| HOME | HELP | FEEDBACK | SUBSCRIPTIONS | ARCHIVE | SEARCH | TABLE OF CONTENTS |
|
|
|
|||||||
|
|||||||||
Original Research |
1 Department of Radiology, Brigham and Women's Hospital, Boston, MA.
2 Present address: Department of Radiology, Lucile Packard Children's Hospital,
725 Welch Rd., Stanford, CA 94305H5654.
3 Present address: Department Radiology, Baylor College of Medicine, Houston,
TX.
4 Department of Biostatistics, Harvard School of Public Health, Boston,
MA.
Received September 15, 2006;
accepted after revision June 20, 2007.
Address correspondence to J. D. MacKenzie.
Abstract
 Top Abstract IntroductionSubjects and MethodsResultsDiscussionReferences |
|---|
SUBJECTS AND METHODS. Emergency department patients were evaluated for pulmonary embolism with standard and simulated reduced-dose CT angiography. Simulated lower-dose CT angiograms obtained at 90, 45, 22, and 10 mAseff were reconstructed by mathematical addition of noise to the standard dose (180 mAseff) data from the images of 18 patients with and 20 patients without pulmonary embolism. Four radiologists blinded to the study parameters separately interpreted each CT angiogram. Dose trends for subjective measures (diagnostic certainty, image quality, and perceived technical limitations) were evaluated, test characteristics for the detection of pulmonary embolism were computed, and clot burden was measured.
RESULTS. Readers indicated significant reductions in diagnostic certainty (p < 0.02) and image quality (p < 0.02) and an increase in perceived technical limitations (p < 0.01) as the simulated radiation dose was decreased. These subjective measures also showed significant adverse dose trends when the mAseff was reduced (p < 0.001). At reduced radiation doses, the sensitivity and positive predictive value for detection of pulmonary embolism diminished significantly. The sensitivity was 0.94 (lower bound of 0.95 CI, 0.92); specificity, 0.99 (lower bound of 0.95 CI, 0.98); positive predictive value, 0.95 (lower bound of 0.95 CI, 0.92); and negative predictive value, 0.99 (lower bound of 0.95 CI, 0.97). All patients had a low to moderate clot burden.
CONCLUSION. Reduction in dose for CT angiography in the detection of pulmonary embolism has a significant adverse effect on readers' subjective assessment of diagnostic confidence and image quality. Detection of pulmonary embolism also decreases as the tube current dose is reduced.
Keywords: CT • CT angiography • pulmonary artery • pulmonary embolism • radiation dosage
|
TopAbstractIntroductionSubjects and MethodsResultsDiscussionReferences |
|---|
From an epidemiologic perspective, an increased rate of radiation-induced malignancy is the cost of increased radiation exposure of the population, particularly the pediatric population [7, 8]. In comparison with other radiologic examinations of adults, chest CT delivers substantial amounts of radiation to the breasts [9]. Ionizing radiation is a well-established cause of breast cancer, the highest risk occurring among young women [10]. Dosimetry measurements for emergency CT examinations of the cervical spine have shown that CT delivers increased radiation to the thyroid [11]. Visualization of the pulmonary artery at least to the subsegmental level [1, 12–15] has been the major emphasis in optimizing CT protocols for the detection of PE. The aim of this study was to contribute to the growing body of literature focused on optimizing pulmonary CT angiographic protocols for PE detection through an examination of the effects of simulated dose reduction according to the as low as reasonably achievable (ALARA) principle.
Various dose-reduction strategies include increasing pitch, online modulation of tube current, and lowering tube current–time product (milliampere–second) presets [5, 16, 17]. Studying dose reduction in CT angiography is challenging because there is only one chance at timing the correct radiation dose with the IV contrast bolus. It also would be unacceptable to study the effect of dose on image quality by assembling a patient population and obtaining multiple CT angiograms of each patient. To compensate for these practical limitations, we simulated the effect of lower-dose CT protocols by mathematically adding gaussian-distributed quantum noise to the raw CT data, effectively generating CT images equivalent to images acquired at lower levels of radiation exposure [18].
This study was designed to assess reader evaluation of lower-dose CT protocols for detecting acute PE. By adding specific levels of noise to raw CT data, we explored the effects of dose reduction on image quality, diagnostic certainty, and PE detection levels. We tested the hypothesis that in comparison with standard-radiation-dose protocols, simulated reduced-dose CT protocols have diminished diagnostic certainty, imaging quality, and accuracy in detection of PE.
|
TopAbstractIntroductionSubjects and MethodsResultsDiscussionReferences |
|---|
|
CT Protocol
Imaging was performed with a 4-MDCT scanner (Somatom Plus 4, Siemens
Medical Solutions). The standard full-dose protocol was as follows: 180
mAseff at 120 kV; slice width, 1.25 mm; collimation, 1.0 mm;
rotation time, 0.5 second; suspended inspiration; caudocranial imaging from
the posterior costophrenic angles to the lung apices. Contrast material (125
mL, 370 mg I/mL) was injected IV at 3 mL/s through an 18- to 20-gauge
catheter. Imaging was triggered at 90 H with bolus tracking at the main
pulmonary artery. Current modulation to reduce dose was not used.
Image Reconstruction with Simulated Dose Reduction
Raw CT data were reconstructed by addition of gaussian-distributed quantum
noise [18] with a computer
program (syngo Explorer, Verfahren und Apparate der Medizinischen Physik) to
simulate the effect of incrementally lowering the mAseff from 180
(standard) to 90 (one-half standard), 45 (one-fourth), 22 (one-eighth), and 10
(lowest). Images sets were reconstructed to visualize soft tissues (B20f
reconstruction kernel at 1.25-mm slice thickness and 1.0-mm reconstruction
increment) and lung (B50f reconstruction kernel at 5.0-mm slice thickness,
5.0-mm reconstruction increment). The dose reconstructions were performed with
identical fields of view.
Verification of Simulated Image Noise
The relation between image noise and dose reduction and differences in
noise produced by the simulation software compared with low-dose images
obtained directly from the CT scanner were investigated by acquisition of
low-dose images of a CT phantom. Images of the phantom were acquired with the
CT scanner at 10, 14, 17, 22, 34, 45, 67, 90, 135, 165, and 180
mAseff. All other scanner settings for the phantom were equivalent
to those used to acquire patient data. The 180-mAseff phantom data
were run through the simulation software for comparison with the low-dose data
acquired from the scanner.
Study Design
The study question was defined before patients were identified and data
were collected. All identifying data were removed from the image sets (38
patients, each with one-half, one-fourth, one-eighth, and lowest simulated
dose reconstructions) and transferred to a research workstation that enabled
viewing of the axial data sets. The readers were required to interpret all
reconstructed lung and soft-tissue reconstructed images. The readers were
provided with predetermined values for soft-tissue, lung, and bone window
width and level settings, but the readers could also adjust and optimize the
width and level settings if they wished (e.g., to decide whether a filling
defect was present).
Four attending emergency department radiologists independently interpreted each CT image set. Radiologists had an average of 3 years (range, 1–5 years) of clinical practice as faculty, and one was fellowship trained in cardiothoracic imaging. All four readers were blinded to the standard and simulated dose levels during image interpretation. When interpreting the examinations, three of the four readers were blinded to the study design, reference standard results, and hypothesis. Thus three of the four readers knew they would be interpreting examinations for PE and rating image quality and diagnostic confidence. They were unaware, however, that the images had been manipulated to simulate lower-dose examinations and that the objective of the study was to evaluate the detection of PE at simulated lower doses.
CT image sets at standard and simulated lower doses were interpreted for level of certainty for the diagnosis or exclusion of acute PE, overall image quality, and number of technical factors affecting image quality. The presence or absence of PE and the distribution of PE were evaluated on the simulated lower-dose images. Images from the complete data set at each simulated dose were systematically evaluated in both soft-tissue and lung window reconstructions as if the examination were being interpreted for clinical purposes for the presence of acute PE. The pulmonary arteries were systematically evaluated from the main to at least the proximal subsegmental arteries of each lobe, when technically adequate, for the presence of acute PE. Acute PE was diagnosed if a pulmonary artery was obstructed completely by nonenhancing thrombus or if nonocclusive filling defects were apparent centrally in the vessel.
Recall bias was minimized through random ordering of the CT image sets and the requirement of at least 1 week of time sequencing between readings of image sets of the same patient. Time between patient visit and reader interpretation was a minimum of 6 months. Reader responses were recorded on a database form (Access, Microsoft) that allowed readers to select check boxes denoting the anatomic location of PE on a model of the pulmonary vasculature and to enter fixed responses from pull-down menus regarding the level of diagnostic certainty, image quality, and potential technical causes of degradation of image quality.
Data Collection
Before interpretation, the readers used consensus to establish criteria for
diagnostic certainty and image quality. Diagnostic certainty was rated high
when PE could be diagnosed or excluded with confidence, moderate when there
was some degree of uncertainty, and low when repeated CT angiography or
alternative imaging would be needed. Image quality was rated adequate when the
pulmonary artery vasculature could be confidently visualized to at least the
proximal subsegmental arteries, limited when interpretation was limited to the
segmental arteries or larger vessels, and inadequate when interpretation was
limited to the lobar pulmonary arteries or less of the vasculature. When image
quality was deemed inadequate or limited and no PE was detected, by definition
the diagnostic certainty had to be rated as moderate or low. Potential
technical causes of image quality degradation (i.e., image noise, motion
artifact, poor contrast bolus, body habitus, underlying disease in the lung
parenchyma, and artifact from a pacemaker) were recorded. For calculation of
the PE burden, the anatomic locations of pulmonary emboli were recorded at the
main, left and right main, lobar, and segmental arteries.
Data Analysis
The distributions of the subjective measures (diagnostic certainty, image
quality, and technical limitation scores) were compared between doses by use
of the nonparametric paired Wilcoxon's rank sum test. To adjust for multiple
comparisons, the p values were obtained by comparison of the observed
statistic to the null distribution of the maximum of the Wilcoxon's rank
statistics across all pairwise comparisons. The null distribution was
generated by the permutation method. A proportional odds logistic regression
model was used to test for dose trends in the subjective measures. To account
for between-reader variability, reader was included as a covariate to allow
reader-specific dose trends. The overall dose trend was summarized with the
average dose trends across four readers.
Test characteristics were computed at each dose for the overall presence of PE. The reference standard for presence or absence of PE was determined by consensus from the full-dose image data sets. Equivalence in diagnostic accuracy for the half-dose images (n = 38) was tested by construction of one-sided 95% CI for the test characteristics to assess whether the lower bound of the 95% CI was above a desired level of 95% for sensitivity, specificity, negative predictive value, and positive predictive value. The diagnostic accuracy measures were averaged over readers, and robust estimates of standard error were used to account for intrasubject correlation. Three simulated doses (one fourth, one eighth, and lowest) were pooled into a one-fourth and less dose group. Equivalence in diagnostic accuracy for the pooled one-fourth and less dose group (n =38) was also tested by construction of one-sided 95% CIs for the test characteristics.
Percentage of clot burden was quantified by calculation of the PE index, as previously described [19, 20]. Briefly, the PE index was derived from the amount and location of thrombus on CT images. The PE index is defined as the product of N x D, where N is the value of the proximal clot site equal to the number of segmental branches arising distally and D is the degree of obstruction (partial obstruction, D = 1; complete obstruction, D = 2). Because variation exists in the in the number and location of subsegmental pulmonary arteries, the location of thrombus at the subsegmental level was indicated by the segmental pulmonary artery feeding the thrombus. The maximal obstruction score for each patient was 36 (16 for the left lung, 20 for the right lung). This PE index score was converted to a percentage. The larger the percentage, the larger was the clot burden. Percentages were compared by use of the Student's t test. Statistical analysis was performed with a free statistical software package (www.r-project.org, Free Software Foundation).
|
TopAbstractIntroductionSubjects and MethodsResultsDiscussionReferences |
|---|
|
|
|
|
Although the slope of the decline in the odds ratio for high diagnostic certainty was consistent among readers, readers disagreed when judging the level of diagnostic certainty at a given dose simulation. In other words, readers were consistent in decreased diagnostic certainty according to dose trend, but each reader varied in diagnostic certainty for any given simulated dose. For image quality, the slope of the decline in the odds ratio for adequate image quality was consistent among readers, and readers agreed when judging the degree of image quality at a given dose. Overall, the agreement among the four readers was reasonable. The kappa statistics for assessing between-rater variability were 0.57 for image quality and total technical limitations and 0.69 for diagnostic certainty.
The distribution of results for PE detection at each dose-simulation level is shown in Table 3. For the overall presence of PE on the one-half dose images, the lower bounds of the 95% CI were greater than the diagnostic accuracy rate of 95% for all test characteristics (Table 4). However, in the pooled one-fourth and less dose group, the 95% CIs were lower than the diagnostic accuracy rate of 95% for sensitivity and positive predictive value. The largest number of false-negative results (a sum of 6 for all four readers) occurred in the lowest simulated-dose group (Table 3). False-positive results occurred on three separate occasions in three separate patients. A different reader made a false-positive diagnosis on each occasion. Each false-positive finding of PE was recorded at the segmental or subsegmental level at one or two locations in the lung. Factors present at the location of the false-positive result that may have contributed to the misdiagnosis included image noise, pulmonary fibrosis, atelectasis, and respiratory motion artifact.
|
|
Table 5 shows the extent and distribution of PE in the study patients. A histogram of clot burden measured with the PE index (Fig. 3) shows the association between the amount of clot burden and the number of patients with false-negative results. All patients with PE had a low to moderate level of clot burden (mean clot burden, 32%; range, 6–50%). This clot burden was significantly higher (p < 0.04) for all patients with PE (mean clot burden, 32%) than for patients with at least one false-negative result (mean clot burden, 16%). Similarly, the clot burden for patients without a false-negative result at any simulated dose level (mean clot burden, 39%) was significantly higher (p < 0.001) than the clot burden in patients who had at least one false-negative examination finding (mean, 16%). In other words, at simulated low-dose examinations, PE was more likely to be missed in patients with low clot burden than in those with greater clot burden. For validation of the dose simulation technique, the noise produced by the simulation software measured as the SD in attenuation was tightly correlated (Pearson's r = 0.99) with the noise of low-dose images acquired directly from the scanner (Fig. 4A, 4B).
|
|
|
|
|
TopAbstractIntroductionSubjects and MethodsResultsDiscussionReferences |
|---|
As tube current is reduced, diagnostic confidence and image quality decline. High diagnostic confidence is essential, particularly given the long-term clinical implications of false-positive and false-negative findings. Results obtained with the proportional odds logistic regression model suggest that the likelihood of deeming images acceptable for diagnosis decreases as radiation dose is reduced. The data also indicate that large reductions in mAseff may significantly compromise PE detection in patients with low to moderate PE burden. Greater than 50% reductions in dose resulted in an unacceptable decline in test accuracy. Although PE was detected with high sensitivity and specificity at a simulation of one half the standard mAseff, a larger sample population may also have shown significant differences in detection of PE after a 50% simulated reduction in dose.
High positive predictive value is essential in interpreting CT angiograms in evaluation for PE because a positive result most often results in treatment with anticoagulation. For the pooled one-fourth and less group, the 95% CI for positive predictive value was below the desired threshold of 95%. In this respect, dose reductions, if applicable at all, would not extend beyond 50% owing to the introduction of an unacceptable rate of error.
Multiple factors contribute to and therefore must be optimized to minimize radiation dose. Jones and Wittram [22] described technical factors, such as excessive patient motion artifact and poor contrast enhancement, that contribute to indeterminate findings at CT examinations for PE. Our data also suggest lower radiation doses may contribute to indeterminate examination findings, as implied by our finding of diminished diagnostic confidence at simulated lower doses. We found that subjective assessment of the number of technical factors increased as simulated dose was reduced. An interesting hypothesis that emerges from our data is that low-dose CT protocols for PE may synergize with a suboptimal technical factor and ultimately magnify the decline in diagnostic confidence. This synergy may lead to indeterminate and nondiagnostic examinations. Adequate images and images affected by a suboptimal technical factor are compared in Figures 5A, 5B and 6A, 6B.
|
|
|
|
Several limitations of this study are acknowledged. The dose-reduction software cannot reconstruct with adaptive filtering, which can improve image quality at lower radiation dose settings [23, 24]. The expected effect of dose reduction with adaptive filtering would be image quality better than that simulated in this study and may allow a reduction in dose with more acceptable image quality. Furthermore, z-axis and in-plane dose-reduction strategies would help limit the amount of dose received. However, we would expect a similar trend in declining diagnostic certainty, image quality, and diagnostic accuracy, as our data suggest, albeit at lower doses owing to the benefits of these dose-reduction strategies. Another limitation was that the data sets were acquired with a 4-MDCT scanner, but 16- and 64-MDCT scanners are becoming increasingly used for the evaluation for suspected PE. In addition, the readers did not have the option to acquire multiplanar reformatted images with each data set. With a standard dose, use of reformatted images may increase diagnostic confidence, but use on higher-noise images has not been tested.
Randomization of the CT image sets and time sequencing between readings of the images were used to minimize recall bias. The possibility exists that one or more of the radiologists might have recognized an examination, but none of the readers reported doing so. This study also did not address the possibility of a reader's experience in interpreting low-dose images. It is conceivable that once a reader becomes accustomed to interpreting images acquired at low doses of radiation, diagnostic confidence may increase and perhaps approach a level near that for standard-dose examinations.
We did not correlate diagnostic accuracy or subjective measures with the clinical symptoms or medical history (e.g., body mass index). General dose-reduction guidelines for tube current, generally based on weight, are becoming increasingly available [25]. Furthermore, although this study did not show a significant decline in each of the subjective measures between the full and simulated one-half and lowest-dose examinations, increasing the sample size might have shown a steady trend in decreasing diagnostic certainty and image quality and in increasing number of perceived technical limitations, as suggested by results of application of the proportional odds logistic regression model. Mayo et al. [26] found that decreasing the dose reduces the subjective assessment the quality of conventional (nonangiographic) chest CT images.
Although our findings suggest a disadvantage of dose reduction through minimizing tube current, the data do not necessarily support maximizing tube current to produce optimal CT images for interpretation. Rather, the study was preliminary, and the findings should be confirmed with a larger sample size or with animal studies. The data support further investigation to optimize dose-reduction strategies with a multipronged approach (e.g., adaptive filtering and z-axis and in-plane dose modulation) with a cautious reduction in tube current. Furthermore, minimizing technical factors such as poor bolus timing and motion artifact are important for successful implementation of low-dose MDCT protocols.
Acknowledgments
We thank the following investigators for their assistance during the
research for thismanuscript: Bernhard Schmidt (Siemens Medical Solutions),
Tomas M. Vargas (Harvard College), Elise M. Blinder and Jeffrey T. Kwasniewski
(Brigham and Women's Hospital), and Roberto Riva (Modena, Italy).
|
TopAbstractIntroductionSubjects and MethodsResultsDiscussionReferences |
|---|
CiteULike 
Complore 
Connotea 
Del.icio.us 
Digg 
Reddit 
Technorati What's this?
This article has been cited by other articles:
|
|
 |
|
  L. M. Hurwitz, T. T. Yoshizumi, P. C. Goodman, R. C. Nelson, G. Toncheva, G. B. Nguyen, C. Lowry, and C. Anderson-Evans Radiation Dose Savings for Adult Pulmonary Embolus 64-MDCT Using Bismuth Breast Shields, Lower Peak Kilovoltage, and Automatic Tube Current Modulation Am. J. Roentgenol., January 1, 2009; 192(1): 244 - 253. [Abstract] [Full Text] [PDF]
|
|
| ||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||
| HOME | HELP | FEEDBACK | SUBSCRIPTIONS | ARCHIVE | SEARCH | TABLE OF CONTENTS |
50% clot burden). Patients with low clot burden were more
likely to have PE missed at simulated low-dose examination (p <
0.04). Open bars indicate all study patients with PE. Solid bars indicate
patients with PE who had false-negative findings on at least one examination
at any simulated dose level. Each bar represents number of patients.
