| HOME | HELP | FEEDBACK | SUBSCRIPTIONS | ARCHIVE | SEARCH | TABLE OF CONTENTS |
|
|
|
|||||||
|
|||||||||
| ||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||
Experimental Studies |
1 From the Departments of Clinical Radiology (J.W., R.F., A.B., H.K., T.A., W.H.) and Medical Informatics and Biomathematics (N.O.), University of Muenster, Albert-Schweitzer-Str 33, 48149 Muenster, Germany. Received February 20, 2005; revision requested April 19; revision received August 3; accepted September 1; final version accepted January 3, 2006. Address correspondence to J.W. (e-mail: weslingj{at}uni-muenster.de).
 |
ABSTRACT |
|---|
 TOP ABSTRACT INTRODUCTIONMATERIALS AND METHODSRESULTSDISCUSSIONADVANCE IN KNOWLEDGEReferences |
|---|
Materials and Methods: A colon phantom with simulated haustral folds and 10 polyps of varying size (2.0–8.0 mm) was imaged at four- and 16-section CT (section thicknesses of 1.25 and 0.75 mm, reconstruction increments of 0.8 and 0.7 mm, and 100 and 10 mAs, respectively, and 120 kV for both) and at 1.5- and 3.0-T MR imaging (three-dimensional gradient-recalled echo sequence, section thickness of 1.4 mm). Three-dimensional endoluminal images were assessed by 10 reviewers for each modality regarding polyp detection. Comparisons of sensitivities were performed by using logistic regression.
Results: Overall, polyps were detected with a sensitivity of 87% (95% confidence interval [CI]: 80%, 94%) at four-section CT, 92% (95% CI: 87%, 97%) at 16-section CT, 56% (95% CI: 46%, 66%) at 1.5-T MR imaging, and 55% (95% CI: 45%, 65%) at 3.0-T MR imaging. The detection of polyps at least 4 mm in diameter was not influenced by the modality or radiation dose (sensitivity of 100%). CT performed in low-dose mode depicted all polyps with a diameter of at least 3 mm. Polyps smaller than 3 mm in diameter were detected with a sensitivity of 7.5% (1.5-T MR imaging), 22.5% (3.0-T MR imaging), and 20% (low-dose CT); detection rates were significantly greater (P < .001) with normal-dose CT (four section, 67.5%; 16 section, 82.5%). Increased spatial resolution (with CT) and higher field strength (with MR imaging) had no significant effect on polyp detection.
Conclusion: With both multi–detector row CT and MR imaging, readers detected polyps above the clinically relevant threshold diameter of 6 mm, with similar sensitivities.
© RSNA, 2006
|
INTRODUCTION |
|---|
|
TOPABSTRACTINTRODUCTIONMATERIALS AND METHODSRESULTSDISCUSSIONADVANCE IN KNOWLEDGEReferences |
|---|
However, radiation exposure is a drawback for CT colonography. MR colonography, in contrast, is not associated with radiation exposure. Furthermore, MR imaging contrast agents have a more favorable safety profile than do CT contrast agents (6,7). The greater potential of contrast resolution with MR imaging theoretically favors MR imaging over CT colonography. Therefore, focus on MR imaging for colorectal cancer screening has been proposed (5). However, to the best of our knowledge, studies that compare the two techniques in regard to polyp detection are still lacking.
The purpose of our study, therefore, was to compare multi–detector row (four- and 16-section) CT, including a low-dose protocol, with high-field-strength (1.5- and 3.0-T) MR imaging for reader detection of colorectal polyps in a colon phantom.
|
MATERIALS AND METHODS |
|---|
|
TOPABSTRACTINTRODUCTIONMATERIALS AND METHODSRESULTSDISCUSSIONADVANCE IN KNOWLEDGEReferences |
|---|
|
|
|
Image Acquisition
The phantom was imaged by using four- and 16-section CT (Somatom Volume Zoom and Sensation 16, respectively; Siemens, Forchheim, Germany) and 1.5- and 3.0-T MR imaging (Gyroscan Intera; Philips).
Multi–detector row CT.—The phantom was placed in the center of the CT gantry, with the longitudinal axis of the phantom parallel to the longitudinal axis of the gantry. CT images (three protocols) were acquired with a detector collimation of 4 x 1.0 mm (for one normal-dose and one low-dose data set) and 16 x 0.75 mm (for one normal-dose data set). Dose per section was 100 mAs, and tube voltage was 120 kV. In addition, the phantom was scanned with a low-dose protocol by reducing the tube current to 10 mAs (clinically evaluated protocol with four-section CT). Effective dose of the low-dose protocol was estimated in a recent study (9) to be 0.9 and 1.2 mSv for male and female subjects, respectively. To compensate for image noise, we used a very smooth filter (B10) for the low-dose protocol and a medium-smooth filter (B30) at 100 mAs. Section thickness and reconstruction interval were 1.25 and 0.8 mm, respectively, for four-section CT and 0.75 and 0.7 mm, respectively, for 16-section CT. Pitch values were kept constant at 1.5 for both CT scanners, since reasonable and acceptable acquisition times were to be ensured. The acquisition matrix was 512 x 512, with a 40-cm display field of view, which resulted in a nominal pixel size of 0.78 x 0.78 mm.
MR imaging.—The phantom was placed in the center of the MR imager along the longitudinal axis. Circularly polarized standard birdcage head coils were used at 1.5 and 3.0 T to provide measurements with similar spatial homogeneity at both field strengths. Abdominal phased-array coils with equal characteristics for both imagers were not available. The phantom was imaged with three-dimensional turbo-field-echo sequences, with identical contrast and geometry parameters at both field strengths (two data sets): magnetization preparation by an inversion pulse every 99 msec, 3.0/1.44 (repetition time msec/echo time msec), 15° flip angle, acquired matrix of 192 x 123 x 60 over a field of view of 350 x 262.5 x 168 mm (frequency x phase x section encoding), and reconstructed voxel size after Fourier interpolation of 256 x 205 x 120 voxels (1.37 x 1.37 x 1.4 mm). Total imaging time was 38.9 seconds. The nominal field of view exceeded the size of the coil. It was selected to simulate measurement parameters for whole-body geometry, although a signal was actually obtained only from within the head coil.
Image Processing and Assessment
Image data (CT and MR imaging) were transferred to a workstation with multiplanar and volume-rendering and surface-shaded display capabilities (Syngo Colonography; Siemens). Endoluminal views (Fig 2) for each of the modalities were created by using a surface-shaded display algorithm with constant rendering parameters. The endoluminal images were assessed with a semiautomated fly-through by 25 radiologists (range of experience in abdominal CT and MR imaging, 2–16 years) who were blinded to the image acquisition parameters and imaging modalities. The readers were classified into three groups according to the level of experience (2–3, 4–6, and 
7 years). From these groups, a constant number of readers were randomly assigned into groups of five per modality (five groups of five readers each, assigned to five protocols [modalities], for a total of 25 readers). Readers independently assessed the presence, number, and location of lesions. The detected lesions were graded for contour definition by using a five-point grading scale. A score of 1 (excellent) indicated the presence of a lesion with excellently visualized margins, a score of 2 (good) indicated the presence of a lesion with well-visualized margins, a score of 3 (fair) indicated the presence of a lesion with fairly visualized margins, a score of 4 (poor) indicated the presence of a lesion with poorly visualized margins, and a score of 5 (almost impossible) indicated the presence of a suspicious lesion or perhaps an artifact that mimicked a lesion. To assess the intergroup variability and to prevent learning effects, each group was assigned to a second reading for data sets from a different modality 8 weeks later (10 readers per modality).
|
|
Logistic regression was used to analyze size-dependent differences in polyp depiction for each modality and to analyze differences in polyp depiction between the modalities, including polyp detection as a dependent variable and polyp size and modality as independent variables. Estimation of logistic model parameters was performed by means of generalized estimating equations. To account for stochastic dependence of repeated observations made by each reader, a working correlation matrix of unstructured type was applied—that is, all observations made by a certain reader are assumed to be equally correlated. The 95% confidence interval (CI) for predicted probabilities of polyp detection is given. To compare "polyp contour definition" for the different modalities (1.5- and 3.0-T MR imaging, four- and 16-section CT, and low-dose four-section CT), a generalized linear model with cumulative LOGIT link was set up. Parameter estimation was performed by means of generalized estimating equations. Categorical variables were expressed as frequency and percentage, whereas continuous variables were presented as mean ± standard deviation. Differences were considered significant at P < .05.
|
RESULTS |
|---|
|
TOPABSTRACTINTRODUCTIONMATERIALS AND METHODSRESULTSDISCUSSIONADVANCE IN KNOWLEDGEReferences |
|---|
The flexible plastic (Dubliplast; Dentaurum) had T1 and T2 relaxation times of 626 msec ± 18 [standard deviation] and 31 msec ± 1, respectively, at 3.0-T MR imaging and 443 msec ± 24 and 35 msec ± 1, respectively, at 1.5-T MR imaging. Lard had T1 and T2 relaxation times of 257 msec ± 4 and 35 msec ± 2, respectively, at 3.0 T and 195 msec ± 1 and 57 msec ± 1, respectively, at 1.5 T. Thus, the relaxation times of the phantom material are on the same order as those of in vivo soft tissue (10).
Polyp Detection at Multi–Detector Row CT
Estimated overall sensitivity (Fig 3) for polyp detection was 87% (95% CI: 81%, 91%) for four-section CT and 92% (95% CI: 80%, 97%) for 16-section CT. Depiction of polyps deteriorated with decreasing polyp size for both modalities. The depiction rate was 100% for polyps 3–8 mm for both four- and 16-section CT (Fig 3). The sensitivity decreased to 82.5% (95% CI: 62%, 93%) at 16-section CT and 67.5% (95% CI: 54%, 78%) at four-section CT for polyps 2 mm in diameter. The difference between four- and 16-section CT for detection of 2-mm polyps was not statistically significant (P = .185). Polyp contour definition for lesions smaller than 6 mm in diameter was rated similar with the two modalities (Fig 4). The average score was 1.92 for four-section CT and 1.69 for 16-section CT. There was a substantial but nonsignificant decrease in polyp contour definition (four-section CT vs 16-section CT) for 2-mm polyps (P = .104) (Fig 4). There was a substantial but nonsignificant difference in the number of false-positive results (Fig 5). By using four-section CT, an average of 0.5 lesion per reader was mistaken for a polyp, whereas by using 16-section CT an average of 0.9 lesion per reader was reported as a false-positive finding.
|
|
|
Polyp Detection at MR Imaging
The overall sensitivity for polyp depiction (Fig 3) was 56% (95% CI: 50%, 62%) for 1.5-T MR imaging and 55% (95% CI: 45%, 64%) for 3.0-T MR imaging. All polyps with a diameter of at least 4 mm were detected, but detection decreased to 7.5% (1.5-T MR imaging) and 22.5% (3.0-T MR imaging) for polyps as small as 2 mm. MR imaging at 1.5 and 3.0 T did not differ significantly in the depiction of polyps of 3 mm (P = .062) or 2 mm (P = .143) in diameter. For polyps smaller than 6 mm in diameter, the average contour definition score was 3 for both modalities, and for lesions 2 mm in diameter, the score was 5.0 at 1.5-T imaging and 4.5 at 3.0-T imaging (Fig 4). There was an average of 0.2 false-positive finding per reader at 3.0 T and none at 1.5 T (Fig 5).
Multi–Detector Row CT versus MR Imaging
To compare multi–detector row CT and MR imaging (Fig 2), the detection rates and contour definition scores were summarized for four- and 16-section CT on the one hand and 1.5- and 3.0-T MR imaging on the other hand. Detection rates for MR imaging decreased from 100% to 15% below a threshold diameter of 4 mm, whereas for CT, sensitivity remained high for polyps as small as 3 mm in diameter. Detection of 2-mm lesions was successful in 75% (95% CI: 63%, 84%) of cases by using CT. CT was significantly better than MR imaging in depicting polyps smaller than 6 mm (four-section CT, 84% [95% CI: 77%, 89%] and 16-section CT, 90% [95% CI: 75%, 96%] vs 1.5-T MR imaging, 45% [95% CI: 38%, 52%] and 3.0-T MR imaging, 44% [95% CI: 32%, 56%]; P < .001). In terms of polyp contour definition, comparable scores of between 1.0 and 1.4 for CT and MR imaging were found for polyps larger than 4 mm (Fig 4). Although there was a trend toward a better delineation of 4-mm polyps with CT, the difference was not significant (P = .106). CT was significantly better than MR imaging in depicting 2-mm (P < .001) and 3-mm (P < .001) polyps.
|
DISCUSSION |
|---|
|
TOPABSTRACTINTRODUCTIONMATERIALS AND METHODSRESULTSDISCUSSIONADVANCE IN KNOWLEDGEReferences |
|---|
Our study used clinically approved protocols tested in vitro (9) and in vivo (11,14,15). However, for MR imaging at 3.0 T, no routine protocols have yet been established. In principle, an increased signal allows for a higher spatial resolution, especially with thinner sections at 3.0 T. This remains valid even when a slight reduction in signal-to-noise ratio is assumed when phased-array body coils are used with patients instead of the head coils used in this phantom study.
The effect of an increased spatial resolution was not tested. Signal strength, which is dependent on field strength and sensitivity of receiving coils, is not the main limiting factor but rather the increased acquisition time (to cover the same field of view with a higher matrix) that might be inappropriate for routine MR colonography in a single breath hold. The effect of longer acquisition times or of additional filtering (eg, for half-Fourier techniques) to keep acquisition times low must be tested principally for disturbances by bowel movement and breathing. The losses that result from increased artifacts must be weighted against gains from higher spatial resolution. For this purpose, studies in volunteers are required. We therefore concentrated on assessing the effect of possibly changed contrast resolution at higher field strength on polyp detection and matched the geometry of the protocol at 3.0 T to a well-established and accepted protocol used at 1.5 T.
When comparing the results of CT and MR imaging, one has to take into account that while the phantom material matches the absorption of ionizing radiation in body tissue for the CT measurement (45 HU), contrast parameters for MR imaging are more difficult to simulate. The plastic material used for the phantom has a similar T2 value to muscular or fatty tissue (10), so that data acquisitions at least at the short echo times used in this study will show a contrast and edge definition similar to that of human tissue. The T1 values of the phantom material are shorter than those of human tissues involved in the colon wall, but this will lead only to a minor deviation of signal intensity in measurements in humans and can be compensated by adjustment of flip angles.
A target lesion size of 6 mm in diameter or larger has been proposed for screening purposes with virtual colonography. Our results showed that CT and MR colonography are equally effective in the depiction of polyps in this size range, independent of spatial resolution or field strength.
Detection of polyps smaller than 6 mm in diameter mainly depends on spatial resolution and, thus, section thickness and imaging matrix. In our study, the voxel size for CT was 0.585 mm3 (0.78 x 0.78 x section thickness) for 16-section CT and 0.975 mm3 for four-section CT. MR imaging provided interpolated voxels of 1.37 x 1.37 x 1.4-mm edge length; thus the voxel volume is about 4.5 times the volume of 16-section CT and 2.7 times the volume of four-section CT. The in-plane resolution of MR imaging is about 40% lower than that of CT. Therefore, CT should be capable of reliably depicting smaller lesions. Indeed we found significant differences between the modalities for detection of lesions below the threshold of 6 mm. The smallest polyp in this study was 2 mm in diameter. Considering pixel size, the smallest polyp in our study was within the limits of given spatial resolution for either four- or 16-section CT. Therefore, increased spatial resolution, as provided at 16-section CT, showed a substantial—but not significant—effect on polyp detection for small polyps (<6 mm). However, theoretically, the faster scanning time within a breath hold (9 seconds with the chosen parameters) clinically favors 16-section CT over four-section CT. Increased signal intensity alone, as provided at MR imaging at 3.0 T, showed no significant effect on detection of polyps smaller than 6 mm compared with MR imaging at 1.5 T.
Thin-section imaging, as preferred in our study, increases image noise, which can cause image degradation. This might be one reason for the higher number of false-positive results for CT compared with MR imaging. Although polyp contour definition for lesions smaller than 6 mm was rated similarly for both CT modalities, we found 16-section CT with the minimum section thickness of 0.75 mm to be most vulnerable for interpretation errors, with an average of 0.9 false-positive result per reader. However, even though the detection rates are comparable, the number of false-positive results decreased to 50% for four-section CT. Therefore, in our opinion, a compromise between spatial resolution and image noise is required. Alternatively, smoothing algorithms may be introduced.
Although impaired image quality caused by image noise is aggravated in a low-dose setting, results of recent studies have shown low-dose thin-section CT colonography to be feasible (9,16). To compensate for the increase in image noise, we used a smoothing filter for the low-dose protocol with 10 mAs. Interestingly, low-dose CT with an effective dose of 0.9 and 1.2 mSv for male and female subjects, respectively (9), was rated equally good for detection of polyps as small as 3 mm in diameter compared with normal-dose CT and was rated superior to MR imaging. Furthermore, for polyps with a size of 3 mm, contour definition was superior to that of both MR imaging modalities. Thus, lower-dose CT colonography appears to be a viable alternative to normal-dose CT colonography and radiation-free MR colonography, especially in a screening setting, when exposure of patients to radiation is a major concern.
Our study does have limitations. Although the CT attenuation of the phantom polyps and their MR imaging signal behavior was comparable with the normal bowel wall, the model is restricted. A phantom does not exhibit bowel movement or patient movement; no fecal or fluid residues are present, which normally affect polyp detection in vivo. Furthermore, placement of the colon model in a homogenous fluid-filled body phantom will exhibit reduced scattering compared with in vivo situations (eg, the pelvic region). Our study results did not allow us to draw any conclusions about the feasibility of low-dose protocols in large or obese patients, since our phantom corresponds to a 75-kg male subject.
Simulation of contrast enhancement was not possible in our colon phantom. Polyp detection with the dark-lumen MR imaging technique, as used in our study, relies not only on the identification of endoluminal filling defects but also on the enhancement of colorectal lesions following intravenous administration of paramagnetic contrast material. Enhancement of colorectal lesions after intravenous administration of contrast material has been described for both CT and MR colonography (2,17) and was seen as a means to help differentiate medium-sized polyps from stool particles and thickened haustral folds (4). This might affect polyp detection rates, especially when two-dimensional images are used for image analysis. The image assessment in our study was performed in a semiautomated fly-through fashion by using endoluminal images. This approach relies on lesion morphology rather than on lesion enhancement. Therefore, in our opinion, the absence of contrast enhancement did not substantially affect polyp detection in our study setting.
We acknowledge discrepancies between our in vitro study and the in vivo situation as a limitation. However, our data may help to outline major differences between the tested CT and MR imaging modalities and directions and open questions for future research.
In summary, at present the depiction of colorectal polyps below a threshold diameter of 6 mm requires CT colonography. However, reader detection of polyps of potentially clinically relevant size is equal at CT and MR colonography. Increased spatial resolution as provided by 16-section CT and higher field strength as provided by 3.0-T MR imaging had no significant effect on polyp detection.
Practical applications: Low-dose multi–detector row CT colonography with thin-section imaging is feasible and therefore might be a viable alternative to MR colonography. Further in vivo studies are warranted to increase spatial resolution and thus polyp detection rates with MR colonography.
|
ADVANCE IN KNOWLEDGE |
|---|
|
TOPABSTRACTINTRODUCTIONMATERIALS AND METHODSRESULTSDISCUSSIONADVANCE IN KNOWLEDGEReferences |
|---|
|
FOOTNOTES |
|---|
Abbreviations: CI = confidence interval
Authors stated no financial relationship to disclose.
Author contributions: Guarantors of integrity of entire study, J.W., R.F., W.H.; study concepts/study design or data acquisition or data analysis/interpretation, all authors; manuscript drafting or manuscript revision for important intellectual content, all authors; manuscript final version approval, all authors; literature research, J.W., A.B., H.K.; experimental studies, J.W., R.F., A.B., H.K., T.A.; statistical analysis, J.W., A.B., N.O.; and manuscript editing, all authors
|
References |
|---|
|
TOPABSTRACTINTRODUCTIONMATERIALS AND METHODSRESULTSDISCUSSIONADVANCE IN KNOWLEDGEReferences |
|---|
This article has been cited by other articles:
|
|
 |
|
  A. H. de Vries, H. W. Venema, J. Florie, C. Y. Nio, and J. Stoker Influence of Tagged Fecal Material on Detectability of Colorectal Polyps at CT: Phantom Study Am. J. Roentgenol., October 1, 2008; 191(4): W181 - W189. [Abstract] [Full Text] [PDF]
|
|
|
|
 |
|
 I. Yamada, S. Okabe, M. Enomoto, K. Sugihara, N. Yoshino, A. Tetsumura, J. Kumagai, and H. Shibuya Colorectal Carcinoma: In Vitro Evaluation with High-Spatial-Resolution 3D Constructive Interference in Steady-State MR Imaging Radiology, December 19, 2007; (2007) 2462070128. [Abstract] [Full Text]
|
|
| ||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||
| HOME | HELP | FEEDBACK | SUBSCRIPTIONS | ARCHIVE | SEARCH | TABLE OF CONTENTS |
| RADIOLOGY | RADIOGRAPHICS | RSNA JOURNALS ONLINE |
