverticula • Urinary bladder wall

4 (3–5) 4.5 (3–5) 4 (3–5) 4 (3–5) 4 (3–5) 4 (3–5) 4 (3–5) 4 (2–5) 4 (3–5) 4 (4–5) 4 4 (2–5) 4 (3–5) 4 (3–5)

3 (2–4) 3 (2–4) 3 (2–5) 3 (2–5) 3 (2–4) 3 (2–5) 3 (2–4) 3 (2–5) 3 (2–5) 3 (2–5) 2 (2–3) 3 (2–4) 3 (2–3) 3 (2–5)

3 (2–4) 3 (3–4) 3 (2–5) 3 (2–5) 3 (2–4) 3 (2–5) 3 (2–4) 3 (2–5) 3 (2–5) 3 (2–5) 2 (2–3) 3 (2–4) 3 (2–3) 3 (2–5)

3 (2–4) 3 (3–5) 3 (2–5) 4 (2–5) 3 (2–5) 3 (2–5) 3 (2–4) 3 (2–5) 3 (32–5) 3 (2–5) 2 (2–3) 3 (2–4) 3 (2–3) 3 (2–5)

a

Numbers represent median scores (inter-quartile range) for quality and RD = reduced dose.

6

R.D.A. Khawaja et al. / European Journal of Radiology 84 (2015) 2–10

Fig. 2. Transverse abdominal CT (CTDIvol , 6 mGy; SubmSv scan, 1.2 mGy) images of a 64 year-old woman (BMI, 29 kg/m2 ) with known Hodgkin disease show a low attenuation liver lesion. Both liver parenchyma and lesion were acceptable for SubmSv IMR1–2 images and unacceptable with SubmSv-FBP (too noisy) and SubmSv-IMR3 images.

suboptimal (4/39) across the three RD IMR series. Liver parenchyma was suboptimal for evaluation in 32/39 patients on RD FBP images and acceptable in RD IMR (23–26/39) patients. Among RD IMR images, RD IMR1 images were significantly better than RD IMR3 images for both liver margins and parenchyma (P < 0.01, Fig. 2). Compared to RD FBP, liver lesions were seen significantly better with RD IMR1 images (P < 0.01, Fig. 3). RD IMR3 images were similar to RD FBP images (P > 0.05). The visibility of gall bladder, adrenal glands, pancreatic contours, peritoneum (including retroperitoneum lymph nodes, kidneys, bowel wall and urinary bladder) were significantly better on RD IMR1 images compared to RD FBP (P < 0.01, Table 5). Compared to RD FBP, renal cysts were seen significantly better with RD IMR1–3 (P < 0.001, Fig. 4). Additionally, colonic diverticula were seen significantly better with RD IMR1–3 images (P < 0.001). Ring artifacts (related to photon starvation) in pelvis were seen in 12/39 patients at RD CT images (≤25 kg/m2 , 0/10 patients; 25.1–29.9 kg/m2 , 3/11 and ≥30 kg/m2 , 9/18; Fig. 5). There was no significant difference in the extent of artifacts on RD IMR and RD FBP images (P > 0.05). 3.5. Objective image quality The mean CT number (Hounsfield Units, HU) values and image noise (standard deviations of attenuation values) are

summarized in Table 6. Overall, mean objective noise was lower in RD IMR1 setting compared to other IR algorithms (11.5 ± 3.5 HU, P < .001). Objective noise was 64–84% (P < 0.001) lower for RD IMR compared to RD FBP (71 ± 26 HU). Compared to SD FBP (26 ± 7 HU), there was 8–56% (P < .001) noise reduction in IMR at 85% dose reduction in abdominal CT. There was no difference in the SNR for patients with acceptable and unacceptable visualization at RD dose (P > 0.05, Table 6).

3.6. Noise spectral density trends and phantom study NSD plots were only measured in two patients that represented a spectrum of patient size (BMI, one with 25 kg/m2 and the other with 30 kg/m2 ) used in our study. NSD was performed to assess noise in frequency domain have been plotted in Fig. 5. RD IMR images showed significantly lower noise in the frequency domain compared to RD FBP and SD FBP images. The results of LCD phantom study showed that at the same dose of 4 mGy, with IMR the detectability index for a low contrast pin is more than 250% better than FBP. In addition, with IMR the dose can be reduced by 80% compared to FBP, and the detectability is not only maintained, but actually improves by 80% over FBP at the higher dose.

R.D.A. Khawaja et al. / European Journal of Radiology 84 (2015) 2–10

7

Fig. 3. Transverse abdominal CT (CTDIvol , 6 mGy; SubmSv scan, 1.2 mGy) images of a 78 year-old man (BMI, 21 kg/m2 ) with known primary hepatocellular carcinoma show a low attenuation renal lesion in right kidney. Both renal margins and lesion were acceptable for all SubmSv IMR1–2 images and unacceptable with SubmSv-FBP (too noisy) and SubmSv-IMR3 images.

4. Discussion Preliminary results of this ongoing prospective study suggest comparable lesion detection for RD and SD CT images in all patients irrespective of BMI. While RD FBP images were unacceptable for visualization of structures, RD IMR images enabled acceptable visualization in smaller sized patients (up to BMI of 25 kg/m2 ). However, when it comes to evaluation of hepatic, renal, and intestinal lesions in terms of lesion margins and conspicuity, IMR outperformed RD FBP images in all patients. Both image noise measurements and

noise spectral density plots showed significant noise reduction with IMR technique. We believe that abdominal and pelvic body regions are the most challenging regions in the body for substantial dose reduction to less than 1 mSv due to presence of low contrast abnormalities and normal structures. To the best of our knowledge, this study provides initial results for the performance of IMR technique at a sub-millisievert radiation dose in abdominal MDCT. We believe that sub-millisievert abdominal CT can be limited to certain clinical indications such as renal stone detection, inflammatory bowel disease, appendicitis, Crohn’s disease, diverticulitis

Fig. 4. Demonstration of artifacts in a patient (BMI, 32 kg/m2 ): compared to standard-of-care abdominal CT (CTDIvol , 14 mGy), pelvic anatomy was unacceptable due to ring artifacts (arrow) across all SubmSv images (1.2 mGy).

8

R.D.A. Khawaja et al. / European Journal of Radiology 84 (2015) 2–10

Fig. 5. Noise spectral density plots for two patients with different sizes (A, 68.8 kg; B, 113.3 kg). All IMR settings (especially IMR1 body soft tissue) showed the highest noise reduction frequency.

etc. However, for the evaluation of liver tumors and pathologies, this reduced dose will not be suitable unless a very small sized patient. Prior studies have assessed the role of IRT in abdominal MDCT at higher radiation doses (CTDIvol , 4.2–17 mGy) compared to our study (CTDIvol , 1.2 mGy) [17–23,31,32]. Deak et al. studied the role of MBIR and ASIR in abdominal CT performed at 2–5-folds higher radiation dose (9 ± 4 mGy) with improved image quality and decreased image noise with MBIR images [28]. At 4.2 mGy radiation dose, BLINDED et al. reported better diagnostic confidence for abdominal CT using ASIR technique [22]. At lower dose of 2.5 mGy (about 50% higher dose than our study), Kataria et al. showed improved subjective image quality using Safire technique [31]. In contrast to above studies exclusive to abdominal MDCT, feasibility of sub-milliseivert radiation doses has been assessed in lumbar and whole spine CT, chest CT, coronary CT angiography, and inflammatory bowel disease [8–14]. O’Neill et al. reported acceptable quality in abdominal CT for evaluating Crohn’s disease at sub-milliseivert radiation dose in patients with low to normal BMI (<25 kg/m2 ) [11].

This is in line with our results where this dose level was acceptable for smaller patients (≤25 kg/m2 ) for visualization of bowel wall and lesions. A review of recent literature showed studies that have been performed using IMR technique [24–28]. In cardiac CT, up to 80% dose reduction can be achieved using IMR [14,24–28]. In chest CT, IMR has been shown to improve delineation of lesion margins as well as diagnostic confidence compared to standard dose FBP images [27]. In abdominal CT, two studies have been reported. Y. Funama et al. studied IMR technique in non-enhanced porcine livers that were scanned multiple times. They investigated noise and accuracy on images acquired with IMR and FBP, and reported no loss of image accuracy in IMR images. They, however, did not investigate dose reduction potential with IMR [27]. Our study differs in two aspects: first our study was performed on patients in a clinical setting. Secondly we demonstrate significant dose reduction in abdominal MDCT for small sized patients. S. Suzuki et al. also presented initial performance evaluation for IMR in abdominal CT [28]. In their patient study, significantly improved image noise

Table 6 Objective image quality. Structure

SD FBPa

RD FBP

RD IMR1

RD IMR2

RD IMR3

101.9 ± 26.7 −92.7 ± 34.5 178.6 ± 68.7

109.1 ± 24.4 −92.0 ± 26.7 140.9 ± 39.7

101.6 ± 25.5 −90.5 ± 34.1 131.2 ± 39.6

100.1 ± 24.3 −91.7 ± 34.2 130.2 ± 40.1

100.3 ± 25.1 −92.2 ± 34.7 130.5 ± 39.8

*

Attenuation (HU) • Liver • Fat tissue • Aorta Noise (HU)** • Liver • Fat tissue • Aorta

26.2 ± 6.7 16.6 ± 3.4 28.9 ± 8.4

70.7 ± 26.1 43.1 ± 18.5 79.2 ± 28.2

12.6 ± 3.4 10.5 ± 4.9 14.9 ± 5.2

14.6 ± 3.5 13.2 ± 5.6 16.5 ± 4.9

16.4 ± 3.4 15.2 ± 3.5 18.5 ± 4.7

Signal-to-noise (SNR) ratio*** • Liver • Fat tissue • Aorta

0.10 ± 0.03 0.09 ± 0.03 0.18 ± 0.07

0.11 ± 0.02 0.09 ± 0.00 0.14 ± 0.04

0.10 ± 0.02 0.09 ± 0.03 0.13 ± 0.04

0.10 ± 0.02 0.09 ± 0.03 0.13 ± 0.04

0.10 ± 0.02 0.09 ± 0.03 0.13 ± 0.04

a * ** ***

SD = Standard dose; RD = reduced dose. Data are means ± standard deviation. No significant differences were found (P values of 0.99 for liver, 0.993 for fat, 0.983 for aorta; ANOVA test). Data are means ± standard deviation. Data were significantly different (P values of 0.001 for liver, fat, and aorta). SNR was measured as: SNR = mean HU/BN, where BN is the background noise.

R.D.A. Khawaja et al. / European Journal of Radiology 84 (2015) 2–10

and low-contrast resolution was reported with IMR. They reported maintained edge sharpness in abdominal CT images compared with iDose and FBP techniques. Our study results are consistent with them at sub-millisievert radiation dose in smaller patients only. As per current study, sub-millisievert dose is not feasible for patients with larger body habitués. Some of the artifacts observed in our study have been reported in prior studies as well. Particularly in larger patients, ring artifacts were a major limitation for acceptable image quality in pelvic cavity. MBIR technique has been shown to produce a pattern of staircase effect and blacked-out artifacts at a higher radiation dose (∼9 mGy) while Safire technique reported ring artifacts at 2.5 mGy radiation dose [31,32]. Altered image texture and the appearance of liver parenchyma have reported with other studies [22,30]. This was also seen in our study especially in larger patients as well as with IMR body sharp-plus setting. This altered image appearance did not compromise the detection of focal liver lesions in our study in smaller sized patients (BMI less than or equal to 25 kg/m2 ). RD IMR and RD FBP images were unacceptable for evaluation of liver lesions and parenchyma in larger patients with a BMI (25 kg/m2 or above). Only smaller patients with a BMI less than 25 kg/m2 had acceptable assessment for liver and lesions on RD IMR images (but not on RD FBP images). Additionally, the altered textural profile of IMR settings as seen in the noise spectral density plots may have resulted into the poor scores mostly for the evaluation of liver parenchyma in larger sized patients. It should be noted that the dose that is needed for a consistent image quality across different patients depend on their sizes. Hence, in this current study we made a choice of a fixed dose regardless of patient size – that may account for the appearance for ring artifacts in the pelvis for larger BMI group. Our study has two major implications. Our study suggests that reduced dose (under 1 mSv radiation dose) images reconstructed with IMR can provide equivalent overall image quality comparable to standard of care images in patients with a BMI less than or equal to 25 kg/m2 for only limited indications in abdominal MDCT. For evaluation of renal cysts, stones, and colonic diverticula IMR outperforms FBP in all patients irrespective of patient BMI. Secondly, we also report the differences in the performance of various settings for IMR at reduced dose scans. We found that IMR body soft tissue and IMR body routine settings performed substantially better than IMR sharp plus setting. From standpoint of image reconstruction times, IMR took less than 5 min on an offline reconstruction facility, which is considerably shorter on the CT user interface (approximately 3 min for majority of reference protocols). Our study has limitations. First, the sample size was low with the data from 41 patients only. Secondly, only one iterative reconstruction technique was studied which limits the application of this study. We wanted to perform pilot analysis using recently introduced knowledge-based IMR at sub-milliseivert radiation dose, assess and improve the technique before embarking on a comparative study. Thirdly, ring artifacts were visualized in the pelvis in RD CT images (particularly in larger patients). These rings are caused by an insufficient X-ray flux reaching the detector, which is an undesired side effect of our choice to operate with a fixed tube current (in order to ensure a final dose of 0.9 mSv) rather than with automatic exposure control. Also, texture analysis was not performed since we based our results on the subjective evaluation across four radiologists as well as objective image quality using noise measurements, noise spectral density plots and a phantom study. In addition, lesion detection was performed by all radiologists in independent sessions (without consensus). Hence, that led to a variable number of lesions detected by each radiologist that could be due to the variable experience as an abdominal radiologist (4–15

9

years across our study radiologists). Another limitation of our study was the selection of sub-milliseivert doses which may be too low for certain lesion types and particular body habitus. Additionally, although the scan parameters were exactly same for both standard of care and reduced dose CT series for each patient, fixed-tube current setting was only applied to reduced dose CT series across all study subjects. As mentioned above, we wanted to reach submillisievert radiation dose for all research scans. We believe that resulted into a disadvantage for larger sized patients. Finally, patients with a BMI of 33 kg/m2 or above were excluded since we did not want to subject them to unnecessary radiation dose without first assessing the feasibility of this dose level in smaller patients. In conclusion, preliminary results from this prospective study suggest that lesion detection is similar in standard dose and reduced dose (sub-milliseivert) abdominal CT. This finding was irrespective of patient body mass indices using a knowledge based Iterative Model Reconstruction (IMR) technique. IMR considerably improves both objective and subjective image quality parameters of reduced dose abdominal multidetector CT images compared to FBP, particularly in smaller patients (BMI less than or equal to 25 kg/m2 ). Additionally, assessment of urinary and gastrointestinal system can be performed at size-specific dose estimate of 1.5 mGy (∼0.9 mSv) using a knowledge-based IMR technique irrespective of patient BMI. Reduced dose abdominal MDCT at 0.9 mSv can only be applied in smaller patients and in limited CT indications using IMR technique.

Conflict of interests and financial disclosure Co-author Sarabjeet Singh M.D. received research funding from the Radiological Society of North America (RSNA), GE Healthcare, Philips Healthcare, and Toshiba America Systems. Coauthor Synho Do Ph.D. received research funding from Philips Healthcare. Coauthors Kevin Brown, Richard Thompson, Thomas Morton, Nilgoun Raihani and Thomas Koehler are employees of Philips Healthcare. The rest of the co-authors have no conflict of interests and financial disclosures.

Appendix 1. Iterative Model Reconstruction (IMR) • A model and knowledge-based iterative reconstruction technique. • IMR overcomes the “idealized” assumptions made in FBP reconstruction. These assumptions introduce deficiencies in the resultant FBP image such as artifacts and excessive noise especially in low dose examinations. • IMR iteratively minimizes a cost function that represents deviation of the volume from ideal data, which is determined through knowledge of the system models while simultaneously introducing a ‘roughness’ penalty. • It incorporates statistical and CT system models to improve the image quality (noise, contrast and contrast-to-noise ratio) of low dose CT. • IMR has three major groups of reconstruction settings for the body CT including (a) Body Soft, (b) Body Routine, and (c) Body SharpPlus. According to the vendor, Body Soft and Body Routine settings are optimal for soft tissue structures of the body. Body SharpPlus settings increase the sharpness of anatomic structures as well as the lesions in bones and lungs. Each setting has three levels of noise reduction (1–3), with 3 being the highest level and 1 being lowest level of noise reduction. • Users can chose appropriate IMR setting for image reconstruction based on noise reduction and spatial resolution requirements. • All IMR settings for body CT take same amount of reconstruction time (up to 5 min) on IMR prototype system as well as in clinical settings.

10

R.D.A. Khawaja et al. / European Journal of Radiology 84 (2015) 2–10

Appendix 2. Image Quality Evaluationa • Liver • Lesions in the liver • Adrenal Glands • Gall Bladder • Pancreas • Kideny • Retroperitoneum • Peritoneum • Urinary bladder • Pelvic lymph nodes • Bowel

Margins Number and Size Lesion Wall Contour and Pancreatic duct Renal pelvis and margins Sub-centimeter Mesenteric fat and vessels Wall Sub-centimeter Wall

Parenchyma Conspicuity Conspicuity Lesions Lesions Lesions Enlarged Lesions Lesions Enlarged Abnormality

a For soft-tissue evaluation, window-width (WW, 400) and window-level (WL, 40) was used.

Appendix 3. Low-contrast detectability phantom study Dose 10 mGy 4 mGy 2 mGy

d’ FBP (median) 0.821 0.554 N/A

d’ IMR (median) N/A 1.986 1.475

N/A = not applicable.

References [1] Brenner DJ, Hall EJ. Computed tomography – an increasing source of radiation exposure. N Engl J Med 2007;357(22):2277–84. [2] Brenner DJ, Elliston CD. Estimated radiation risks potentially associated with full-body CT screening. Radiology 2004;232(3):735–8. [3] Hansmann J, Schoenberg GM, Brix G, Henzler T, Meyer M, Attenberger UI, et al. CT of urolithiasis: comparison of image quality and diagnostic confidence using filtered back projection and iterative reconstruction techniques. Acad Radiol 2013;20(September (9)):1162–7. [4] Fleischmann D, Boas FE. Computed tomography—ideas and new technology. Eur Radiol 2011;21(3):510–7. [5] Thibault JB, Sauer KD, Bouman CA, Hsieh JA. A three-dimensional statistical approach to improved image quality for multislice helical CT. Med Phys 2007;34(11):4526–44. [6] Kalmar PI, Quehenberger F, Steiner J, Lutfi A, Bohlsen D, Talakic E, et al. The impact of iterative reconstruction on image quality and radiation dose in thoracic and abdominal CT. Eur J Radiol 2014;83(August (8)):1416–20. [7] Kahn J, Grupp U, Rotzinger R, Kaul D, Schäfer ML, Streitparth F. CT for evaluation of potential renal donors – how does iterative reconstruction influence image quality and dose? Eur J Radiol 2014;83(August (8)):1332–6. [8] Chen MY, Shanbhag SM, Arai AE. Submillisievert median radiation dose for coronary angiography with a second-generation 320-detector row CT scanner in 107 consecutive patients. Radiology 2013;267(April (1)):76–85. [9] Schuhbaeck A, Achenbach S, Layritz C, Eisentopf J, Hecker F, Pflederer T, et al. Image quality of ultra-low radiation exposure coronary CT angiography with an effective dose <0.1 mSv using high-pitch spiral acquisition and raw data-based iterative reconstruction. Eur Radiol 2013;23(March (3)):597–606. [10] Yin WH, Lu B, Hou ZH, Li N, Han L, Wu YJ, et al. Detection of coronary artery stenosis with sub-milliSievert radiation dose by prospectively ECGtriggered high-pitch spiral CT angiography and iterative reconstruction. Eur Radiol 2013;23(November (11)):2927–33. [11] O‘Neill SB, Mc Laughlin PD, Crush L, O‘Connor OJ, Mc Williams SR, Craig O, et al. A prospective feasibility study of submillisievert abdominopelvic CT using iterative reconstruction in Crohn’s disease. Eur Radiol 2013;23(September (9)):2503–12.

[12] Kalra MK, Quick P, Singh S, Sandborg M, Persson A. Whole spine CT for evaluation of scoliosis in children: feasibility of sub-milliSievert scanning protocol. Acta Radiol 2013;54(March (2)):226–30. [13] Heilbron BG, Leipsic J. Submillisievert coronary computed tomography angiography using adaptive statistical iterative reconstruction – a new reality. Can J Cardiol 2010;26(January (1)):35–6. [14] Khawaja RD, Singh S, Gilman M, Sharma A, Do S, Pourjabbar S, 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 2014;March. [15] McCollough CH, Chen GH, Kalender W, Leng S, Samei E, Taguchi K, et al. Achieving routine submillisievert CT scanning: report from the summit on management of radiation dose in CT. Radiology 2012;264(August (2)):567–80. [16] Boone JM, Hendee WR, McNitt-Gray MF, Seltzer SE. Radiation exposure from CT scans: how to close our knowledge gaps, monitor and safeguard exposure – proceedings and recommendations of the Radiation Dose Summit, sponsored by NIBIB, February 24–25, 2011. Radiology 2012;265(November (2)):544–54. [17] May MS, Wüst W, Brand M, Stahl C, Allmendinger T, Schmidt B, et al. Dose reduction in abdominal computed tomography: intraindividual comparison of image quality of full-dose standard and half-dose iterative reconstructions with dual-source computed tomography. Invest Radiol 2011;46(7):465–70. [18] Gervaise A, Osemont B, Lecocq S, Noel A, Micard E, Felblinger J, et al. CT image quality improvement using Adaptive Iterative Dose Reduction with wide-volume acquisition on 320-detector CT. Eur Radiol 2012;22(2):295–301. [19] Martinsen AC, Sæther HK, Hol PK, Olsen DR, Skaane P. Iterative reconstruction reduces abdominal CT dose. Eur J Radiol 2012;81(7):1483–7. [20] Meyer M, Klein SA, Brix G, Fink C, Pilz L, Jafarov H, et al. Whole-body CT for lymphoma staging: feasibility of halving radiation dose and risk by iterative image reconstruction. Eur J Radiol 2014;83(February (2)):315–21. [21] Martinsen AC, Sæther HK, Hol PK, Olsen DR, Skaane P. Iterative reconstruction reduces abdominal CT dose. Eur J Radiol 2012;81(July (7)):1483–7. [22] Singh S, Kalra MK, Hsieh J, Licato PE, Do S, Pien HH, et al. Abdominal CT: comparison of adaptive statistical iterative and filtered back projection reconstruction techniques. Radiology 2010;257(November (2)):373–83. [23] Mitsumori LM, Shuman WP, Busey JM, Kolokythas O, Koprowicz KM. Adaptive statistical iterative reconstruction versus filtered back projection in the same patient: 64 channel liver CT image quality and patient radiation dose. Eur Radiol 2012;22(1):138–43. [24] Halpern EJ, Gingold EL, White H, Read K. Evaluation of coronary artery image quality with knowledge-based iterative model reconstruction. Acad Radiol 2014;21(June (6)):805–11. [25] Yuki H, Utsunomiya D, Funama Y, Tokuyasu S, Namimoto T, Hirai T, et al. Value of knowledge-based iterative model reconstruction in low-kV 256-slice coronary CT angiography. J Cardiovasc Comput Tomogr 2014;8(March–April (2)):115–23. [26] Oda S, Utsunomiya D, Funama Y, Katahira K, Honda K, Tokuyasu S, et al. A knowledge-based iterative model reconstruction algorithm: can super-lowdose cardiac CT be applicable in clinical settings? Acad Radiol 2014;21(January (1)):104–10. [27] Funama Y, Taguchi K, Utsunomiya D, Oda S, Katahira K, Tokuyasu S, et al. Image quality assessment of an iterative reconstruction algorithm applied to abdominal CT imaging. Phys Med 2014;30(June (4)):527–34. [28] Suzuki S, Haruyama T, Morita H, Takahashi Y, Matsumoto R. Initial performance evaluation of iterative model reconstruction in abdominal computed tomography. J Comput Assist Tomogr 2014;38(May–June (3)):408–14. [29] Burgess AE. Comparison of receiver operating characteristic and forced choice observer performance measurement methods. Med Phys 1995;22:643. [30] Deák Z, Grimm JM, Treitl M, Geyer LL, Linsenmaier U, Körner M, et al. Filtered back projection, adaptive statistical iterative reconstruction, and a modelbased iterative reconstruction in abdominal CT: an experimental clinical study. Radiology 2013;266(January (1)):197–206. [31] Kataria B, Smedby O. Patient dose and image quality in low-dose abdominal CT: a comparison between iterative reconstruction and filtered back projection. Acta Radiol 2013;March. [32] Gay F, Pavia Y, Pierrat N, Lasalle S, Neuenschwander S, Brisse HJ. Dose reduction with adaptive statistical iterative reconstruction for pediatric CT: phantom study and clinical experience on chest and abdomen CT. Eur Radiol 2014;24(January (1)):102–11.