Gastrointestinal Imaging
Original Research
Reducing the Radiation Dose for CT Colonography Using Adaptive Statistical Iterative Reconstruction: A Pilot Study
Audio Available | 
Share
OBJECTIVE. The purpose of our study was to evaluate the feasibility of preserving image quality during CT colonography (CTC) using a reduced radiation dose with adaptive statistical iterative reconstruction (ASIR).
MATERIALS AND METHODS. A proven colon phantom was imaged at standard dose settings (50 mAs) and at reduced doses (10–40 mAs) using six different ASIR levels (0–100%). We assessed 2D and 3D image quality and noise to determine the optimal dose and ASIR setting. Eighteen patients were then scanned with a standard CTC dose (50 mAs) in the supine position and at a reduced dose of 25 mAs with 40% ASIR in the prone position. Three radiologists blinded to the scanning techniques assessed 2D and 3D image quality and noise at three different colon locations. A score difference of ≥ 1 was considered clinically important. Actual noise measures were compared between the standard-dose and low-dose acquisitions.
RESULTS. The phantom study showed image noise reduction that correlated with a higher percentage of ASIR. In patients, no significant image quality differences were identified between standard- and low-dose images using 40% ASIR. Overall image quality was reduced for both image sets as body mass index increased. Measured image noise was less with the low-dose technique using ASIR.
CONCLUSION. The results of this pilot study show that the radiation dose during CTC can be reduced 50% below currently accepted low-dose techniques without significantly affecting image quality when ASIR is used. Further evaluation in a larger patient group is warranted.
Keywords: adaptive statistical iterative reconstruction (ASIR), colonography, CT, radiation dose
| Introduction |   |
|---|
The radiation dose used during CT colonography (CTC) has recently been a topic of debate among medical professionals [1]. The real risk from low intermittent exposure to radiation is unknown because nearly all available data have been generated from Japanese survivors of the atomic bombs dropped in 1945 on Hiroshima and Nagasaki [2–4]. Despite the lack of scientific data measuring future cancer risk from radiation exposure, patients remain concerned about radiation exposure. In response to public concern and the desire for the best patient care, we sought to reduce the radiation dose during CT whenever possible without compromising image quality.
Currently, the CTC dose is one half of that administered for a conventional CT examination because of the high contrast between the soft-tissue wall and the air attenuation in the colon lumen [5]. Further dose reductions would be welcome, especially for screening studies that are likely to have an impact on a large number of patients. Adaptive statistical iterative reconstruction (ASIR) is a new method for reconstructing CT data that markedly reduces image noise [6, 7]. The potential exists for maintaining low image noise while reducing the radiation dose without affecting image quality.
We sought to evaluate ASIR for use during CTC. Our hypothesis was that the radiation dose during CTC could be further reduced by 50% without affecting diagnostic image quality by using ASIR versus currently accepted techniques.
| Materials and Methods | |
|---|
The study was conducted after institutional review board approval and informed patient consent with adherence to HIPAA guidelines. The study was divided into two parts conducted in series: the initial phantom study to determine what dose and what ASIR level should be selected for the patient arm of the study, and the patient study with the supine position acquisition as an internal control (standard dose settings) and the prone position with a 50% dose reduction using ASIR.
ASIR represents a novel approach aimed at decreasing image noise without sacrificing image quality for reduced-dose CT. Instead of relying on radiologists to accept the increased noise inherent in low-dose imaging or applying postprocessing filters that can reduce lesion conspicuity, ASIR is a noise-reducing reconstruction algorithm that replaces filtered back projection (FBP), the reconstruction algorithm typically used for CT. Unlike FBP, ASIR does not assume noise is evenly distributed across the entire image but instead uses a mathematic model to identify and remove individual projections that are noisy and deviate from the model [6]. It is this ability to selectively reduce image noise that allows ASIR to produce a higher quality image at a lower radiation dose compared with FBP techniques.
The percentage of ASIR (10–100%) is operator selectable at the console. It reflects a linear combination of the original FBP image (0% ASIR) and an essentially noise-free image created by full compliance with the mathematic model (100% ASIR). A choice of 40% ASIR implies that 40% of the ASIR image was blended with the FBP image. A preliminary phantom analysis and clinical feasibility study of low-dose body CT using 40% ASIR determined that it provided quantitative and qualitative image noise and quality nearly identical to those of routine-dose CT [7].
 View larger version (100K) | Fig. 1A —Comparison of 3D and 2D images in phantom without and with adaptive statistical iterative reconstruction (ASIR) at various dose levels. 3D endoluminal images show that as dose is decreased, mucosal nodularity and endoluminal floaters become more prominent. Many of these artifacts are less conspicuous or absent with ASIR. |
 View larger version (94K) | Fig. 1B —Comparison of 3D and 2D images in phantom without and with adaptive statistical iterative reconstruction (ASIR) at various dose levels. 2D images show that image noise is clearly visible in low-dose images without ASIR. Much of noise is removed with use of ASIR. Standard-dose CT colonography is 50 mAs without ASIR compared with 25 mAs (not shown) with ASIR. The crosshair and line are software navigation aids. |
A previously described silastic, air-filled colon phantom containing 16 soft-tissue attenuation 6-mm polyps was scanned while submersed in water in a 30-cm round water bath [8]. The morphology of the polyps was flat, with dimensions that were wider (6 mm) than tall (3 mm). CT was performed on a 64-MDCT scanner (Discovery CT750HD, GE Healthcare) using 120 kilovolt peak (kVp), 0.5-second rotation time, 0.8-mm reconstruction interval, 1.25-mm slice thickness, 0.625-mm slice collimation, 0.9 pitch, and the standard reconstruction algorithm. The phantom was scanned at 50 mAs without ASIR and at various mAs settings (40, 30, 20, and 10 mAs) with and without ASIR. Image data at each mAs setting were reconstructed using 0%, 20%, 40%, 60%, 80%, and 100% weighting of ASIR. Noise measurements were made by selecting a region of interest in air above the phantom and recording the SD. Overall image quality was judged on a scale of 0 (nondiagnostic) to 4 (highest image quality) by a single experienced gastrointestinal radiologist using the same software tools as described for the patient validation study.
Eighteen consecutive CTC examinations were performed in patients prescheduled for CTC screening examinations between July 2008 and September 2009. Patient age ranged from 34 to 85 years (mean, 68 years). Eight patients were women, and 10 were men.
The methods for patient preparation and mechanical insufflation followed the technique reported for the National CT Colonography Trial [9]. The protocol acquired both supine and prone acquisitions. The prone position used a 50% dose reduction (25 fixed mAs). The supine acquisition used the standard CTC technique (50 fixed mAs without ASIR), which served as an internal control. If the transverse diameter of a patient exceeded 40 cm on the supine scout image, the mAs was doubled for both the supine (100 mAs) and the prone (50 mAs) studies. All examinations were performed on the same CT scanner (Discovery CT750 HD). The prone (25 mAs) data sets were reconstructed at 40% ASIR. No delays in reconstruction time were experienced using ASIR compared with normal reconstruction times without ASIR.
 View larger version (140K) | Fig. 2A —Axial images for comparison of technique at level of splenic flexure. A–C, 2D axial images—supine image using standard-dose CT colonography with 50 mAs and 0% adaptive statistical iterative reconstruction (ASIR) (A), prone image using 25 mAs and 0% ASIR (B), and prone image using 25 mAs and 40% ASIR (C) show ASIR reduces noise artifact to produce improved image. |
 View larger version (180K) | Fig. 2D —Axial images for comparison of technique at level of splenic flexure. D–G, 3D axial images without and with adaptive statistical iterative reconstruction (ASIR)—50 mAs without ASIR (D), 50 mAs with ASIR (E), 25 mAs without ASIR (F), and 25 mAs with ASIR (G) show addition of ASIR reduces image artifact (mucosal nodularity and endoluminal floaters) with both dose settings. Images D and G are comparable despite reduced dose (2.1 vs 4.2 volume CT dose index). |
Image noise was recorded by selecting a region of interest (≤ 1,600 mm2) within the paraspinal muscles and by recording the SD. The region of interest was placed in the same location on prone and supine data sets. Body mass index (BMI [weight in kg / height in m2]) was calculated using a standardized calculator from measured height and weight as recorded in the medical record.
Image data for the patient study were reviewed by three experienced (> 500 proven cases) CTC radiologists using an advanced workstation (GE AW 4.3_05 with CTC software version Voxtool 6.12.3, GE Healthcare). The axial 0.625-mm-thick images were used for 2D scoring. Observers independently recorded their scores after previewing four test cases and agreeing after discussion how ratings would be determined. The same images were reviewed by all three reviewers.
Image quality, noise, and sharpness assessments were evaluated at three similar anatomic sites in the colon in each patient: at a prominent fold within the rectosigmoid junction, at the splenic flexure of the colon, and at the ileocecal valve. These sites were chosen to assess the technique for a simulated lesion (ileocecal valve) and in areas with and without potential beam hardening (rectosigmoid within bony pelvis and splenic flexure, respectively).
Image quality was graded on a 5-point scale from 0 to 4 (0 for nondiagnostic, 1 for severe artifact with low confidence, 2 for moderate artifact or moderate diagnostic confidence, 3 for mild artifact or high confidence, and 4 for well seen without artifacts and high confidence of detecting a lesion ≥ 5 mm).
Image noise was subjectively graded on a similar scale from 0 to 4 (0 for nondiagnostic, 1 for severe noise, 2 for moderate perceptible noise, 3 for mild perceptible noise, and 4 for no perceptible noise).
Image sharpness, assessed by evaluating the aortic contour in the upper abdomen, was graded on a 5-point scale (0 for nondiagnostic, 1 for poorly defined edge sharpness, 2 for moderately unsharp edges, 3 for mildly unsharp edges, and 4 for very sharp edges).
Scores were averaged for each reader for routine-dose and low-dose images. The mean scores of all readers were compared between routine-dose and low-dose images. A clinically significant difference in qualitative assessments of image noise and quality was considered to be a change in score of ≥ 1. The intraclass correlation coefficient was calculated to measure the interobserver agreement of the three ratings. A correlation of ≥ 0.7 was considered excellent, 0.3–0.7 good, and < 0.3 was considered poor.
The statistical software program SAS 9.1 (SAS Institute) was used for data analysis, and statistical significance was defined as p ≤ 0.01.
| Results | |
|---|
The results of the phantom study are summarized in Table 1. Image noise decreased as the percentage of ASIR was increased and as the mAs was increased. At the standard dose of 50 mAs without ASIR, noise was measured at 12.7 HU. With varying amounts of ASIR, the noise levels at lower doses could match or exceed this baseline value. For example, 30 mAs at 60% ASIR was associated with a noise level of 10.7 HU. Even at the lowest dose setting of 10 mAs with progressively higher levels of ASIR, noise measurements might match or be lower than the standard-dose noise measurement without ASIR.
Ratings of 2D image quality could achieve the maximum score of 4 for all dose levels when ASIR was applied (20–100%). For 50 mAs without ASIR, the baseline rating for 3D image quality was 3 of 4. Ratings of 3 and 4 could be achieved with the dose setting reduced to 20 mAs, regardless of the ASIR setting. Low rankings of image quality were more dependent on dose settings than on ASIR values. Higher image quality ratings were modestly improved using ASIR, especially with dose levels of ≤ 30 mAs (Fig. 1A, 1B).
Table 2 summarizes the results of the patient study. Averaged data for all 18 patients as well as per BMI size group (< 20, 20–29, and ≥ 30) are shown.
Objective measurements of image noise varied from 28.3 to 61.8 HU among the three patient size groups using standard CTC dose settings in the supine position. With a 50% dose reduction (i.e., 25 mAs) and 40% ASIR, the mean noise measurement was lower, ranging from 23.7 to 51.2 HU. Objective noise increased as patient size increased.
There were no differences of one or greater when comparing subjective image noise and image quality scores between routine- and low-dose images.
There were a total of six averaged image noise comparisons (three sites: rectosigmoid, splenic flexure, ileocecal valve with two comparisons [2D and 3D] per site). For four of the six comparisons, there was more subjective noise on the low-dose images, although the difference did not exceed 0.3. For the two remaining comparisons, noise was graded as essentially identical (difference ≤ 0.1). Image noise scores were worse in the rectosigmoid compared with the splenic flexure and ileocecal valve (ICV).
Similarly, averaged subjective measures of image quality were generally rated slightly lower with low-dose images in five of the six comparisons; however, these differences did not exceed 0.4. In only one of the six image quality comparisons, the low-dose images with ASIR rated better than routine dose (3D image quality splenic flexure, difference of 0.2). Image quality scores were worse in the rectosigmoid compared with the splenic flexure and ICV. Image sharpness as assessed by the aortic edge was essentially identical between routine- and low-dose images (routine dose, 2.5; low dose, 2.4).
Evaluation by patient size is limited because the number of patients in each group varied (three for BMI < 20, 10 for BMI 20–29, and five for BMI ≥ 30). In general, for both routine- and low-dose images, the subjective image noise and image quality scores worsened as patient size increased (i.e., lower scores). For example, the average routine-dose noise scores ranged from 2.9 to 3.8 HU in the smallest patients and 2.0–2.8 HU in the largest patients. The low-dose noise scores ranged from 2.4 to 3.3 HU in the smallest patients and 1.9–2.5 HU in the largest patients.
The differences in image quality were most marked between routine- and low-dose images when evaluating 2D rectosigmoid images in patients with a BMI of < 20 (difference of 0.8). The worst image sharpness scores were found in the smallest patients for both routine- and low-dose images.
The intraclass correlation coefficient values among readers for routine-dose noise, low-dose noise, routine-dose quality, and low-dose quality were 0.70 (95% CI, 0.28–0.99), 0.75 (0.36–0.99), 0.74 (0.35–0.99), and 0.74 (0.35–0.99), respectively.
| Discussion | |
|---|
Our hypothesis, that the standard radiation dose for CTC (i.e., 50 mAs) could be reduced 50% when combined with ASIR without substantially affecting image quality, was validated both in a phantom and in humans. Objective image noise measures were actually lower at 25 mAs with ASIR than at a routine dose of 50 mAs without ASIR for all patients regardless of BMI.
The phantom study indicated that to maintain 3D image quality scores of 3 or higher, a dose threshold of 20–30 mAs should be used. Therefore, a dose of 25 mAs was chosen for the patient study, which represents a 50% reduction compared with conventional CTC. A 40% ASIR level was chosen on the basis of our anecdotal experience using higher ASIR levels in other patients that produced 2D images with an artificial, noiseless, over-smoothed appearance, which is generally not acceptable to most radiologists. Therefore, we sought to use the lowest ASIR level to maintain image noise and subjective quality.
In the patient study, when subjective scores of routine- and low-dose images were compared, low-dose images (both 2D and 3D) always had slightly lower scores than routine-dose images. The difference between the routine- and low-dose scores, however, was always less than 1, and therefore not considered to be clinically important. This finding was somewhat surprising because objective measures of image noise showed a mean 24% noise reduction using the 40% ASIR setting. This much noise reduction may not be appreciated subjectively because CTC studies are routinely performed at a low dose and are inherently noisier than conventional CT examinations.
In general, the best subjective scores of image quality and image noise were found in the smallest patients, and, not surprisingly, both scores tended to worsen with increased patient size. For example, in small patients, average scores were in the low to mid 3 range, whereas larger patients had scores in the mid to low 2 range (the highest achievable score was 4). Interestingly, however, for both routine- and low-dose images, image sharpness was graded slightly worse in smaller patients than average or large patients. This may be explained by less periaortic fat in the smaller patients so that the contrast between the unenhanced aorta and surrounding structures is less apparent, an issue exacerbated by lower radiation dose. These differences based on patient size, however, need to be further evaluated and confirmed in a larger patient group.
This study was performed using a manual technique setting. It may be that implementation of the automated exposure control now available on most CT scanners would provide higher-quality images of the pelvis, where the lowest image quality and noise scores were encountered in our patients. These higher doses could be modified by using the lower settings in the upper abdomen. Although it was possible to use automated exposure control for our patients, we chose not to do so to keep the dose constant for image analysis and scientific comparison. Dividing the patient groups by BMI allowed us to provide valid comparisons between techniques. In clinical practice, a low-dose technique and automated exposure control with ASIR reconstruction on scanners can be used when ASIR is available [7].
Higher percentages of ASIR were not evaluated in the patient study because the phantom study did not show a clear advantage to using higher levels and because of our anecdotal experience of an artificial appearance to the images with higher ASIR settings. Future studies may be helpful to evaluate whether a higher percentage of ASIR in 3D images in clinical CTC produces better image quality than we found in this study. If so, 2D images could be evaluated at 40% ASIR and 3D images at 100% ASIR, although this would require two different reconstructions and double the number of images generated. We did not study whether image quality would be improved using standard-dose settings because the aim of this study was to reduce radiation dose if possible.
Only a few published reports have evaluated the clinical effectiveness of CTC below the standard dose of 50 mAs. Cohnen et al. [10] reported sensitivity of 81% and 62% for polyps ≥ 1 cm and 6–9 mm, respectively, with a dose of 10 mAs. In an animal study, Branschofsky et al. [11] used a mathematic filter that reduced noise by 50–70% for a 10-mAs scan of two pig colons. They reported that using such a low dose resulted in distortions of lesion size and shape. In the current study, we did not observe image distortion. A study by Iannaccone et al. [12] using 10 mAs had sensitivity of 86% for polyps ≥ 6 mm in diameter but missed all flat polyps. Our study did not evaluate the detection of flat polyps; however, all the phantom polyps could be classified as flat (6 mm in width, 3 mm in height) and were readily visible. Further study to confirm similar diagnostic performance for polyp detection should be considered on the basis of this pilot evaluation.
There were several limitations to our study. First, the different ASIR levels of the phantom images were not masked for the radiologist reviewing them. However, bias was unlikely to have occurred because observer scores were quite similar to objective measures of image noise. Second, it was not possible using the phantom to assess the effects of ASIR at the sites of different polyps or for polyps of different sizes. The phantom did test the smallest-size polyp (6 mm) that is clinically important and a morphology (flat) that is the most difficult to detect. Because these polyps represent the most difficult clinical scenario, we anticipate that larger and more sessile polyps should be more easily detected. Third, although our assessment of interobserver agreement was good, with a correlation coefficient of 0.7, our image quality measurements are subjective and results could be different using a different cohort of readers. Finally, only a small patient cohort was evaluated. Nonetheless, a larger patient study is warranted in view of this encouraging pilot data. We also did not evaluate the effect of reduced radiation doses in the detection of extracolonic abnormalities, which would be of interest for future studies.
In summary, this pilot study supports the hypothesis that the radiation dose for CTC can be reduced 50% below current accepted techniques when combined with ASIR (at a 40% setting), without having a clinically significant impact on image noise or image quality. Further clinical evaluation is recommended.
Address correspondence to C. D. Johnson.
| References | |
|---|
