Introduction

Although CT colonography is now widely accepted as an effective tool for screening for colorectal cancer [1, 2], radiation exposure remains as a drawback. Although debatable, the risk of inducing fatal cancer by means of CT colonography with radiation doses ranging from 4 to 12 mSv has been estimated to be 0.01–0.02% in a 50-year-old individual [3, 4]. The population-based risk in the future might not be negligible, considering the increased use of screening CT colonography.

The use of automatic exposure control (AEC) techniques [5] enables a significant reduction in radiation exposure without loss of image quality in CT colonography [6] as in any other CT applications. The principle of the AEC techniques is to adjust x-ray intensity according to regional x-ray attenuation in the body, thereby adapting radiation exposure for individual patients while maintaining constant image quality. However, when the region undergoing imaging contains a highly attenuating material, the AEC techniques can overcompensate suggesting excessive radiation exposure. For instance, it has been reported that the presence of a metallic prosthesis is associated with a 34.1% increase in the radiation exposure in abdominal–pelvic CT obtained with an AEC technique [7].

For CT colonography, a similar concern regarding the excessive radiation exposure can be raised if an AEC technique is used in conjunction with stool or fluid tagging agents such as barium or iodine. The use of the tagging agents is essential to reduce false-positive diagnoses, especially if the colon is to be cleansed in a less vigorous manner to improve patient compliance in a screening programme [8]. The presence of large amounts of the highly attenuating material in the bowel might result in an inadvertent increase in the radiation exposure by an AEC technique.

Our study aimed to determine whether the amount of tagged stool and fluid significantly affects the radiation exposure in low-dose screening CT colonography performed with an AEC technique.

Materials and methods

Our institutional review board approved this prospective study and waived the patient informed consent as all CT were obtained following our clinical protocol and patient confidentiality was maintained. One author is an employee of Philips Healthcare, the manufacturer of the imaging device used in this study. Those authors who are not employees of Philips Healthcare had control of the data and the information submitted for publication and control of inclusion of any data and information that might present a conflict of interest for the author who is an employee for Philips Healthcare.

Study sample

The study initially included 316 consecutive patients screened with low-dose CT colonography in our institution between August 2008 and November 2008. The rationale for the sample size determination is described in Appendix 1. The patients were typically asymptomatic and had average risk of colorectal cancer. By reviewing the CT colonography images, a study coordinator excluded three patients with metallic prosthesis within the imaging range. Additionally, two patients were excluded as the imaging showed severe motion artefacts hindering subsequent measurement of the tagged stool and fluid by using image post-processing techniques (described below).

The remaining 311 patients (age range, 40–82 years; mean±SD, 54.7 ± 8.2 years) were finally included (Fig. 1). They were 179 men (40–82; 54.0 ± 8.0 years) and 132 women (40–81; 55.6 ± 8.4 years). The abdominal circumference (cm) at the umbilicus level was measured for each patient.

Fig. 1
figure 1

Patient flow diagram

Full size image

Bowel preparation

For CT colonography, the selection of the cathartic agent and tagging agent was made at the discretion of the referring physician, considering the patients’ physical state, particularly the renal and cardiovascular functions. In 271 patients who were administered magnesium citrate (50 g Magcorol powder; Taejoon Pharmaceutical, Seoul, Korea) (n = 252) or sodium phosphate (n = 19) (90 mL Colclean; Taejoon Pharmaceutical) for bowel cleansing, three doses of 200 mL of 4.6% w/v barium (EZ-CT; Taejoon Pharmaceutical) were used for stool tagging. In the remaining 40 patients who were administered polyethylene glycol electrolyte (4 L Colyte-F; Taejoon Pharmaceuticals), a single dose of 100 mL of water-soluble iodine agent (diatrizoate meglumine and diatrizoate sodium; Gastrografin, Bracco Diagnostics, Princeton, NJ, USA) was used for fluid tagging. All patients were prescribed low-residue diet on the day before CT colonography and were administered 10 mg of bisacodyl suppository (Dulcolax; Boehringer Ingelheim, Seoul, Korea) in the morning on the day of the examination.

CT colonography

An automated CO2 delivery system (PROTOCO2L; E-Z-Em, Westbury, NY, USA) was used. No spasmolytics were used. Non-contrast low-dose prone and supine acquisitions were obtained using 64-detector-row CT equipments (Brilliance; Philips Healthcare, Cleveland, OH, USA) with the use of an AEC system. Transverse CT images were reconstructed with the section thickness of 0.67 mm and the reconstruction interval of 0.6 mm. Other imaging parameters are tabulated in Table 1. Mean volume CT dose index, CTDI vol (mGy), was recorded for each of the prone and supine helical scans. The AEC maintains approximately the same preferred image noise level across patients (patient-to-patient modulation; Automatic Current Selection [ACS], Philips Healthcare) and across images along the craniocaudal axis within an acquisition (z-axis modulation [Z-DOM]; Philips Healthcare) (Appendix 2). We did not use an angular modulation, as simultaneous z-axis and angular modulations were not available for the imaging device we used and the vendor recommended choosing the former rather than the latter. The effective dose of the prone and supine scans, estimated by multiplying the dose-length products reported on the CT console by a conversion factor (0.017 mSv·mGy1·cm1) [9], was 1.3–4.7 (mean±SD, 2.3 ± 0.6) mSv. This radiation dose level was empirically determined for our Asian population based on our previous experience [10]. The images were interpreted as part of routine practice, as described elsewhere [10]. We did not analyse the diagnostic performance, as it was not the purpose of our study.

Table 1 Imaging parameters
Full size table

Tagged stool and fluid

From the image dataset of each prone or supine image, the study coordinator segmented the voxels corresponding to the tagged stool and fluid remaining inside the colon as well as the other segments of the alimentary tract with a semi-automatic method (Fig. 2, Appendix 3). Although either the barium or iodine agent was used in each patient, it was considered that each tagging agent could tag both residual stool and fluid. From the segmentation results, a computer scientist calculated the volume (vol) (cm3) as well as the mean (HU mean ) and median (HU median ) CT numbers (HU) of the tagged stool and fluid using the Matlab program (version R2007a, MathWorks, Natick, MA, USA). In addition, the x-ray attenuation by the tagged stool and fluid (ATT) was estimated with the Beer-Lambert law as \( {e^{{\mu_{water}} \times \left( {1 + \frac{{H{U_{mean}}}}{{1000}}} \right){ } \times { }\sqrt[3]{{vol}}}} \), where μ water is the linear attenuation coefficient of water (0.193 cm−1) [11] (Appendix 4).

Fig. 2
figure 2

Segmentation results in a 57-year-old woman. a Transverse CT colonography image. Dark and grey areas indicate segmented air inside the colon and small bowel, respectively. b Volume rendering image of segmentation results for tagged stool and fluid inside the alimentary tract. The measured vol, HU mean , HU median and ATT were 64.2 cm3, 411.2 HU, 416.0 HU, and 2.97, respectively. c Endoluminal image. The surface irregularity was rated as grade 1 by both radiologists. ATT, estimated x-ray attenuation by the tagged stool and fluid; HU mean , mean CT number of the tagged stool and fluid; HU median , median CT number of the tagged stool and fluid; vol, volume of the tagged stool and fluid.

Full size image

Image noise

Although the AEC mechanisms we used were designed to maintain an approximately constant image noise level [5, 12], this constancy was assessed with two separate analyses.

First, the image noise was quantitatively measured as the standard deviation of CT numbers of intracolonic air voxels segmented from the images using a semi-automatic method by the study coordinator (Fig. 2, Appendix 3). We chose to measure the image noise in the intracolonic air rather than in the background air outside the body, as the former is more relevant than the latter in assessing the quality of three-dimensional (3D) endoluminal fly-through images.

Second, two independent radiologists visually assessed the irregularity of the endoluminal surface of the colon, as an indicator of the image noise. Each radiologist had experience of interpretation of at least 700 CT colonography examinations. The radiologists were unaware of the study design and blinded to the measured image noise. This subjective analysis was performed in a retrospective manner to maintain consistency in the radiologists’ visual assessment. As this analysis aimed to confirm that the image noise level was approximately constant across the patients, we included only 20 patients, ten patients with the lowest image noise measurements (the mean of prone and supine scans) and ten patients with the highest image noise measurements. The order of patients was randomised for each radiologist. After a complete 3D fly-through review (rectum-to-cecum and cecum-to-rectum for the prone and supine images), each radiologist graded the surface irregularity on a four-grade scale: grade 0, very smooth and shiny surface; grade 1, slightly irregular surface, but polyp detection would not be affected; grade 2, the detection of a flat polyp can be potentially affected because of the irregular surface; and grade 3, unacceptable quality due to the severely irregular surface. Visualisation parameters of the reviewing software (Aquarius, Terarecon, San Mateo, CA, USA) were set to default settings.

Statistical analysis

For all measurements (vol, HU mean , HU median , image noise and CTDI vol ), the mean values of the prone and supine scans were used in the statistical analyses. The Pearson correlation was measured between CTDI vol and each of vol, HU mean , HU median and ATT of the tagged stool and fluid. For the variables with skewed data distribution, natural logarithm or square root transforms of the data were used instead of the original data. CTDI vol , vol and ATT were transformed to ln(CTDI vol ), \( \sqrt {{vol}} \) and ln(ATT), respectively.

Multiple linear regression analysis was used to determine the effect of ATT on CTDI vol while adjusting for other variables including age, sex, abdominal circumference and the type of tagging agent. In a similar manner, another multiple linear regression analysis was performed to determine the effect of ATT on the image noise, while adjusting for other variables including age, sex, abdominal circumference, the type of tagging agent and CTDI vol . The vol, HU mean and HU median did not enter the multiple linear regression models because they were represented by ATT. To ensure linearity between the continuous variables included in the multiple linear regressions, the noise and abdominal circumference were transformed to \( \frac{1}{{nois{e^2}}} \) and \( {e^{0.719 \times {\mu_{water}} \times \left( {\frac{{abdominal circumference}}{\pi } - 32} \right)}} \), respectively, before entering the regression models (Appendix 4). A p value of <0.05 was considered to indicate statistical significance. Statistical software (SPSS, version 15.0; SPSS, Chicago, IL, USA) was used.

Finally, the concordance rate was measured for the two radiologists’ grading results.

Results

Correlation

CTDI vol varied from 0.88 to 2.54 (mean±SD, 1.45 ± 0.32) mGy. The mean vol, HU mean , HU median and ATT were 61.2 cm3, 403.9 HU, 388.6 HU and 2.72, respectively (Table 2). There was no significant correlation between ln(CTDI vol ) and each of \( \sqrt {{vol}} \) (r = −0.03 [95% CI, −0.14, 0.08], p = 0.59), HU mean (r = −0.08 [−0.20, 0.03], p = 0.15), HU median (r = −0.09 [−0.20, 0.02], p = 0.11), and ln(ATT) (r = −0.03 [−0.14, 0.08], p = 0.61).

Table 2 Measured data
Full size table

The effect of ATT on CTDI vol

The results of the multiple linear regression analysis showed that ATT did not significantly affect CTDI vol (p = 0.93) when the other four variables were adjusted. Of the tested variables, the transform of abdominal circumference (Table 2), \( {e^{0.719 \times {\mu_{water}} \times \left( {\frac{{abdominal circumference}}{\pi } - 32} \right)}} \), was the only factor significantly affecting CTDI vol (p < 0.001) (Table 3).

Table 3 Results of multiple linear regression analysis for variables affecting CTDI vol
Full size table

The effect of ATT on Image Noise

The image noise measured in the intracolonic air regions ranged from 59.5 to 64.1 (mean±SD, 61.5 ± 0.9) HU (Table 2). The regression model including ATT and the five other variables did not explain the transform of noise, \( \frac{1}{{noise}^2} \), in a statistically significant manner (R 2 = 0.02, p = 0.38).

Visual analysis

The visual analysis included the ten patients with the lowest image noise measurements (range, 59.5–59.7 HU) and the ten patients with the highest image noise measurements (63.2–64.1 HU). One of the two radiologists rated the surface irregularity in all 20 cases as grade 1 (Fig. 2). The other radiologist rated the surface irregularity as grade 1 in 19 cases and as grade 2 in one patient who had a measured noise level of 63.2 HU. The concordance rate between the two radiologists was 95% (19/20).

Discussion

In our results, CTDI vol varied from 0.88 to 2.54 (mean±SD, 1.45 ± 0.32) mGy, while the image noise was maintained within the range of 59.5 to 64.1 (mean±SD, 61.5 ± 0.9) HU, and the radiologists rated the surface irregularity as grade 1 in all but one of the 20 tested cases. These results indicate that the AEC technique adapted radiation exposure for individual patients appropriately to maintain approximate constancy of the image noise.

There was no significant correlation between ln(CTDI vol ) and each of the parameters \( \sqrt {{vol}} \) , HU mean , HU median and ln(ATT). The regression analysis showed that ATT did not significantly affect CTDI vol . These results indicate that the amount of tagged stool and fluid did not significantly affect the radiation exposure. Therefore, when the stool or fluid tagging technique was used, the concern regarding excessive radiation exposure due to overcompensation of the AEC was not substantiated.

Our results demonstrating an insignificant effect of the amount of the tagged stool and fluid on the radiation exposure might be considered counterintuitive, as tagged stool and fluid should contribute to the overall x-ray attenuation of the region undergoing imaging. A plausible explanation for our results would be that the attenuation by the tagged stool and fluid might be considerably small, or even negligible, compared with the attenuation by the patient’s body. In other words, the AEC system might not be precise enough to respond to a small change in the x-ray attenuation by the tagged stool and fluid, though the system competently responded to the relatively larger change in the x-ray attenuation according to the patient’s body size. In theory, an excessive amount of tagged stool and fluid could potentially increase the radiation exposure, as does a metallic prosthesis [7]. Therefore, it is unclear whether our results can be reproduced in patients with different colon preparations leaving more residues of tagging agents by using no cathartic agent or by tagging with both barium and iodine agents. Nevertheless, it should be noted that our results can be still generalised to patients undergoing bowel preparations similar to the regimens used in our study.

Our study has limitations. First, we used imaging devices of a single type. We did not use angular modulation as simultaneous z-axis and angular modulations were not available for the imaging device we used. Investigations testing other types of device capable of such simultaneous modulations are needed to further generalise our results. Second, the analysis of image quality with regard to the image noise relied on the radiologists’ subjective rating for the surface irregularity. It would have been ideal if polyp detection performance could have been evaluated to corroborate the finding that the image noise level was maintained within an acceptable range. However, we were unable to perform such an analysis because many patients did not undergo subsequent optical colonoscopy.

Our study demonstrated that the amount of tagged stool and fluid did not significantly affect the radiation exposure in low-dose screening CT colonography performed with an AEC technique. In patients undergoing bowel preparations similar to the regimens used in our study, the concern regarding excessive radiation exposure by overcompensation of the AEC can be mitigated when a stool or fluid tagging technique is used.