| HOME | HELP | FEEDBACK | SUBSCRIPTIONS | ARCHIVE | SEARCH | TABLE OF CONTENTS |
|
|
|
|||||||
|
|||||||||
1 Department of Medical Radiology, Institute of Diagnostic Radiology, University
Hospital Zurich, Rämistrasse 100, Zurich CH-8091, Switzerland.
2 Present address: Department of Radiology, Kantonsspital, Loestrasse 170, Chur
CH-700, Switzerland.
Received November 3, 2003;
accepted after revision January 18, 2004.
Address correspondence to T. Boehm
(thomas_boehm{at}gmx.net).
Abstract
 Top Abstract IntroductionSubjects and MethodsResultsDiscussionReferences |
|---|
SUBJECTS AND METHODS. Retrospective ECG-gated MDCT and non–ECG-gated MDCT of the chest were performed in 25 patients. ECG-gated MDCT data were reconstructed mid diastole using a fixed interval of –400 msec in 25 patients and then additionally at –200, –300, and –500 msec in 10 of those patients. Shaded surface display and volume rendering of the bronchial tree combined with virtual bronchoscopy were performed using all data sets. The extent of bronchial tree visualization in shaded surface display–virtual bronchoscopy and volume rendering–virtual bronchoscopy and the presence of artifacts in volume-rendered images were scored by three blinded reviewers. The effective radiation doses of the ECG-gated and nongated acquisitions were compared.
RESULTS. The summary scores of all bronchial segments for gated shaded surface display–virtual bronchoscopy and gated volume rendering–virtual bronchoscopy did not differ significantly. The summary scores for nongated shaded surface display–virtual bronchoscopy and nongated volume rendering–virtual bronchoscopy were not significantly different. Non-gated acquisition yielded significantly better visualization of the bronchial tree for both post-processing techniques, regardless of the time interval used for reconstruction of the ECG-gated series. Artifact scores in volume-rendered images were significantly higher for ECG-gated MDCT compared with nongated MDCT. Effective radiation dose was significantly higher for the ECG-gated acquisition.
CONCLUSION. Given the advantage of volume rendering for representing the entire data set and given the lower radiation dose and better 3D image quality of nongated acquisition, volume rendering performed on nongated MDCT data is the method of choice for 3D visualization of the bronchial tree.
|
TopAbstractIntroductionSubjects and MethodsResultsDiscussionReferences |
|---|
Virtual bronchoscopy is a reconstruction technique based on shaded surface display or on volume-rendering techniques, which render axial CT data into simulated endobronchial views [4]. This alternative imaging technique has been shown to be feasible for the diagnosis of various airway abnormalities [4–7].
Currently, most virtual bronchoscopies are based on shaded surface display, which requires relatively little computing power and is not time-consuming. Shaded surface display generates triangle-based isosurfaces in 3D space and is based on threshold segmentation [8].
Volume rendering also allows the visualization of objects in three dimensions. Volume rendering of CT data differentiates tissues on the basis of Hounsfield units and assigns different colors, opacity, and brightness attributes to the separated volumes, thus yielding overlay displays. The main difference between shaded surface display and volume rendering is that volume rendering preserves density information. Furthermore, in contrast to shaded surface display, volume rendering displays the complete volume of the data, whereas shaded surface display reduces the amount of data by replacing the density information with a surface model [9].
Artifacts caused by cardiac motion are not only confined to the heart and to the central blood vessels, but also affect the adjacent lung. These artifacts may decrease 3D image quality and additional efforts in suppressing such artifacts may be justified.
Retrospective ECG gating is a relatively new method that is based on the simultaneous acquisition of CT data and the ECG signal. The acquired projectional raw data are selected for image reconstruction with respect to a predefined cardiac phase (i.e., with a certain temporal relation to the peak of the R waves in ECG) [10]. Retrospective ECG-gated helical scanning provides continuous and fast volume coverage with an overlapping image increment with the possibility of individual reconstructions in every part of the cardiac cycle [11]. Many studies have shown the advantages of ECG gating in modern CT imaging [10, 12–15]. It can be thus hypothesized that this type of artifact suppression could be useful for 3D purposes. On the other hand, we were able to show in a recent study that ECG gating did not improve the quality of 3D visualization of the bronchial tree when using shaded surface display [16], despite a significantly better signal-to-noise ratio in the ECG-gated CT images.
To the best of our knowledge, a direct comparison of shaded surface display and volume rendering regarding 3D visualization of the bronchial tree and virtual bronchoscopy has not yet been published. Similarly, the influence of ECG gating on volume-rendered image quality has not been previously evaluated. The aims of this study were therefore to compare the 3D visualization of the bronchial tree on the same CT data sets using both shaded surface display and volume rendering and to assess the influence of ECG gating on volume-rendered image quality.
|
TopAbstractIntroductionSubjects and MethodsResultsDiscussionReferences |
|---|
|
Imaging Technique
All CT scans were obtained with a 4-MDCT scanner (Somatom VolumeZoom,
Siemens Medical Solutions). First, a scout view of the thorax was used to plan
CT data acquisition, and then a 20-mL test bolus of iodixanol (Visipaque 320,
Amersham Health) was administered IV at a rate of 3 mL/sec using a power
injector via an 18-gauge catheter placed in the cubital vein to measure the
optimal time delay for data acquisition. Contrast material (150 mL) was
injected at a rate of 3 mL/sec using the optimal delay of 20–30 sec. The
use of contrast material was necessary only for assessing coronary artery
bypass graft patency and for ruling out aortic dissection. For generating 3D
models of the bronchial tree and for virtual bronchoscopy, the administration
of contrast material would not be required. The CT data set was acquired
craniocaudally during the patient's inspiratory breath-hold. The imaging
volume extended from the proximal supraaortic vessels to the apex of the
heart. For ECG-gated data acquisition, a coronary standard protocol was used:
tube voltage, 120 kV; tube current, 300 mAs; collimation, 4 x 2.5 mm;
pitch, 0.38; slice thickness, 3 mm; increment, 1.5 mm. The patient's ECG was
recorded simultaneously with the scanning data. For image reconstruction in
mid diastole, a fixed interval of –400 msec before the onset of the next
R wave was used [17].
Segmented adaptive cardiac volume reconstruction based on a half rotation
(180°, 250 msec) reconstruction technique was used, which provided
improved temporal resolution using segmented reconstruction techniques while
maintaining a high z-resolution
[18]. In this technique,
depending on the patient's heart rate, two reconstruction algorithms were
applied: a single-segmental reconstruction (
65 beats per minute) that
required data from only one half-rotation and an adaptive two-segmental
reconstruction (> 65 beats per minute) that required data from at least two
half-rotations [19]. For the
non–ECG-gated series, a 360° (500 msec) reconstruction algorithm was
used; tube current was set to 160 mAs and pitch was 1.2. The remaining
acquisition parameters were the same as for the ECG-gated series.
Radiation Dose
To compare the radiation exposure, we calculated effective doses using a
commercially available computer program (WinDose version 2.1a,
Scanditronix-Wellhöfer Dosimetrie)
[20]. The calculations of
effective radiation dose in this software program are based on Monte Carlo
calculations for anthropomorphic mathematic phantoms that were obtained by the
GSF National Research Center for Environment and Health (Neuherberg, Germany)
[20]. By entering different
scanning parameters including collimation, pitch, kerma, tube current, tube
voltage, scanning range, anatomic area, and the patient's sex, the software
program provided an estimate of the effective radiation dose
[20].
3D Image Reconstruction
Axial CT data from the ECG-gated and nongated series were transferred to an
Octane SGI workstation (Silicon Graphics) with commercially available 3D
reconstruction software (ProVision version 2.2, Algotec Systems). Shaded
surface display models were reconstructed and were used for simultaneous
shaded surface display visualization of the bronchial tree and for shaded
surface display–virtual bronchoscopy
(Fig. 1A). Image segmentation
for shaded surface display was based on thresholding. The threshold was set
such that the endobronchial air-to-bronchial wall interface was segmented
(window width, 310 H; window center, –1,000 H).
|
For volume-rendering postprocessing, all CT data were transferred to a VolumeZoom Wizard workstation (Siemens Medical Solutions). Volume-rendering models were reconstructed with the same threshold values as for shaded surface display and used for simultaneous volume-rendering visualization of the bronchial tree and for volume rendering–virtual bronchoscopy (Fig. 1B).
|
Navigation through the tracheobronchial tree based on shaded surface display and volume-rendering data was performed in the fly-through mode beginning in the trachea and ending in the most distal bronchus that could be visualized. Navigation was done interactively using the computer mouse and was aided by simultaneous multiplanar reformation display of the current intrabronchial position. A fly-through video clip was prepared for reviewing images.
Rating of Bronchial Tree Visualization
Shaded surface display and volume rendering and the corresponding virtual
bronchoscopy flythrough video clips were presented in random order to three
experienced independently working radiologists. The reviewers were unaware of
patient data, clinical history, and CT acquisition technique.
The shaded surface display–virtual bronchoscopy and volume rendering–virtual bronchoscopy readout was aimed at quantifying the degree of visualization of peripheral bronchi. Cases in which smaller and more distally located bronchi were accessible in virtual bronchoscopy were rated higher than cases in which only central bronchi were accessible. In cases with discrepant ratings for the 3D visualization and the corresponding virtual bronchoscopy (i.e., access to the more peripherally located parts of the bronchial tree was hindered by an artifact in virtual bronchoscopy, but the distal parts were well visualized with volume rendering or shaded surface display), the rating for virtual bronchoscopy was recorded. Rating of bronchial segment visualization was performed separately for the upper, middle–lingula, and lower lobes. The following scores were applied: 0, not visualized; 1, visualized up to right or left main bronchus; 2, upper lobe, middle lobe, or lower lobe bronchi; 3, lingula or segmental bronchus; 4, fourth-generation bronchi; and 5, fifth-generation bronchi (Fig. 2).
|
Effects of the Different Diastolic Reconstructions
For the 10 patients assessed for follow-up of coronary artery bypass
grafts, the projectional raw data had been stored. After rating of the initial
study with only one mid diastolic image reconstruction at –400 msec
before the R wave and obtaining unexpected results (e.g., the nongated
technique resulted in better 3D visualization of the bronchial tree than the
gated mid diastolic series), we performed additional reconstructions at
–200, –300, and –500 msec before the R wave to assess the
influence of different diastolic reconstructions on image quality. The shaded
surface display and volume-rendering reconstructions with virtual bronchoscopy
were prepared similarly to the initial mid diastolic series. All shaded
surface display reconstructions with virtual bronchoscopy (e.g., shaded
surface display–virtual bronchoscopy for the nongated series and for the
ECG-gated series at –200, –300, –400, and –500 msec
before the R wave) were then presented to the same three radiologists. Image
analysis was performed in the same way as for the mid diastolic
reconstructions. The reviewers were unaware of the patient data, clinical
history, MDCT acquisition technique, and reconstruction interval. Because of
the limited number of patients, we decided to compare only the summary scores
for all bronchial segments. For comparison with the nongated images, the gated
series with the best rating was used. The results were compared with the
results of the initial image analysis (e.g., –400 msec before the R wave
and nongated) in the 10 patients.
Assessment of Volume-Rendered Image Quality
Rating of 3D image quality was performed using the data sets acquired at
–400 msec before the R wave in two steps: assessment of stairstep
artifacts in the main bronchi and assessment of artifacts in the lung
parenchyma.
The image analysis aimed at quantifying the stairstep artifacts in the left and right main bronchus separately. The following scores were used: 1, well-defined bronchial wall without stairstep artifacts; 2, small regularly distributed stairstep artifacts measuring 1 mm or less; 3, large irregularly distributed stairstep artifacts greater than 1 mm, causing significant distortion of the anatomic image; and 4, severely distorted bronchial wall with real anatomy not sufficiently represented by the image.
A rating was performed to quantify the distribution of artifacts in the lung parenchyma for the right and left lungs separately. The following scores were applied: 1, no artifacts; 2, minimal artifacts randomly distributed in the image (not grouped, evenly distributed over the image, or repeating in a certain spatial frequency) causing no difficulties in assessing the bronchial tree; 3, distinct artifacts affecting all parts of the image (evenly distributed over the image or repeating with a certain spatial frequency) without limiting assessment of the bronchial tree; and 4, heavy artifacts limiting assessment of the bronchial tree.
Statistical Analysis
For statistical analysis, SPSS version 11.0 (Statistical Package for the
Social Sciences) for Windows (Microsoft) was used. The mean values of every
bronchial segment score and the bronchial segments' sums for every reviewer
were calculated. Wilcoxon's signed rank test was used to compare the mean
values between ECG-gated and nongated CT. This test was also used to compare
stairstep artifacts in the main bronchi and effective radiation doses between
the two CT techniques. Interobserver agreements (kappa statistics) were
calculated for shaded surface display–virtual bronchoscopy quality
assessments. Level of agreement were interpreted as poor (
= 0), slight
( = 0.01–0.20), fair ( = 0.21–0.40), moderate
( = 0.41–0.60), good ( = 0.61–0.80), and almost
perfect ( = 0.81–1.00)
[21].
|
TopAbstractIntroductionSubjects and MethodsResultsDiscussionReferences |
|---|
Radiation Dose
The mean effective radiation dose (± standard deviation [SD]) was
11.3 ± 1.7 mSv for the ECG-gated series and 4.6 ± 1.1 mSv for
the nongated MDCT series (p < 0.005).
Volume Rendering–Virtual Bronchoscopy
Comparison between shaded surface display–virtual bronchoscopy
and volume rendering–virtual bronchoscopy using nongated CT
data.—Visualization of the segmental and subsegmental bronchi in
the lower lobes was significantly better when using volume
rendering–virtual bronchoscopy (right lower lobe, p < 0.002;
left lower lobe, p < 0.002). The left upper lung segment was
better visualized (p < 0.013) with shaded surface display using
nongated data. Comparison of summary scores for all bronchial segments
revealed nonsignificant differences in visualization of the bronchial tree
between shaded surface display–virtual bronchoscopy and volume
rendering–virtual bronchoscopy (p = 0.18). Mean values of the
summary scores using nongated data amounted to 20.6 ± 0.2 and 21.2
± 0.3 for shaded surface display–virtual bronchoscopy and volume
rendering–virtual bronchoscopy, respectively.
Comparison between shaded surface display–virtual bronchoscopy and volume rendering–virtual bronchoscopy using ECG gating.—No significant differences were found between shaded surface display–virtual bronchoscopy and volume rendering–virtual bronchoscopy either in the scores for the single segments or in the scores for the sums of all bronchial segments. Mean values of the summary scores using ECG-gated data were 19.2 ± 0.2 and 20.0 ± 0.3 for shaded surface display–virtual bronchoscopy and volume rendering–virtual bronchoscopy, respectively.
Comparison between ECG-gated and non-gated data using volume rendering–virtual bronchoscopy.—A significantly better visualization of single bronchial segments was achieved in the right lower lobe (p < 0.03) and the left lower lobe (p < 0.04) using a nongated acquisition. The remaining lung segments, with the exception of the left upper lobe, were better depicted with the non-gated technique. However, these differences were not statistically significant. The summary scores for all bronchial segments amounted to 20.0 ± 0.3 and 21.2 ± 0.3 for volume rendering–virtual bronchoscopy using ECG-gated and nongated MDCT data, respectively, and were significantly higher for the nongated MDCT than for the ECG-gated technique (p < 0.04).
Bronchial Tree Visualization
Kappa statistics.—The mean interobserver agreement (kappa
statistics) for the three reviewers of the bronchial segments was 0.7 (good)
for shaded surface display and 0.8 (good) for volume rendering.
Effects of diastolic reconstructions on bronchial tree visualization using volume rendering–virtual bronchoscopy.—Assessment of bronchial tree visualization using volume rendering–virtual bronchoscopy in the nongated series resulted in a mean summary score of 21.4 ± 0.2 per patient. The summary scores for the four additional diastolic reconstructions at –200, –300, –400, and –500 msec were 19.6 ± 0.2, 19.8 ± 0.3, 19.9 ± 0.3, and 18.9 ± 0.3, respectively. The summary scores for the reconstruction at –200 to –400 msec were not significantly different, whereas the summary score for the reconstruction at –500 msec was significantly lower compared to the scores for the other diastolic reconstructions. When comparing the ECG-gated reconstruction yielding the highest score for a particular patient with nongated MDCT, the mean summary score per patient was 19.9 ± 0.3. The difference between nongated and ECG-gated MDCT (when using the best gated series) was statistically significant (p < 0.004)— that is, nongated CT performed significantly better than ECG-gated CT.
Assessment of Volume-Rendered Image Quality
Stairstep artifacts in the left and right main
bronchi.—Using nongated data, we found that stairstep artifacts
were more frequent and significantly more pronounced in the left main bronchus
(2.0 ± 0.4) compared to the right main bronchus (1.4 ± 0.5,
p < 0.001). When using ECG-gated data reconstructed at –400
msec, there were also significantly more stairstep artifacts in the left main
bronchus (2.9 ± 0.7) than in the right main bronchus (2.5 ± 0.7,
p < 0.002).
Stairstep artifacts were scored significantly higher using ECG gating than in the nongated acquisition for both sides (p < 0.001).
Artifacts in the lung parenchyma.—No difference occurred in the number and severity of artifacts between the right and left lung parenchyma for ECG-gated data sets (right and left, 2.6 ± 0.8) and for nongated data (right and left, 1.6 ± 0.5). Nongated data sets (1.6 ± 0.5) showed significantly lower arti-fact scores compared with ECG-gated data (2.6 ± 0.8, p < 0.001). Figure 3A, 3B, 3C, 3D, 3E, 3F shows a synopsis of shaded surface display, volume rendering, and multiplanar reformation images of a patient with no bronchial pathology. Multiple bandlike artifacts that are only seen in the ECG-gated volume-rendered images were found (Fig. 3D).
|
|
|
|
|
|
|
TopAbstractIntroductionSubjects and MethodsResultsDiscussionReferences |
|---|
In addition to the results of the previous study [16], we are now able to show that 3D imaging of the bronchial tree is not improved by cardiac gating with either postprocessing technique (shaded surface display or volume rendering). This result did not change when computing different diastolic reconstructions (i.e., using R wave anticipations of –200, –300, –400, and –500 msec) and using the data with the best results for comparison with the non-gated series. These results are of clinical importance because ECG-gated acquisition results in a significantly higher radiation dose to the patient compared to nongated CT.
No difference was detected between shaded surface display–virtual bronchoscopy and volume rendering–virtual bronchoscopy in visualizing peripheral bronchi when using an ECG-gated data set. When using nongated data, we found that volume rendering performed significantly better in both lower lobes, but shaded surface display was superior in the left upper lobe. These data are difficult to interpret. We suppose that volume-rendering segmentation may be more susceptible to contrast material inflow artifacts, which would explain its performance in the upper lobe. The use of contrast material was necessary for the main purpose of these examinations (i.e., assessment of cardiac bypasses), but in fact the application of contrast material was not necessary for generating 3D models of the bronchial tree and virtual bronchoscopy. However, in clinical practice, 3D models of the bronchial tree are mostly reconstructed from standard contrast-enhanced chest CT data sets. Further studies are needed to evaluate the influence of contrast material on the image quality of 3D visualization of the bronchial tree.
On the other hand, volume rendering may be less susceptible to cardiac motion artifacts, which are more pronounced in the lower lobes. Nevertheless the summary scores for all bronchial segments were not significantly different between shaded surface display and volume rendering when using nongated data. We therefore conclude that volume rendering does not perform better for 3D depiction of the bronchial tree than shaded surface display.
Both postprocessing techniques have additional advantages and shortcomings that should be taken into account. The shaded surface display technique optimally renders depth impression, requires minimal computing power, and saves time. Nevertheless, the pitfalls are numerous and can be classified into three categories [23]. These categories are threshold range, slice thickness (either responsible for partial volume or stairstep artifacts), and motion artifacts during data acquisition (either because of the patient's inability to maintain strict apnea or because of organ pulsation). As with some patients of this study, setting a threshold in shaded surface display entails the potential inclusion or exclusion of details that may produce artifacts or omit relevant anatomic or pathologic structures [24]. Pitfalls related to the threshold are also common with volume rendering. It shows various shades of anatomy not by absolute thresholding, but by gradation of thresholding (trapezoid representations of a gray or color scale), with the additional capability of changing brightness and opacity [25].
A significant advantage of volume rendering over shaded surface display is that volume rendering does not lose a large amount of data in the final reconstruction [26]. This argument may be considered to be one in favor of volume rendering, especially if bronchial and extrabronchial disease should be visualized simultaneously.
The unexpected results regarding the effects of ECG-gated acquisition remain to be explained. Two groups of factors may be responsible for the better performance of non-gated acquisition. First, different signal-to-noise ratios of the two CT acquisition techniques may cause differences in 3D image quality. However, in our previous study [16], the signal-to-noise ratio was determined for the same data sets and showed a slight but significant difference in favor of the ECG-gated technique. Second, different reconstruction algorithms and algorithm-specific artifacts may be the reason for the differences in 3D image quality. The heart rate in all patients was within the limits advised by the manufacturer of the CT scanner. Therefore, excessive heart rate cannot be blamed for the differences. The reconstruction algorithms used for ECG-gated CT are based on the patient's heart rate, which may vary during data acquisition. Depending on the current heart rate, CT data from one or two heart cycles are used for reconstruction to enhance temporal resolution [19]. These heart rate–dependent changes in reconstruction may lead to a certain craniocaudal inhomogeneity of the CT data set. However, the same algorithms work well in the imaging of the coronary arteries, the heart valves, and the ascending aorta. All these applications are characterized by a high contrast between the contrast-enhanced vessels and the adjacent tissues. Therefore, technique-related artifacts of such CT studies may not be detrimental because of the high signal-to-artifact ratio.
Our hypothesis is that certain artifacts in the ECG-gated series may be responsible for the poorer performance, despite the reduction of motion artifacts. We therefore evaluated the ECG-gated and nongated volume-rendered images systematically in a blinded manner for the presence and extent of artifacts. The summary scores clearly showed a higher prevalence of these artifacts in the ECG-gated series. In addition, significantly more stairstep artifacts were detected at the level of the tracheal bifurcation using the ECG-gated technique. We therefore suggest that algorithm-related artifacts are the cause of the inferior performance of ECG-gated reconstructions.
In conclusion, the shaded surface display–virtual bronchoscopy and volume rendering–virtual bronchoscopy techniques visualized peripheral bronchi to a similar degree. Non-gated CT together with any of the postprocessing techniques performed better than the ECG-gated counterpart. Given the additional advantages, such as presentation of the whole data set with volume rendering and reduced radiation dose with the nongated CT acquisition, we recommend volume rendering performed on nongated CT data as the method of choice for virtual bronchoscopy and 3D visualization of the bronchial tree.
Acknowledgments
We thank Burkhardt Seifert for contributions to the statistical
analyses.
|
TopAbstractIntroductionSubjects and MethodsResultsDiscussionReferences |
|---|
CiteULike 
Complore 
Connotea 
Del.icio.us 
Digg 
Reddit 
Technorati What's this?
| ||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||
| HOME | HELP | FEEDBACK | SUBSCRIPTIONS | ARCHIVE | SEARCH | TABLE OF CONTENTS |
