| HOME | HELP | FEEDBACK | SUBSCRIPTIONS | ARCHIVE | SEARCH | TABLE OF CONTENTS |
|
|
|
|||||||
|
|||||||||
Thoracic Imaging |
1 From the Department of Radiology, Vienna Medical University, Währinger Gürtel 18-20, A-1090 Vienna, Austria (P.P., J.S., M.W., C.J.H.); Department of Radiology, Utrecht Medical Center, Utrecht, the Netherlands (M.P.); and Department of Radiology, Academic Medical Center, University of Amsterdam, Amsterdam, the Netherlands (C.S.). Received December 16, 2005; revision requested February 10, 2006; revision received May 30; accepted June 20; final version accepted September 5. Supported by grant P17083-N04 from the Austrian Funds for Scientific Research. Address correspondence to P.P. (e-mail: philipp.peloschek{at}meduniwien.ac.at).
 |
ABSTRACT |
|---|
 TOP ABSTRACT INTRODUCTIONMATERIALS AND METHODSRESULTSDISCUSSIONADVANCES IN KNOWLEDGEReferences |
|---|
Materials and Methods: This institutional review board–approved prospective study included 20 oncology patients (eight women and 12 men; mean age, 56 years ± 16 [standard deviation]) who underwent clinically indicated standard-dose thoracic multidetector CT and provided informed consent. Transverse thin slabs of the chest (thickness, 7 mm; reconstruction increment, 3.5 mm) were created by using MIP and VR techniques to reconstruct CT data (collimation, 16 x 0.75 mm) and were reviewed in interactive cine mode. Mean, minimum, and maximum reading time per examination and per radiologist was documented. Three radiologists digitally annotated all nodules seen in a way that clearly determined their locations. The maximum number of nodules detected by the three observers and confirmed by consensus served as the reference standard. Descriptive statistics were calculated, with P < .05 indicating a significant difference. The Wilcoxon matched-pairs signed rank test and confidence intervals for differences between methods were used to compare the sensitivities of the two methods.
Results: VR performed significantly better than MIP with regard to both detection rate (P < .001) and reporting time (P < .001). The superiority of VR was significant for all three observers and for nodules smaller than 11 mm in diameter and was pronounced for perihilar nodules (P = .023). Sensitivities achieved with VR ranged from 76.5% to 97.3%, depending on nodule size.
Conclusion: VR is the superior reading method compared with MIP for the detection of small solid intrapulmonary nodules.
© RSNA, 2007
|
INTRODUCTION |
|---|
|
TOPABSTRACTINTRODUCTIONMATERIALS AND METHODSRESULTSDISCUSSIONADVANCES IN KNOWLEDGEReferences |
|---|
Maximum intensity projection (MIP) and volume rendering (VR) represent two commercially available techniques for displaying a subvolume of the 3D data set. Results of previous studies (13,18,19) have already proved the superiority of MIP over the reading of transverse sections for the detection of pulmonary nodules. The VR technique represents a more advanced 3D volume reconstruction method that is based on the assignment of opacity values to CT numbers: High opacity values produce an appearance similar to surface rendering, and low opacity values allow the user to "see through" structures. MIP is confined to the display of voxels with the maximum intensity found along a projection line from the viewer's eye through the 3D volume of interest (3,13,18–20). In addition, VR leads to an apparent enlargement of structures with an absorption higher than that of the surrounding tissues, depending on the opacity values applied.
We hypothesized that transverse thin-slab VR would be more sensitive in the detection of intrapulmonary nodules than MIP by improving the spatial orientation of the reporting radiologist and virtually increasing the size of pulmonary nodules. Thus, the purpose of our study was to prospectively compare MIP and VR for the detection of small intrapulmonary nodules at multidetector CT.
|
MATERIALS AND METHODS |
|---|
|
TOPABSTRACTINTRODUCTIONMATERIALS AND METHODSRESULTSDISCUSSIONADVANCES IN KNOWLEDGEReferences |
|---|
Patients and Nodule Distribution
We performed a prospective analysis of clinically indicated CT data. All 65 patients (37 men and 28 women; mean age, 62 years ± 21 [standard deviation]; age range, 35–87 years) included in the study underwent chest CT for clinical indications. Reasons for referral included detection of metastasis, as well as follow-up of known pulmonary metastasis. The study protocol was approved by the internal review board (ethics committee) of the Vienna Medical University, and written informed consent was obtained from all patients.
Patient inclusion was determined by a chest radiologist who was not involved in the reading process (M.P., with 15 years of experience in thoracic imaging). Patient inclusion was not consecutive owing to the application of exclusion criteria (Fig 1); patients were selectively included in the study group if CT showed between one and 20 intrapulmonary nodules that were smaller than 2 cm in diameter. We aimed for a sample size of between 160 and 180 nodules, as was suggested before the study by the results of a power analysis (nQuery Adviser, version 5.0; Statistical Solutions, Cork, Ireland), to ensure a power of 80% for detecting a medium effect size of 
= .5 at a significance level of P < .05, according to Cohen (21). Therefore, patient inclusion was terminated when more than 160 isolated intrapulmonary nodules had been detected.
|
Finally, eight women and 12 men (mean age, 56 years ± 16; age range, 35–71 years) were included; the total number of isolated intrapulmonary nodules, as detected by three observers and confirmed by consensus, was 163. Four of the 20 patients had up to two nodules, 10 patients had between three and 10 nodules, and six patients had between 10 and 20 nodules. The distribution of lesions with regard to size was as follows: 51 (31.3%) of the 163 nodules were 5 mm or smaller, 75 (46.0%) were between 6 and 10 mm, and 37 (22.7%) were larger than 10 mm in diameter.
Nodules were grouped according to their location by one of the authors who was not involved in the reading process (M.P). Nodules classified as perihilar were located in a 2-cm zone around the hilum; peripheral nodules were defined as those within 1 cm of the pleural surface, including the space along the cardiomediastinal silhouette; and the remaining nodules located in the core of the parenchyma were classified as central. The final distribution included 39 (24%) perihilar nodules, 67 (41%) peripheral nodules, and 35 (35%) central nodules. This distribution of nodules was similar to data reported in previous studies (5,13).
Technical Parameters
Images were obtained with a subsecond 16-section multidetector CT scanner (Somatom Sensation; Siemens Medical Solutions) in a single breath hold at 120 kV, 100 mAs (effective), and a collimation of 16 x 0.75 mm. Scanning was performed from the lung apices to the upper abdomen. An intravenous contrast agent (iopromide, Ultravist; Schering, Berlin, Germany) was administered (100 mL [300 mg iodine per milliliter]) with an injection rate of 4 mL/sec. Scans were acquired with a delay of 40 seconds. In our institution, we routinely administer intravenous contrast material in oncology patients known to have or suspected of having pulmonary metastases because of the resulting improved anatomic orientation in the mediastinum and interpretation of potential intrapulmonary consolidations. We did not expect contrast material use to have an effect on pulmonary nodule detection.
Thin sections (width, 1 mm) were reconstructed with a reconstruction increment of 0.75 mm by using a high-resolution reconstruction algorithm with standardized window level and width settings for the lung parenchyma (window level, –650 HU; window width, 1500 HU). Scanning length varied between 30 and 35 cm and resulted in 300–350 transverse images per patient.
Three-dimensional Transverse Thin-Slab Reconstructions
The 20 data sets included in the study group were sent to a dedicated workstation (Leonardo; Siemens Medical Solutions) with software (Syngo LungCare; Siemens Medical Solutions). This software allows the real-time 3D processing of sliding thin-slab MIP and VR reconstructions without user interaction.
Thin-slab 7-mm-thick MIPs were reconstructed from the 1-mm transverse sections; MIP slabs were reconstructed at an increment of 3.5 mm. Thin-slab VR images with a 7-mm section thickness and a 3.5-mm reconstruction increment were also reconstructed from the 1-mm transverse sections. In our study, 50% opacity was set at –400 HU, and voxels with an attenuation of 200 HU or greater were assigned 100% opacity.
Image Interpretation
Data sets were reviewed at the dedicated workstation by using the commercially available software, which allowed for real-time reconstruction and cine-mode display of thin-slab VR and MIP images. Three radiologists independently analyzed the MIP and VR data sets. Observers A and C (J.S. and P.P., each with 5 years of experience) were trained in general body imaging, and observer B (C.S., with 15 years of experience) was specialized in thoracic imaging.
All observers were familiar with the appearance of MIP and VR images and with the workstation used. Studies were randomly presented with regard to patient order and to the order of the technique used (MIP or VR). No observer saw both MIP and VR images in a particular patient within the same reading session. To avoid memory effects, we ensured that at least 4 weeks elapsed between reading sessions.
An isolated intrapulmonary nodule was defined as a round opacity completely surrounded by lung with a ratio of less than 2 between its maximum and minimum diameter (19).
Criteria and methods for reporting had been defined before the study, and observers were trained before study initiation by evaluating a training data set obtained in five additional patients with a total of 37 intrapulmonary nodules that were not included in the final data analysis.
Observers were not informed about the number of lesions per patient and were asked to annotate each nodule they saw in a way that clearly determined its location.
Statistical Analysis
Our study lacked an absolute standard of truth because histopathologic proof was not available, and we are aware of no other imaging method that provides performance superior to CT for the detection of intrapulmonary nodules. We therefore aimed to compare the sensitivities of the two 3D techniques rather than to determine absolute sensitivity. The maximum number of nodules detected by all three readers and confirmed by consensus served as the reference standard. We further classified the nodules into three subcategories according to their size and location, as described above. We also documented the mean, minimum, and maximum reading time per examination and per radiologist.
Statistical analysis was performed with software (SPSS for Windows, version 11.5.1, SPSS, Chicago, Ill; EGRET, version 2.0.31, Cytel Software Corporation, Cambridge, Mass). The significance level was set at P < .05.
The following tests were performed:
1. First, we calculated the sensitivity of the two 3D techniques on a nodule-by-nodule basis: A Wilcoxon matched-pairs signed rank test assessed nodule by nodule whether the number of observers who had detected that nodule significantly differed between the two 3D modes. In this test, the nodule size was the dependent variable and the 3D mode was the independent variable. The significance level was set at P < .05. A two-way analysis was performed to also consider the influence of nodule location. In that test, the nodule size was the dependent variable and nodule location and 3D technique were the independent variables.
2. We then calculated the sensitivity of the two 3D techniques on a reader-by-reader basis. A significant difference was indicated when zero was not included within the bounds of the 95% confidence intervals.
3. We applied the McNemar test to determine whether one 3D technique yielded a significantly higher number of nodules that were seen exclusively with that 3D technique.
4. In addition, a logistic regression analysis with random effects was performed for each reader separately to compare the sensitivity of the two methods, taking multiple nodules per patient into account.
5. Interobserver agreement was determined in a pairwise fashion by comparing observers A and B, observers B and C, and observers A and C in turn. The significance of differences was tested by calculating the 95% confidence intervals.
6. The efficiency of the two 3D techniques, with regard to required reading time, was determined by calculating the mean, minimum, and maximum time needed to read the examinations. Reading times were averaged over the three observers for MIP and VR. Data were compared with a two-way analysis of variance with repeated measures, with time being the dependent variable and readers and viewing technique being the independent variables.
|
RESULTS |
|---|
|
TOPABSTRACTINTRODUCTIONMATERIALS AND METHODSRESULTSDISCUSSIONADVANCES IN KNOWLEDGEReferences |
|---|
|
|
|
|
|
|
|
|
|
DISCUSSION |
|---|
|
TOPABSTRACTINTRODUCTIONMATERIALS AND METHODSRESULTSDISCUSSIONADVANCES IN KNOWLEDGEReferences |
|---|
|
We compared MIP and VR in our study and found a superior detection performance for VR over MIP. In addition, the number of nodules seen exclusively with one technique was higher for VR. VR techniques display the entire volume of data—not just the maximum intensity voxels—and, therefore, convey relevant information in a complex anatomic context better than MIP techniques. VR algorithms assign relative opacity values from 0% to 100% to each voxel along a line from the viewer's eye through the data set. It is noteworthy that, with VR, opacity settings have an effect on the apparent size of structures in the image: Higher opacity values make objects look larger, and lower opacity values make them look smaller (25). In our study setup, pulmonary nodules (>40 HU) were assigned high opacity values (>70%). Although the apparent increase in diameter based on the assigned opacity value may not have been of considerable magnitude, it may have contributed to the superior performance of VR.
In addition to being affected by the surrounding lung parenchyma, the detection of pulmonary nodules is also dependent on lesion properties, such as size and location. Our results indicate that VR is superior to MIP only for nodules smaller than 10 mm. For larger nodules (>10 mm), we could not find a significant performance difference, most likely due to the fact that evaluation of these lesions is less prone to visual perception errors. Nodules in a perihilar location are more difficult to detect than more peripherally located nodules owing to the close vicinity to the central vascular structures (1,13,23). Our results suggest that the superiority of VR over MIP is more pronounced for perihilar nodules. This is not surprising, considering the superior capability of VR compared with MIP in demonstrating the 3D spatial relationship in the two-dimensional display.
The magnitude of sensitivities we found with MIP (45.1%–91.9%), is in accord with data published by Diederich et al (4), who reported sensitivity ranges of 36%–67% for nodules smaller than 5 mm, 72%–89% for nodules from 6 to 10 mm, and 91%–100% for lesions larger than 10 mm in diameter. The sensitivity for nodule detection with VR, however, was substantially higher in our study (76.5%–97.3%). These numbers are comparable to the detection rates achievable with double reading of transverse images, which were recently reported to amount to 74%–79% for small (size, 3.9 mm ± 3.2) pulmonary nodules (9).
We decided to evaluate 7-mm-thick slabs reconstructed with a 3.5-mm increment. Slab thickness and reconstruction increment were somewhat arbitrarily determined, driven by the goal to enable smooth cine viewing and an optimal display of lesions of interest and overlying anatomy. In contrast to previous studies (13,19), in which slabs of 10–30 mm thickness were evaluated, we used a slightly smaller slab thickness.
Further studies are warranted to evaluate the performance of VR in association or in comparison with computer-aided detection algorithms for pulmonary nodule detection. Most previous investigators assessing the effect of computer-aided detection evaluated performance differences of computer-aided detection compared with reading of transverse sections (15–17,26). Our data suggest that there is a substantially increased performance of the human visual system when 3D displays rather than transverse sections are used. Researchers who evaluate the additional benefit of computer-aided detection should consider the potentially best display technique as a reference. We found that for that specific application, VR is superior to MIP.
Average reading time with VR (4.5 minutes) was significantly shorter than with MIP (5.4 minutes). These times include the time required for digital annotation of each localized nodule, which explains the increased reporting time compared with experience in routine clinical work. The reading time did not substantially exceed the times reported in previous studies (13). The shorter reading times with VR may serve as an indirect sign of the superior spatial orientation for the readers with this technique.
The major limitation of our study was the fact that there was no absolute reference standard for the number of nodules with a method of histopathologic proof or a different, but superior, detection method. Therefore, the reference standard in this study was the maximum number of nodules detected by three readers and confirmed by consensus. Because the aim of our study was not to determine the absolute sensitivity of a particular technique but rather to compare the performance of two methods, we think that this limitation did not affect our results.
Our decision to restrict patient recruitment to only those patients referred for evaluation of possible metastatic disease represents a selection bias in favor of well-defined, solid nodules. We specifically excluded focal opacities adjacent to pleural surfaces and airways that did not meet the morphologic criteria of a round, solid lesion. We did so to minimize the discussion of what was or was not a potentially malignant nodule and accepted this limitation to achieve more constrained data in order to compare the two display techniques. Although we expected a superiority of VR over MIP for more complex focal lesions such as ground-glass nodules and nodules with irregular borders, we did not test this. We therefore must limit our conclusion to the fact that VR was superior to MIP for the depiction of solid round opacities.
Another limitation was the fact that we did not determine specificities. To determine sensitivities, we accepted the consensus of all readers as a reference standard. Because it seems even more difficult to assess the presence of potentially false-positive lesions without the presence of an absolute reference standard, we did not determine specificity values. By including only patients with one or more nodules seen on transverse images or MIPs, we missed the opportunity to find patients with one or more nodules seen exclusively on VR images, where the detection of nodules would have changed patient care. It has, however, to be noted that we focused our study solely on detection performance and did not include potential interpretative errors or effects on patient care. We also did this because our study lacks an absolute standard of histopathologic truth. From previous publications, it is known that even detected nodules may be misinterpreted if potentially abnormal features are not sufficiently recognized. This interpretation error can lead to a further increase in misdiagnosis: Of 38 missed lung cancers in a screening setting, 23 were missed owing to detection errors, and 15 were missed owing to interpretation errors (27). Because we did not aim to assess the effects on patient care, we did not include patients without a known nodule and performed a data analysis on nodule-by-nodule and reader-by-reader bases but not on a patient-by-patient basis.
Our results pertain to the particular scanning and reconstruction parameters that we used. We did not evaluate the performance of variable slab thicknesses or the effect of low-dose acquisition on VR image quality.
We conclude that VR displays are useful as an additional diagnostic review of thoracic multidetector CT data sets to highlight small solid intrapulmonary nodules in patients examined for the diagnosis of lung metastases or bronchogenic carcinoma.
|
ADVANCES IN KNOWLEDGE |
|---|
|
TOPABSTRACTINTRODUCTIONMATERIALS AND METHODSRESULTSDISCUSSIONADVANCES IN KNOWLEDGEReferences |
|---|
|
FOOTNOTES |
|---|
Abbreviations: MIP = maximum intensity projection • 3D = three-dimensional • VR = volume rendering
See Materials and Methods for pertinent disclosures.
Author contributions: Guarantor of integrity of entire study, P.P.; 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, P.P., C.J.H., M.P.; clinical studies, P.P., J.S., C.S.; statistical analysis, P.P., J.S., M.W., M.P., C.S.; and manuscript editing, all authors
|
References |
|---|
|
TOPABSTRACTINTRODUCTIONMATERIALS AND METHODSRESULTSDISCUSSIONADVANCES IN KNOWLEDGEReferences |
|---|
This article has been cited by other articles:
|
|
 |
|
  N. Kawel, B. Seifert, M. Luetolf, and T. Boehm Effect of Slab Thickness on the CT Detection of Pulmonary Nodules: Use of Sliding Thin-Slab Maximum Intensity Projection and Volume Rendering Am. J. Roentgenol., May 1, 2009; 192(5): 1324 - 1329. [Abstract] [Full Text] [PDF]
|
|
|
|
 |
|
 J. Vikgren, S. Zachrisson, A. Svalkvist, A. A. Johnsson, M. Boijsen, A. Flinck, S. Kheddache, and M. Bath Comparison of Chest Tomosynthesis and Chest Radiography for Detection of Pulmonary Nodules: Human Observer Study of Clinical Cases Radiology, December 1, 2008; 249(3): 1034 - 1041. [Abstract] [Full Text] [PDF]
|
|
|
|
|
|
F. Girvin and J. P. Ko Pulmonary Nodules: Detection, Assessment, and CAD Am. J. Roentgenol., October 1, 2008; 191(4): 1057 - 1069. [Abstract] [Full Text] [PDF]
|
|
| ||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||
| HOME | HELP | FEEDBACK | SUBSCRIPTIONS | ARCHIVE | SEARCH | TABLE OF CONTENTS |
| RADIOLOGY | RADIOGRAPHICS | RSNA JOURNALS ONLINE |
