HOME HELP FEEDBACK SUBSCRIPTIONS ARCHIVE SEARCH TABLE OF CONTENTS		QUICK SEARCH:   [advanced] Author: Keyword(s): Year:  Vol:  Page:

This Article Abstract Figures Only Full Text (PDF) All Versions of this Article: 2361041288v1 236/1/295    most recent Submit a response Alert me when this article is cited Alert me when eLetters are posted Alert me if a correction is posted Citation Map Services Email this article to a friend Similar articles in this journal Similar articles in PubMed Alert me to new issues of the journal Download to citation manager Citing Articles Citing Articles via HighWire Citing Articles via Google Scholar Google Scholar Articles by Kim, J.-S. Articles by Bae, K. T. Search for Related Content PubMed PubMed Citation Articles by Kim, J.-S. Articles by Bae, K. T.

Automated Detection of Pulmonary Nodules on CT Images: Effect of Section Thickness and Reconstruction Interval—Initial Results¹

¹ From the Mallinckrodt Institute of Radiology, Washington University School of Medicine, 510 S Kingshighway Blvd, St Louis, MO 63110. Received July 23, 2004; revision requested September 29; revision received October 23; accepted December 10. J.H.K. supported by a postdoctoral fellowship program from the Korea Science and Engineering Foundation (KOSEF). Address correspondence to K.T.B. (e-mail: baet{at}mir.wustl.edu).

	ABSTRACT

TOP ABSTRACT INTRODUCTION Materials and Methods Results Discussion References

 
Institutional review board approval was obtained. Informed patient consent was not required. Study was compliant with HIPAA. Performance of an automated pulmonary nodule detection program was evaluated on multi–detector row CT images that were acquired once but reconstructed retrospectively at different section thicknesses and reconstruction intervals. From raw CT data in 10 patients with pulmonary nodules, three sets of CT images were reconstructed separately in each patient by selecting two section thickness and reconstruction combinations, respectively: thin group, 1 and 1 mm; overlap group, 5 and 1 mm; and thick group, 5 and 5 mm. Nodules 3 mm in diameter and larger were detected in each group (thin group, 126 nodules; overlap group, 121 nodules; and thick group, 114 nodules) by means of consensus of two radiologists. Findings were used as the reference standard for evaluation of the computer-aided detection (CAD) program. Sensitivity and number of false-positive findings per patient by CAD were: thin group, 95.2% (120 of 126 nodules) and 5.4 findings; overlap group, 94.2% (114 of 121 nodules) and 9.7 findings; and thick group, 88.6% (101 of 114 nodules) and 23.6 findings, indicating that nodule detection degraded with increase in section thickness but improved substantially with a small reconstruction interval.

© RSNA, 2005

	INTRODUCTION

TOP ABSTRACT INTRODUCTION Materials and Methods Results Discussion References

 
With single–detector row computed tomography (CT), spiral CT images of the thorax are typically obtained with 6–10-mm section thickness. The acquisition of thin-section (ie, 1–2-mm section thickness) images of the whole thorax is impractical, because it requires multiple breath-hold sets of contiguous spiral scans to cover the thorax completely. Spatial limitations due to thick sections may be compensated for partially by means of using small reconstruction intervals that would improve nodule detection and diagnostic confidence (1). Multi–detector row CT, with its fast scanning speed and superb spatial resolution, allows us to routinely acquire thin-section images of the entire thorax in less than 10 seconds. This improvement in spatial and temporal resolution increases the sensitivity for detection of small pulmonary nodules (2). Furthermore, in multi–detector row CT, multiple spiral data are acquired during a single CT gantry rotation that allows us to generate CT images of different section thicknesses.

A main drawback of CT scanning with thin sections or small reconstruction intervals is that the size of image data sets is large. A considerable amount of interpretation time is required to review the entire set of thin-section CT images, which may impair practical implementation of screening examinations for pulmonary nodule detection. For this reason, we often intentionally generate and review CT images with 3–5-mm section thickness (which are thicker than those capable of being produced with multi–detector row CT) from the CT scan projection data. Alternatively, a computer-aided detection (CAD) system could be used as a clinical tool to help reduce the radiologist's workload and enhance the diagnostic performance of interpreting thin-section multi–detector row CT images.

Various CAD systems have been proposed and tested for the detection of lung nodules (3–8). Some of the assessments were performed on thick-section CT images (3–6) and others on thin-section CT images (7,8). The performance of CAD systems varies substantially with the quality of the images used for analysis. In general, the accuracy for nodule detection increases with thinner CT images. To our knowledge, however, no study has been reported to assess and compare the performance of a CAD system on CT images with different section thicknesses and reconstruction intervals. We believe that this type of comparative study is crucial to understand the image quality requirements for a CAD system to function as a practical and useful tool with a high degree of accuracy.

We also postulate that a CAD program, for a practical image interpretation tool, should be able to process CT images of variable section thicknesses and reconstruction intervals that are readily acquired at multi–detector row CT. Thus, the purpose of this study was to evaluate the performance of lung nodule detection with a three-dimensional (3D) morphologic matching CAD program on multi–detector row CT images that were acquired once but reconstructed retrospectively at different section thicknesses and reconstruction intervals.

	Materials and Methods

TOP ABSTRACT INTRODUCTION Materials and Methods Results Discussion References

 
Institutional review board approval was obtained. Informed patient consent was not required for our retrospective study, which involved review of previously obtained image data. Patient confidentiality was protected; our study was compliant with the Health Insurance Portability and Accountability Act.

Patients and CT Image Data
Diagnostic thoracic CT scans from 10 consecutive patients with pulmonary nodules (seven men and three women; age range, 53–89 years; mean age, 68.8 years) were reviewed retrospectively, and CT projection data were collected. These patients were different from those studied in the evaluation of a CAD program we developed (9). The patients were referred to undergo thin-section CT for the evaluation of documented (on previous CT or chest radiography studies) or clinically suspected pulmonary nodules. In selecting the study patients, we did not target any specific diagnosis or pathologic condition that was manifested by pulmonary nodules. The nodules could be associated with metastasis, primary lung cancer, or granulomas. CT studies with substantial parenchymal or pleural diseases, such as consolidation, fibrosis, and pleural effusion, were excluded from our study population.

The CT images were acquired with a multi–detector row CT scanner (Plus 4 Volume Zoom; Siemens Medical Systems, Erlangen, Germany) by using scanning parameters of 120 kVp, 120 effective mAs, 0.5-second scanning, 4 x 1-mm collimation, and a standard thin-section lung image-reconstruction kernel. From the collected CT projection data, three sets of CT images were retrospectively reconstructed separately in each of 10 patients by selecting three combinations of different section thicknesses and reconstruction intervals: (a) For the thin group, CT data sets were generated with 1-mm section thickness and 1-mm reconstruction interval; (b) for the overlap group, CT data sets were generated with 5-mm section thickness and 1-mm reconstruction interval; and (c) for the thick group, CT data sets were generated with 5-mm section thickness and 5-mm reconstruction interval. The mean number of CT images generated per patient was 274.4, with a range of 191–363 images for the thin and overlap groups and a mean of 53.8 and a range of 41–56 images for the thick group.

CAD Program
Our 3D morphologic matching CAD program that is described in detail elsewhere (9) can process CT images with different section thicknesses and/or reconstruction intervals. In brief, the section thickness and reconstruction interval information of a CT scan was obtained from the Digital Imaging and Communications in Medicine image header. After the lung region was segmented on each section, the two-dimensional segmented lung regions were stacked to generate a 3D volumetric data set of the lung region. For the data sets in the thick group, finer-resolution intermediate sections were interpolated every 1 mm by using adjacent 5-mm sections above and below the interpolation level and were integrated into the expanded 3D volumetric data set.

The reconstruction interval information was used in the process of evaluating the status of nodule candidates to adjust the parameters for geometric criteria, such as elongation factor and compactness. For the data sets in the overlap and thick groups, nodule candidates were compared first against the two-dimensional elongation factor, and any surviving candidates were compared against the 3D elongation factor. In this process, the upper limit of the 3D elongation factor was relaxed from 3 to 5, while the two-dimensional elongation factor remained at 3. The upper limit of compactness was also relaxed from 1.5 to 2. Other image processing parameter values remained unchanged between the thin, overlap, and thick groups.

These rule-based threshold values for the shape parameters were determined empirically and heuristically on the basis of a preliminary analysis of training CT image sets, which were different from the 10 cases evaluated in this study, with different section thicknesses and reconstruction intervals that contain detected pulmonary nodules. For instance, we observed in our previous phantom experiment that spherical synthetic nodules became more oblong or ellipsoidal in shape (stretched along the z-direction) as the section thickness of CT images increased (10). To adapt this apparent shape change in nodules with section thickness, we relaxed the elongation and compactness parameters in the CAD implementation when data sets from the overlap and thick groups were processed. After the implementation was complete, the CAD program could read a CT data set and automatically detect pulmonary nodules on the images.

Image Interpretation
Two radiologists (J.H.K. and K.T.B., with 14 and 12 years of experience in interpreting chest CT images, respectively) established the reference standard by means of consensus. Each reviewer first searched freely through the CT images that included the findings of pulmonary nodules detected by the CAD program and then performed the detection and documentation of pulmonary nodules. Then, the reviewers assessed together each finding on the CT images and arrived at a final decision by means of consensus if the finding was a true nodule or a false-positive finding. The number, size, and location of nodules were documented. The review process was performed separately for each of three groups in the order of the thick, overlap, and thin groups, with 2–3 week intervals between the group reviews. This order of reviews was chosen such that the detection of nodules on lower-resolution images would not be potentially assisted by the memory of subtle pulmonary nodules that are detectable only on higher-resolution images. The number of nodules detected according to the size distribution for each group was as follows: for the thin group, 55 nodules larger than 5 mm in diameter and 71 3–5-mm nodules; for the overlap group, 54 nodules larger than 5 mm and 67 3–5-mm nodules; and for the thick group, 52 nodules larger than 5 mm and 62 3–5-mm nodules.

Data and Statistical Analysis
The number, size, and location of nodules detected by the CAD program alone and by means of consensus reading of the radiologists combined with CAD detection were tabulated (J.S.K.). Sensitivity and number of false-positive findings detected with the CAD program were computed for each group by using the radiologists' consensus reading on each corresponding group's CT images as a reference standard. Statistical analyses were performed with JMP Statistical Software, version 5.1 (SAS, Cary, NC).

	Results

TOP ABSTRACT INTRODUCTION Materials and Methods Results Discussion References

 
The total number of nodules detected by the observers was 126 in the thin group, 121 in the overlap group, and 114 in the thick group (Table). The number of nodules detected by the CAD program and by means of consensus readings shows that the performance of the CAD system on images in the thin group was superior to that on images in the overlap and thick groups (Figures 1–3). With the radiologists' consensus reading on each corresponding group's CT images as a reference, the mean sensitivity of the CAD system for individual cases was 97.9% ± 3.6 for the thin group, 90.5% ± 11.8 for the overlap group, and 83.9% ± 14.9 for the thick group. The sensitivity and number of false-positive findings detected per patient in each group were 95.2% (120 of 126 nodules) and 5.4 for the thin group, respectively, 94.2% (114 of 121 nodules) and 9.7 for the overlap group, and 88.6% (101 of 114 nodules) and 23.6 for the thick group.

View larger version (145K): [in this window] [in a new window] [Download PPT slide]   Figure 1a. (a, b) Two thin-group transverse CT images from two patients with a nodule (arrow) detected in each patient. These images of 1-mm section thickness are slightly noisier (more apparent in the thoracic soft-tissue body wall) than the 5-mm-thick CT images in Figure 2, but the pulmonary vessels are more sharply delineated, particularly in b. No false-negative or false-positive nodules were observed on these two images.

View larger version (162K): [in this window] [in a new window] [Download PPT slide]   Figure 1b. (a, b) Two thin-group transverse CT images from two patients with a nodule (arrow) detected in each patient. These images of 1-mm section thickness are slightly noisier (more apparent in the thoracic soft-tissue body wall) than the 5-mm-thick CT images in Figure 2, but the pulmonary vessels are more sharply delineated, particularly in b. No false-negative or false-positive nodules were observed on these two images.

View larger version (143K): [in this window] [in a new window] [Download PPT slide]   Figure 2a. (a, b) Two transverse CT images from two patients in the overlap group with a nodule (arrow) detected in each patient. These images of 5-mm section thickness are less noisy but are slightly blurrier than the 1-mm-thick CT images in Figure 1. No false-negative or false-positive nodules were observed on these two images.

View larger version (146K): [in this window] [in a new window] [Download PPT slide]   Figure 2b. (a, b) Two transverse CT images from two patients in the overlap group with a nodule (arrow) detected in each patient. These images of 5-mm section thickness are less noisy but are slightly blurrier than the 1-mm-thick CT images in Figure 1. No false-negative or false-positive nodules were observed on these two images.

View larger version (143K): [in this window] [in a new window] [Download PPT slide]   Figure 3a. (a, b) Two transverse CT images from two patients in the thick group with missed nodules (arrow) and false-positive findings (arrowheads). On these images of 5-mm section thickness without an overlap, pulmonary vessels that were fragmented as a result of volume averaging and incomplete segmentation were falsely detected as nodules. Two true nodules that were identified with CAD on CT images in the thin and overlap groups (Figs 1, 2) were not detected (ie, false-negative findings). Although the CT images in Figures 2 and 3 have the same section thickness and look similar, the images in Figure 2 were reconstructed slightly differently from those in Figure 3 (Fig 2a at 2 mm craniad to Fig 3a, and Fig 2b at 1 mm caudad to Fig 3b). The section positions were selected to best illustrate the findings in each image group.

View larger version (151K): [in this window] [in a new window] [Download PPT slide]   Figure 3b. (a, b) Two transverse CT images from two patients in the thick group with missed nodules (arrow) and false-positive findings (arrowheads). On these images of 5-mm section thickness without an overlap, pulmonary vessels that were fragmented as a result of volume averaging and incomplete segmentation were falsely detected as nodules. Two true nodules that were identified with CAD on CT images in the thin and overlap groups (Figs 1, 2) were not detected (ie, false-negative findings). Although the CT images in Figures 2 and 3 have the same section thickness and look similar, the images in Figure 2 were reconstructed slightly differently from those in Figure 3 (Fig 2a at 2 mm craniad to Fig 3a, and Fig 2b at 1 mm caudad to Fig 3b). The section positions were selected to best illustrate the findings in each image group.

	Discussion

TOP ABSTRACT INTRODUCTION Materials and Methods Results Discussion References

With multi–detector row CT, thin-section CT of the entire thorax has become routine. Prior to the advent of multi–detector row CT, however, the acquisition of thin-section CT images of the entire thorax was impractical. Complete coverage of the thorax with 1-mm section thickness would take several breath holds and may require several sets of contiguous spiral CT scans. These prolonged scans of multiple data sets often cause omission of some anatomic levels and misregistration due to inconsistent levels of inspiration from scan to scan. Consequently, in single–detector row CT, CT images of the thorax are obtained with 6–10-mm section thickness. A common approach in spiral CT for compensating for the effect of thick sections is the use of small reconstruction intervals. CT with small reconstruction intervals improves the detection of nodules by increasing the number of images to review and the likelihood that the center of the nodule, with maximal contrast, is visualized on the images (1).

Likewise, with the reconstruction interval reduced from 5 to 1 mm in our study, the sensitivity for nodule detection increased by 6% (seven of 114) for the radiologist's reading and 13% (13 of 101) for the CAD system. More important, the number of false-positive findings detected per patient with the CAD program declined markedly from 23.6 to 9.7 with a 1-mm reconstruction interval.

Our study and the studies of others have shown that the performance of the CAD system depends highly on the quality of the images and likely on the CAD algorithm. In our study, use of CAD analysis on CT images in the thin group resulted in the highest sensitivity for nodule detection with the fewest false-positive findings. This trend was observed in previously published studies. Brown et al (8) assessed their model-based CAD program on selected thin-section single–detector row CT image sets (1-mm collimation, 0.5–1.0-mm reconstruction intervals, 20-mm scan length) and reported 100% sensitivity for detection of nodules larger than 3 mm in diameter with 15 false-positive findings per case. It is noteworthy that the CT images in their study represented subvolumes, not the entire thorax.

Qian et al (7) tested their CAD system on thin-section multi–detector row CT images (1.25-mm section thickness, 1-mm reconstruction interval) and obtained 87.1% sensitivity for solid nodules 3 mm and larger, with 2.5 false-positive findings per case. In contrast, a CAD system applied to CT images of 10-mm section thickness (10- or 7-mm reconstruction interval) by Armato et al (4) resulted in a lower sensitivity with a high false-positive detection rate: an overall nodule detection sensitivity of 70% and 1.5 false-positive findings per section (mean of 28 sections per case). A similar result, 72% sensitivity with 31 false-positive findings per case, was reported by Lee et al (6) on CT images with 10-mm section thickness. A CAD study performed by Ko and Betke (5) on CT images of 5- or 10-mm section thickness resulted in 91% sensitivity for detection of nodules larger than 3 mm in diameter, with 2.3 false-positive findings per image.

With thin-section CT of the thorax, CAD systems will become a practical necessity and will likely achieve an acceptable sensitivity and false-positive detection rate to be a clinically useful tool. Our current CAD algorithm was initially designed and developed to process thin-section images and was modified later to operate on thick-section images. We believe that if the algorithm was designed from its inception to process thick-section images, the performance of the CAD program would be better than our current results. As shown in previously published CAD studies (4–6), however, intrinsic limitations of thick-section images for achieving a high accuracy of nodule detection may not be overcome simply by applying sophisticated CAD algorithms.

On thick-section CT images, if nodules are visible faintly across neighboring sections, their 3D nodule features may not be characterized adequately. As a result, the discrimination ability of the nodule classifier will be compromised substantially. The lack of connectivity between the sections, combined with volume averaging, leads to a high rate of false-positive findings on images in the thick group. For example, blood vessels or bronchial walls that were fragmented as a result of volume averaging were classified into small nodules. This misclassification would be avoided if the 3D features of nodule candidates were characterized sufficiently on contiguous sections. When overlapped images with a small reconstruction interval are available, some of the 3D features of nodules could be retrieved to help improve the performance of the CAD system, as shown in our study. We are continuously investigating different algorithmic approaches to reduce the number of false-positive findings on thick-section CT images.

There are several limitations to our study. First, the number of cases in our data set may be too small to arrive at definitive conclusions. We may need a larger data set for the generalization of our results. However, the preliminary results from our study appear in good agreement with the results reported in previous clinical and CAD studies in which the detection rates of pulmonary nodules were assessed on CT images acquired with different section thicknesses and reconstruction intervals.

Second, we did not address any quality difference between 1-mm- and 5-mm-thick CT images. Although they are from the same raw projection data and the same scan, the 1-mm-thick CT images are likely noisier than the 5-mm-thick CT images because of the reduced number of photons per image. This difference in image noise profile would potentially affect the 1-mm-thick CT images by increasing false-positive detection rates in small pulmonary nodules, particularly on low-radiation-dose CT scans.

Third, the determination of nodules relied on consensus readings of two experts. No histologic verification or follow-up study was performed. The problem of defining "truth" is critical and challenging in the assessment of CAD or clinical performance of nodule depiction (12,13). Fourth, thresholds and criteria for the geometric features for classifying nodules were determined empirically and heuristically. In the current implementation, no parametric analysis was performed to optimize or to evaluate valid ranges of the values we used in our study. Alteration of the thresholds and criteria would affect the sensitivity and false-positive detection rate in our results.

Finally, the images in our study contained multiple nodules. The multiplicity of nodules per case may introduce a bias in the analysis of CAD performance. As an attempt to address this problem, we have computed and reported the mean and standard deviation of the sensitivity per case, as well as the sensitivity for the pooled nodules divided by the number of the cases. The sensitivities calculated with either approach for each of three image groups were similar.

In summary, our study demonstrated that the detection of nodules with a CAD system improved with a decrease in both section thickness and reconstruction interval of multi–detector row CT images. A CAD system tested on thick-section CT images without overlapped sections resulted in a substantially higher number of false-positive findings and thus may not be acceptable in a practical clinical setting. The accuracy for detection of nodules was markedly enhanced with the use of a small reconstruction interval.

	FOOTNOTES

 

Abbreviations: CAD = computer-aided detection • 3D = three dimensional

² Current address: Korea Advanced Institute of Science and Technology, Daejeon, South Korea

³ Current address: Department of Radiology, Choongnam National University, Daejeon, South Korea

See also the article by Bae et al in this issue

Authors stated no financial relationship to disclose.

Author contributions: Guarantors of integrity of entire study, J.S.K., J.H.K., K.T.B.; study concepts and design, all authors; literature research, J.S.K., J.H.K., K.T.B.; clinical studies, J.H.K., K.T.B.; data acquisition and analysis/interpretation, all authors; statistical analysis, J.S.K., J.H.K.; manuscript preparation, definition of intellectual content, editing, revision/review, and final version approval, all authors

	References

TOP ABSTRACT INTRODUCTION Materials and Methods Results Discussion References

 

1. Buckley JA, Scott WW Jr, Siegelman SS, et al. Pulmonary nodules: effect of increased data sampling on detection with spiral CT and confidence in diagnosis. Radiology 1995; 196:395–400.[Abstract/Free Full Text]
2. Fischbach F, Knollmann F, Griesshaber V, Freund T, Akkol E, Felix R. Detection of pulmonary nodules by multislice computed tomography: improved detection rate with reduced slice thickness. Eur Radiol 2003; 13:2378–2383.[CrossRef][Medline]
3. Giger M, Bae KT, MacMahon H. Computerized detection of pulmonary nodules in computed tomography images. Invest Radiol 1994; 29:459–465.[CrossRef][Medline]
4. Armato SG 3rd, Giger ML, MacMahon H. Automated detection of lung nodules in CT scans: preliminary results. Med Phys 2001; 28:1552–1561.[CrossRef][Medline]
5. Ko JP, Betke M. Chest CT: automated nodule detection and assessment of change over time—preliminary experience. Radiology 2001; 218:267–273.[Abstract/Free Full Text]
6. Lee Y, Hara T, Fujita H, Itoh S, Ishigaki T. Automated detection of pulmonary nodules in helical CT images based on an improved template-matching technique. IEEE Trans Med Imaging 2001; 20:595–604.[CrossRef][Medline]
7. Qian J, Fan L, Wei G, et al. Knowledge-based automatic detection of multi-type lung nodules from multi-detector CT studies. SPIE 2002; 4684:689–697.[CrossRef]
8. Brown MS, Goldin JG, Suh RD, McNitt-Gray MF, Sayre JW, Aberle DR. Lung micronodules: automated method for detection at thin-section CT—initial experience. Radiology 2003; 226:256–262.[Abstract/Free Full Text]
9. Bae KT, Kim JS, Na YH, Kim KG, Kim JH. Pulmonary nodules: automated detection on CT images with morphologic matching algorithm—preliminary results. Radiology 2005; 236:286–294.[Abstract/Free Full Text]
10. Goo JM, Tongdee T, Tongdee R, Yeo K, Hildebolt CF, Bae KT. Volumetric measurement of synthetic lung nodules with multi–detector row CT: effect of various image reconstruction parameters and segmentation thresholds on measurement accuracy. Radiology 2005; 235:850–856.[Abstract/Free Full Text]
11. Paranjpe DV, Bergin CJ. Spiral CT of the lungs: optimal technique and resolution compared with conventional CT. AJR Am J Roentgenol 1994; 162:561–567.[Abstract/Free Full Text]
12. Armato SG 3rd, McLennan G, McNitt-Gray MF, et al. Lung image database consortium: developing a resource for the medical imaging research community. Radiology 2004; 232:739–748.[Abstract/Free Full Text]
13. Austin JH, Muller NL, Friedman PJ, et al. Glossary of terms for CT of the lungs: recommendations of the Nomenclature Committee of the Fleischner Society. Radiology 1996; 200:327–331.[Free Full Text]

This article has been cited by other articles:

				R. Iezzi, A. R. Cotroneo, A. Filippone, F. Di Fabio, M. Santoro, and M. L. Storto MDCT Angiography in Abdominal Aortic Aneurysm Treated with Endovascular Repair: Diagnostic Impact of Slice Thickness on Detection of Endoleaks Am. J. Roentgenol., December 1, 2007; 189(6): 1414 - 1420. [Abstract] [Full Text] [PDF]

				P. Peloschek, J. Sailer, M. Weber, C. J. Herold, M. Prokop, and C. Schaefer-Prokop Pulmonary Nodules: Sensitivity of Maximum Intensity Projection versus That of Volume Rendering of 3D Multidetector CT Data Radiology, May 1, 2007; 243(2): 561 - 569. [Abstract] [Full Text] [PDF]

This Article Abstract Figures Only Full Text (PDF) All Versions of this Article: 2361041288v1 236/1/295    most recent Submit a response Alert me when this article is cited Alert me when eLetters are posted Alert me if a correction is posted Citation Map Services Email this article to a friend Similar articles in this journal Similar articles in PubMed Alert me to new issues of the journal Download to citation manager Citing Articles Citing Articles via HighWire Citing Articles via Google Scholar Google Scholar Articles by Kim, J.-S. Articles by Bae, K. T. Search for Related Content PubMed PubMed Citation Articles by Kim, J.-S. Articles by Bae, K. T.