Radiology
HOME HELP FEEDBACK SUBSCRIPTIONS ARCHIVE SEARCH TABLE OF CONTENTS
QUICK SEARCH:   [advanced]


     

Published online before print June 26, 2006, 10.1148/radiol.2401050727
This Article
Right arrow Abstract Freely available
Right arrow Figures Only
Right arrow Full Text (PDF)
Right arrow All Versions of this Article:
2401050727v1
240/2/537    most recent
Right arrow Submit a response
Right arrow Alert me when this article is cited
Right arrow Alert me when eLetters are posted
Right arrow Alert me if a correction is posted
Right arrow Citation Map
Services
Right arrow Email this article to a friend
Right arrow Similar articles in this journal
Right arrow Similar articles in PubMed
Right arrow Alert me to new issues of the journal
Right arrow Download to citation manager
Right arrow reprints & permissions
Citing Articles
Right arrow Citing Articles via Google Scholar
Google Scholar
Right arrow Articles by Plathow, C.
Right arrow Articles by Kauczor, H.-U.
Right arrow Search for Related Content
PubMed
Right arrow PubMed Citation
Right arrow Articles by Plathow, C.
Right arrow Articles by Kauczor, H.-U.
(Radiology 2006;240:537-545.)
© RSNA, 2006

Technical Developments

Quantification of Lung Tumor Volume and Rotation at 3D Dynamic Parallel MR Imaging with View Sharing: Preliminary Results1

Christian Plathow, MD, MSc, Max Schoebinger, MSc, Christian Fink, MD, Holger Hof, MD, Jürgen Debus, MD, PhD, Hans-Peter Meinzer, PhD, MSc and Hans-Ulrich Kauczor, MD

1 From the Departments of Radiology (C.P., C.F., H.U.K.) and Medical and Biological Informatics (M.S., H.P.M.), German Cancer Research Center, Heidelberg, Germany; Department of Diagnostic Radiology, Eberhard-Karls University, Hoppe-Seyler 3, 72076 Tuebingen, Germany (C.P.); Department of Diagnostic Radiology, Ludwig Maximilian University, Munich, Germany (C.F.); and Department of Radiation Therapy, University of Heidelberg, Heidelberg, Germany (H.H., J.D.). Received May 2, 2005; revision requested June 23; revision received June 30; accepted August 2; final version accepted September 14. Address correspondence to C.P. (e-mail: c.plathow{at}dkfz.de, christian.plathow{at}med.uni-tuebingen.de).


   ABSTRACT
TOP
 ABSTRACT
INTRODUCTION
 MATERIALS AND METHODS
 RESULTS
 DISCUSSION
 ADVANCE IN KNOWLEDGE
 References
 
The purpose of this study was to prospectively evaluate the volumes and rotations of pulmonary nodules during respiration by using three-dimensional fast low-angle shot dynamic magnetic resonance (MR) imaging (1.5/0.6 [repetition time msec/echo time msec], 3.8 x 3.8 x 3.8-mm voxel size, imaging time per three-dimensional data set of 1 second). The feasibility of the technique was verified by using 130-, 40-, and 12-cm3 phantoms made of meatballs and in five patients with solitary intrapulmonary tumors (four men, one woman; median age, 60 years) at computed tomography and histologic analysis. All patients provided written informed consent, and the study was institutional review board approved. It was proved that there were no substantial differences among the 21 algorithms used to correct partial volume effects. The most precise algorithm (r > 0.9, P < .01) used to correct partial volume effects—with which mean phantom volumes of 120.8 cm3 ± 4.1, 36.1 cm3 ± 3.98, and 13.1 cm3 ± 1.5 were calculated—yielded a root mean square error of 14%. The MR imaging–derived nodule volume and rotation during respiration could be quantified by using oriented bounding box techniques.

© RSNA, 2006


   INTRODUCTION
TOP
 ABSTRACT
 INTRODUCTION
MATERIALS AND METHODS
 RESULTS
 DISCUSSION
 ADVANCE IN KNOWLEDGE
 References
 
With the development of diagnostic tools such as multisection computed tomography (CT), increasing numbers of small pulmonary nodules are being found incidentally and radiologists are being faced with the challenge of estimating the probability of the malignancy of these nodules with use of noninvasive methods (1). With one approach, changes in nodule size are measured during follow-up imaging examinations; however, this method requires excellent reproducibility. A 1-mm increase in the cross-sectional diameter of a 10-mm nodule corresponds to a 10% increase in diameter and a 33% increase in volume. Such a change observed at two examinations performed 3 months apart corresponds to a doubling time consistent with malignancy.

It is easier to detect an increase in the volume than an increase in the diameter of a spherical nodule. This fact led Yankelevitz et al (2) to propose using segmentation software with static CT images to identify small growing pulmonary nodules. This technique has been investigated in several CT studies (3–5). To our knowledge, nodule segmentation performed by using dynamic imaging techniques during respiration and the influence of respiration on tumor volume have not been investigated because adequate pre- and postprocessing techniques are not available. However, this information is of great importance not only in helping to correctly determine the nodule volume during follow-up but also in performing high-precision radiation therapy techniques (6–8). Information about potential tumor rotations during the respiratory cycle is important in these applications as well—for example, for optimizing the collimator position and rotation (9).

Dynamic magnetic resonance (MR) imaging is an approach for visualizing respiratory motion with high spatial and temporal resolution (10–12). A significant correlation between dynamic MR imaging findings and lung function test results has been shown (13). This technique was recently applied to measure tumor mobility in two dimensions (14). Because of the limited temporal resolution of conventional MR imaging techniques, the continuous three-dimensional (3D) documentation of a tumor during the entire respiratory cycle that serves as the basis for detecting volumetric and rotational changes had not been investigated.

The use of parallel MR imaging techniques, such as sensitivity encoding and simultaneous acquisition of spatial harmonics, enables the acquisition of a substantially reduced sampled quantity of k-space data. This results in increased temporal resolution without trade-offs in spatial resolution and anatomic coverage (15–17). View sharing is an alternative k-space sampling method in which some parts of the k-space are updated more frequently than others. This results in a shortened total acquisition time. Recently, the use of parallel imaging combined with view sharing in time-resolved MR angiography of the lung has been proposed (18). Thus, the objective of this study was to prospectively evaluate the volumes and rotations of pulmonary nodules during respiration by using 3D dynamic MR imaging.


   MATERIALS AND METHODS
TOP
 ABSTRACT
 INTRODUCTION
 MATERIALS AND METHODS
RESULTS
 DISCUSSION
 ADVANCE IN KNOWLEDGE
 References
 
MR Imaging
All MR examinations were performed by using a clinical 1.5-T whole-body imaging unit (Magnetom Symphony; Siemens Medical Solutions, Erlangen, Germany) equipped with eight receiver channels and a high-performance (30 mT/m) gradient system. A six-channel coil system was used for signal reception. Dynamic MR imaging was performed by using an isotropic time-resolved 3D fast low-angle shot pulse sequence in which parallel imaging was combined with view sharing (18). For dynamic imaging of the respiratory cycles, the following parameters were used: 1.5/0.6 (repetition time msec/echo time msec), 10° flip angle, 1500 Hz/pixel receiver bandwidth, generalized autocalibrating partially parallel acquisition, acceleration factor of three, 24 reference lines for calibration, view sharing in four regions, 375 x 400-mm field of view, 77 x 128 matrix, 198-mm slab thickness, 52 partitions, and a voxel size of 3.8 x 3.8 x 3.8 mm. Twenty consecutive 3D data sets were acquired, resulting in an imaging time of 20 seconds.

Phantoms
For validation of the 3D dynamic MR imaging segmentation and selection of the appropriate volumetry algorithm (description to follow), three meatballs of different volumes—130, 40, and 12 cm3, each verified on the basis of the volume of the meatball in water—served as nodule surrogates (ie, phantoms [Fig 1]) (19,20). The phantoms were centrally fixed on a wooden board and manually moved (by M.S.; speed controlled by C.P.) 1 cm per second in the y direction, in and out of the MR imaging unit, during the MR imaging measurements, in accordance with changes in the apicodiaphragmatic distance in patients performing a slow respiratory maneuver (13). These measurements were performed three times. Therefore, 180 measurements (three phantoms times three measurements times one image per second times 20 seconds) were performed.


Figure 1
View larger version (51K):
[in this window]
[in a new window]
[Download PPT slide]
  		Figure 1: Photographs of tumor phantoms (far left) and tumor surface renderings depicted on every fifth image (second to far right images). One image was obtained every second for 20 seconds. t1, t5, t10, t15, t20 = 1, 5, 10, 15, and 20 seconds, respectively.

 
Patients
The volume and rotation of lung tumors during the respiratory cycle were evaluated in five patients (four men, one woman; median age, 60 years; age range, 49–68 years) with histologically confirmed stage I non–small cell lung cancer. All patients had been smokers. After histologic verification of the tumor, no treatment had been administered before the diagnostic imaging examination. After the nature of the procedure had been fully explained to these patients, they all provided written informed consent for participation in our human research protocol, which was approved by the institutional review board of Ruprecht-Karls University Heidelberg. Correlation of the radiologic and histologic findings was possible for three patients, who underwent lobectomy. The remaining two patients were treated with radiation therapy because of their poor health status. The excised lung tumors were embedded in paraffin after being fixed in 4% formalin (buffered in phosphate-buffered saline). The surrounding normal lung tissue was carefully resected, and the tumor volume was measured. The intervals between MR imaging, CT, and histologic examination did not exceed 10 days.

The patients were instructed to perform various respiratory maneuvers, as follows: They were instructed to maximally expire before imaging and then to slowly inspire to the maximal possible volume (within 5 seconds). After this exercise, they were instructed to once again expire slowly (Fig 2). Thus, the MR imaging measurements started during maximal expiration and continued to encompass both maximal inspiration and maximal expiration. A slow respiratory maneuver was performed to account for the low temporal resolution of one 3D data set per second. To ensure reproducibility, the patients were allowed to rehearse performing these maneuvers several times (13,15).


Figure 2
View larger version (58K):
[in this window]
[in a new window]
[Download PPT slide]
  		Figure 2: Top: Fast low-angle shot 3D MR images (1.5/0.6) of respiratory cycle from maximal inspiration to maximal expiration, with a time resolution of one image per second, obtained in a 68-year-old man with a history of smoking. Bottom: Three-dimensional surface renderings of the segmented lung and tumor volumes.

 
Thin-Section CT
CT imaging was performed during inspiration by using a standard CT scanner (Somatom Plus 4 Volume Zoom; Siemens Medical Solutions). Thin-section (1-mm collimation, 4 x 1.00-mm detector array, pitch of 1.5, rotation time of 0.5 second per rotation, 137 kV, 110 mAs) and contrast material–enhanced (3-mm collimation, 4 x 3.00-mm detector array, pitch of 1.5, rotation time of 0.5 second per rotation, 137 kV, 180 mAs; with 80 mL of iomeprol [Imeron 300; Altana, Konstanz, Germany]) CT images were acquired, per the routine in our radiology department. The acquisition field of view ranged from 290 to 390 mm, depending on the patient's size and shape. The acquisition matrix was 512 x 512.

Image Analyses
Analysis of the 3D data sets involved the following steps: First, the 3D image data sets of the phantoms were segmented. Second, to measure the segmented phantom volumes and compare these results with the real volumes of the phantoms, mathematical algorithms to correct partial volume effects were investigated. Third, 3D dynamic MR imaging was performed in the patients to quantify tumor volume and rotation. In this step, the dynamic MR imaging findings were compared with the CT and histologic findings.

Phantoms and Segmentation
With use of MR imaging, an experienced radiologist (C.P., 5 years segmentation experience) segmented the imaged phantoms as follows: Each voxel was classified as phantom or nonphantom surrounding tissue. This manual segmentation served as the basis on which the software tools quantified the imaged phantom volumes.

The phantoms were segmented interactively by using a combination of manual and semiautomatic segmentation tools (eg, interactive region-growing techniques; CHILI, Heidelberg, Germany) (21). The procedure was applied to each time frame of the volume sequence. Segmentation inconsistencies between neighboring sections and successive time frames were minimized by means of morphologic image processing. Three-dimensional and, optionally, four-dimensional median filtering was applied to reduce these intra– and inter–time frame segmentation irregularities with use of a structuring element with a radius of 1 voxel in both cases (22). The segmentation of one image data set took about 30 seconds. Therefore, each phantom required about 5 minutes for segmentation. The two-dimensional segmentation results were recombined to obtain a voxel-based 3D-over-time description of the phantom during motion.

Volumetry
The most frequent method of measuring volume is that of multiplying the number of voxels included in the segmentation by the volume of a single voxel. This technique was used in the quantification of the CT images, as described before (23). This is an obvious approach for many applications. However, if the spatial resolution rendered by the imaging device is low and/or the object surface–to–object volume ratio is disadvantageous—as in the case of small nodular volumes depicted at MR imaging—then the partial volume effect will substantially influence the volume measurement. Consistent with Kuhnigk et al (23), we divided the tumor volume into three areas: the tumor core, the parenchymal area, and the partial volume area. In addition, we subdivided the partial volume area into a segmented partial volume (SPV) region and a non-SPV region, as shown in Figure 3. To measure a volume corrected for partial volume effects, the SPV and non-SPV area voxels were weighted according to the properties derived from the tumor core or the parenchymal area.


Figure 3
View larger version (53K):
[in this window]
[in a new window]
[Download PPT slide]
  		Figure 3: Illustration of different areas used for partial volume correction (24,25).

 
We used a set of 21 algorithms (26) (Table 1), which include two classes of available volumetry methods: binary methods, in which only the information obtained from the segmentations is used in the volume calculation process, and image-based methods, with which the gray values of the tumor also are taken into consideration. Binary volumetry methods can be subdivided again into two groups: In the first group, all voxels have a weighting of one. A volume is calculated by using the tumor core voxels (method B-1), the tumor core plus SPV area voxels (method B-2), and the tumor core plus SPV area plus non-SPV area voxels (method B-3) (Fig 3). Thus, method B-2 is equivalent to the commonly used volumetry approach. In the second binary method group, partial volume voxels are weighted on the basis of the number of adjacent voxels. Again, three approaches can be distinguished: (a) Only SPV area voxels are weighted (method 1), (b) only non-SPV area voxels are weighted (method 2), and (c) both SPV and non-SPV area voxels are weighted (method 3). In addition, two types of neighborhoods (ie, the sets of voxels considered to be neighboring the currently processed voxels) are used (six or 26 neighborhoods). These approaches are referred to as methods B6-1 to B6-3 and methods B26-1 to B26-3.


View this table:
[in this window]
[in a new window]

  		Table 1. Relative Deviations of Phantom Volume, Measured by Using 21 Algorithms, from Real Phantom Volume

 
With the image-based volumetry methods, the volume of a partial volume voxel is weighted on the basis of two properties derived from the segmented object: the mean of the gray values and the standard deviation of this mean. If the gray value of a partial volume voxel is exactly the same as the mean gray value of the segmented object, the volume is weighted by one. If the gray value of a partial volume voxel differs more than a given number times the standard deviation of the gray values of the segmented object, the volume is weighted by zero. To derive a weighting factor for a given gray value between these two extreme values, different interpolation methods may be used. In this study, linear and exponential interpolations were used. These methods were named according to the following variables: the gray value with either linear interpolation (Glin) or exponential interpolation (Gexp); the variation factor (V), by which the standard deviation of the mean gray value of the objects is multiplied (to define the range of gray values, which will be weighted), with a factor of 2 (V2) or 3 (V3) corresponding to deviations of approximately 66% or 99%, respectively, of all gray values within the tumor; and the method type (1, 2, or 3). As an example, method GlinV2-1 involved the use of linear interpolation, a standard deviation factor of 2, and method 1.

The volumes of the segmented 3D dynamic MR images of the phantoms during motion and of the tumors (in the patients) during respiration were determined. For all measurements, the discrepancies between the measured and true volumes were calculated and expressed as absolute errors and errors relative to the phantom size. For each algorithm, the set of 180 error values was tested for significant deviations from zero.

In the patients, the MR imaging and CT tumor segmentations performed during inspiration were compared. With CT, tumors were segmented automatically (C.P.) by using a threshold-based region-oriented method to delineate the nodule boundaries, and the tumor volume was calculated by using the resulting segmented representation.

Orientation and Rotation
Changes in geometric orientation caused by respiration and deformation can be expressed as rotations about the x-, y-, and z-axes. However, this is possible with only nonspherical—that is, not absolutely round—tumors. To determine the orientation of a tumor during a specific time frame, we used an approach derived from principal component analysis called the oriented bounding box method (27). This method has already been used to measure the extent of a tumor along its principal axes (28) and to support dose calculations for brachytherapy planning (29). To determine the principal axes of a segmented object, we calculated the covariance matrix of the segmented voxels. The normalized eigenvectors of this matrix defined the principal axes of the tumor, which were used to calculate the minimal object-oriented bounding box totally enclosing the structure of interest. The principal axes were calculated for each time frame. To express changes in the spatial orientation of a tumor in terms of rotation, the principal axes were interpreted as the basis vectors of a local orthogonal coordinate system. A rotation matrix describing the transformation between two coordinate systems could then be calculated by using the pairwise dot product of the basis vectors. This rotation matrix was then transformed into Euler angles that described the rotations around the patients' (global) axes.

In this study, rotation angles were calculated between the current time frame and a reference time frame at maximal expiration. The x-axis represented a horizontal line from the right to the left, the y-axis represented a vertical line from the caudal to the cranial region, and the z-axis represented a horizontal line from the anterior to the posterior region. The surface of the tumor was subdivided into eight segments to facilitate the detection of rotation.

Statistical Analyses
Statistical analyses of the data were performed with SAS software (SAS Institute, Cary, NC). Correlations were investigated by using Spearman correlation analysis. Mean values and standard deviations were calculated.

Statistical analyses of the volumetry data were performed by using WinSTAT (version 2001; Fitch Software, Staufen, Germany) and Microsoft Excel (version 2002 [10.2614.3501], Service Pack 1; Microsoft, Redmond, Wash) software. The deviation between the measured and true volumes was expressed in terms of the mean error, standard deviation, RMSE, and 95% confidence interval of the RMSE. To determine significant deviations from the true volumes, a Wilcoxon test with a confidence level of 95% was performed.


   RESULTS
TOP
 ABSTRACT
 INTRODUCTION
 MATERIALS AND METHODS
 RESULTS
DISCUSSION
 ADVANCE IN KNOWLEDGE
 References
 
In all phantoms and tumors, 3D dynamic MR imaging yielded good contrast between the tumor and the surrounding parenchymal tissue. In Figure 2, a respiratory cycle, from maximal inspiration to maximal expiration, is shown.

Phantom Volumetry
Three of the 21 algorithms—methods B6-1, B26-1, and GlinV2-1—proved to be the most effective for measuring phantom volumes (Fig 4, Table 1). The mean errors and standard deviations for deviations between measured and true volumes were 4% ± 13 (RMSE, 14%) with method B6-1, –3% ± 11 (RMSE, 11%) with method B26-1, and –2% ± 14 (RMSE, 14%) with method GlinV2-1. Method GlinV2-1 was found to be the most precise because it was the only algorithm that yielded measured phantom volumes that were not significantly different from the real phantom volumes. Method B-2, which was equivalent to the commonly used volumetry approach, yielded a mean error of 22% ± 19 (RMSE, 29%). The mean segmented phantom volumes measured by using the GlinV2-1 algorithm were 120.8 cm3 ± 4.1, 36.1 cm3 ± 3.98, and 13.1 cm3 ± 1.5 versus the actual volumes of 130, 40, and 12 cm3, respectively. The correlation between the two volume sets was highly significant (r > 0.9, P < .01).


Figure 4
View larger version (32K):
[in this window]
[in a new window]
[Download PPT slide]
  		Figure 4: Box and whisker plot depicts the deviation of the phantom volumes measured by using 21 algorithms from the real phantom volumes. With each algorithm, 180 measurements were applied. The best volumetric estimations were obtained by using method GlinV2-1 (see Volumetry section in Materials and Methods). + = minimal and maximal values, – = median deviation. Boxes range from first to third quartiles; whiskers range from fifth to 95th percentiles.

 
Volumetric Verification at CT and Histologic Analysis
The mean real tumor volume measured by using inspiratory CT was 45.5 cm3 ± 15. In three patients, histologic volumetry was possible and yielded a mean tumor volume of 41.2 cm3 ± 12. In these three patients, the mean CT tumor volume was 43.3 cm3 ± 14. The mean inspiratory tumor volume measured by using the GlinV2-1 algorithm at MR imaging in these three patients was 44.1 cm3 ± 11. The difference between histologic, CT, and MR imaging volume measurements was not significant.

In all five patients, there were substantial changes in the tumor volume during the respiratory cycle (Table 2), independent of the volumetry algorithm used. During expiration, tumor volume was the lowest, whereas during inspiration, tumor volume increased significantly by a factor of 1.32–2.00.


View this table:
[in this window]
[in a new window]

  		Table 2. Changes in Volume and Orientation of Tumors during Respiration in Five Patients

 
Rotation
It was possible to quantify the tumor rotation with 3D dynamic MR imaging, at which substantial changes in the orientation of the tumor during the respiratory cycle were observed (Fig 5). Rotational values are highly individualistic and depend on the characteristics of the tumor and the intensity of the respiratory maneuver (Table 2). We observed no clear evidence of a strong relationship between tumor size and tumor rotation. The maximal tumor rotation was approximately 46° about the x-axis, and the minimal rotation was about 8° about the y-axis.


Figure 5
View larger version (46K):
[in this window]
[in a new window]
[Download PPT slide]
  		Figure 5: Rotation of principal axes of tumors (in same patient as in Figure 2) during respiration, from maximal inspiration to maximal expiration and back. The surface of the tumor was subdivided into eight segments to facilitate detection of rotation. Numbers (1–12) on the horizontal axis are numbers of images and the time scale. (One image per second was acquired.)

 

   DISCUSSION
TOP
 ABSTRACT
 INTRODUCTION
 MATERIALS AND METHODS
 RESULTS
 DISCUSSION
ADVANCE IN KNOWLEDGE
 References
 
Pulmonary nodules are generally detected and monitored by using CT. If growth is detected, the histologic characteristics of the nodule must be studied (30). However, some malignant nodules grow asymmetrically, and such growth may be missed if only a subset of the data is used to perform conventional two-dimensional measurements on a single section (24). A working group reported the occurrence of an asymmetrically growing nodule that was considered to have a doubling time consistent with benignity according to two-dimensional measurements but a doubling time consistent with malignancy according to 3D measurements (24). Moreover, the interobserver variability of manual measurements of small nodules is poor. Thus, a 3D CT imaging technique, as was also used for volume verification in this study, is the generally accepted tool for quantifying the volume of intrapulmonary nodules (3–5,25). Since all available 3D techniques involve the use of static images obtained during a single breath hold, the potential effect of the respiratory level (3) cannot be adequately investigated or duly considered.

The current dynamic MR imaging examination is a single-section two-dimensional technique that enables visualization of tumor mobility and lung motion during the respiratory cycle (14). Despite the high spatial and temporal resolution achieved with this examination, as a two-dimensional technique it has limitations: It does not enable the acquisition of volumetric and rotational information about a tumor during the respiratory cycle. The temporal resolution that is achievable with current MR imaging techniques is inadequate for 3D approaches. Parallel imaging facilitates a substantial improvement in temporal resolution. However, a major drawback of parallel imaging is the reduction in achievable signal-to-noise ratio (31). Consequently, with the hardware configuration that was available at the time of our study, the acceleration factor could not be increased to more than three without compromising tumor visualization. In contrast, view sharing does not affect the signal-to-noise ratio since the same quantity of data is used to calculate the 3D data sets (18). By using a combination of parallel imaging and view sharing, we were able to acquire isotropic 3D data sets with a temporal resolution of 1 second. To conform to this limited temporal resolution, patients were instructed to breathe very slowly from maximal expiration to maximal inspiration, as has been recently proposed (32).

Although volumetry based on the counting of voxels might be appropriate for quantification of large objects with high spatial resolution (32), it is not effective when performed by using 3D dynamic MR image data from small objects such as nodules and tumors. The main reason for this is the limited spatial resolution of the images and the small size of the tumors. The disadvantageous surface area–to-volume ratio intensifies the partial volume effect (33). Although the use of intraobserver techniques can help to confirm the reproducibility and qualitative aspects of volumetric techniques (3–5,25), as has been undertaken increasingly, potential miscalculations cannot be disregarded since there is no real volumetric reference standard. Therefore, we applied another approach and performed phantom experiments for verification (2).

To solve the problem of partial volume effects, we used algorithms that have already been successfully applied in the field of liver surgery (26) and weighted voxels on the tumor surface (which are candidates for segmentation errors and ambiguities introduced by the partial volume effect) according to object properties. The most effective algorithm (method GlinV2-1) is based on the gray value and involves weighting of only the SPV area voxels according to their deviations from the mean gray value of the tumor core. The main reason that this algorithm was the most effective may be that using this weighting scheme causes the tumor border to become "fuzzy" and thereby influences the partial volume effect. Potential segmentation errors are minimized by downgrading these voxels. The weighting of only the SPV area voxels might indicate that interactive segmentation of small objects tends to lead to overestimated segmentation results. In contrast, conventional volumetry (ie, method B-2, which involves counting the interactively segmented voxels without weighting) yielded an RMSE of 29%. Therefore, 3D dynamic MR imaging volumetry with use of the identified partial volume correction algorithm (ie, GlinV2-1) enables accurate quantification of tumor volumes.

It is interesting that we detected a significant change in tumor volumes during the respiratory cycle. Tumors were bigger during inspiration. Whether a stretching of the tumor parenchyma during inspiration might be a reason for this observation must be investigated in additional studies. Since these volumetric changes proved to be very significant and were detected with use of all 21 algorithms, this might be an important indication of the necessity to carefully and correctly perform the breath hold maneuver, especially during follow-up examinations of pulmonary nodules.

We also quantified tumor rotation. Rotation of the principal axes does not necessarily correspond to a real rotation of the tumor. In fact, it indicates changes in the tumor orientation caused by rotation and deformation over time. The accuracy of this measurement is directly dependent on the size of the tumor (expressed in number of voxels) being investigated. The accuracy of this measurement also depends on the shape of the tumor. If the tumor is elongated and nonspherical, the orientation and rotation can be calculated accurately. The more spherical a tumor is, the more susceptible the calculation of the principal axis rotations will be to segmentation errors. It follows, then, that the calculated rotation values will be less accurate. Additional studies with a larger group of patients must address the importance of rotational deformation, as well as the dependency of observed rotational values on tumor size and location.

The major limitation of this study was the low temporal resolution of 3D dynamic MR imaging and the still lower T2 contrast of the 3D fast low-angle shot sequence, as compared with the temporal resolution and T2 contrast possible with two-dimensional true fast imaging with steady-state precession (34). Therefore, the described technique does not enable segmentation of nonsolid nodules, even though such nodules—especially partially solid ones—have a higher risk of malignancy (35).

In conclusion, 3D dynamic MR imaging performed by using parallel imaging and view sharing combined and with a postprocessing technique represents a potential method of noninvasively quantifying the volume and rotation of pulmonary nodules during respiration. The described postprocessing technique was validated by using phantom experiments and in patients with CT and histologic analysis. In addition, the rotation of nodules during respiration could be quantified. Volumetry and rotation quantification were observed to have a substantial dependence on the respiration level. This might affect follow-up measurements and high-precision radiation therapy.


   ADVANCE IN KNOWLEDGE
TOP
 ABSTRACT
 INTRODUCTION
 MATERIALS AND METHODS
 RESULTS
 DISCUSSION
 ADVANCE IN KNOWLEDGE
References
 


   ACKNOWLEDGMENTS
 
The authors gratefully acknowledge the statistical support of Ivan Zuna, PhD. Furthermore, the assistance Susanne Yubai and Kathleen Knauer provided with the patient examinations is greatly appreciated.


   FOOTNOTES
 

Abbreviations: RMSE = root mean square error • SPV = segmented partial volume • 3D = three-dimensional

Authors stated no financial relationship to disclose.

Author contributions: Guarantors of integrity of entire study, C.P., M.S., H.U.K.; 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, C.P., C.F.; clinical studies, C.P., H.H., J.D.; experimental studies, C.P., M.S.; statistical analysis, M.S., H.P.M.; and manuscript editing, C.P., M.S., C.F., H.U.K. C.P. and M.S. contributed equally to this work.


   References
TOP
 ABSTRACT
 INTRODUCTION
 MATERIALS AND METHODS
 RESULTS
 DISCUSSION
 ADVANCE IN KNOWLEDGE
 References
 

  1. Henschke CI, McCauley DI, Yankelevitz DF. Early lung cancer action project: overall design and findings from baseline screening. Lancet 1999;354:99–105.[CrossRef][Medline]
  2. Yankelevitz DF, Gupta R, Zhao B, Henschke CI. Small pulmonary nodules: evaluation with repeat CT—preliminary experience. Radiology 1999;212:561–566.[Abstract/Free Full Text]
  3. Revel MP, Lefort C, Bissery A, et al. Pulmonary nodules: preliminary experience with three-dimensional evaluation. Radiology 2004;231:459–466.[Abstract/Free Full Text]
  4. Kostis WJ, Yankelevitz DF, Reeves AP, Fluture SC, Henschke CI. Small pulmonary nodules: reproducibility of three-dimensional volumetry measurement and estimation of time to follow-up CT. Radiology 2004;231:446–452.[Abstract/Free Full Text]
  5. Wormanns D, Kohl G, Klotz E, et al. Volumetric measurements of pulmonary nodules at multi-row detector CT: in vivo reproducibility. Eur Radiol 2004;14:86–92.[CrossRef][Medline]
  6. Shimizu S, Shirato H, Kagei K, et al. Impact of respiratory movement on the computed tomographic images of small lung tumors in three-dimensional (3D) radiotherapy. Int J Radiat Oncol Biol Phys 2000;46:1127–1133.[CrossRef][Medline]
  7. van Sornsen de Koste JR, Lagerwaard FJ, Schuchhard-Schipper RH, et al. Dosimetric consequences of tumor mobility in radiotherapy of stage I non-small cell lung cancer: an analysis of data generated using "slow" CT scans. Radiother Oncol 2001;61:93–99.[CrossRef][Medline]
  8. Stroom JC, de Boer JC, Huizenga H, et al. Inclusion of geometrical uncertainties in radiotherapy treatment planning by means of coverage probability. Int J Radiat Oncol Biol Phys 1999;43:905–919.[CrossRef][Medline]
  9. Schreibmann E, Lahanas M, Uricchio R, Theodorou K, Kappas C, Baltas D. A geometry based optimisation algorithm for conformal external beam orientation. Phys Med Biol 2003;48:1825–1841.[CrossRef][Medline]
  10. Kondo T, Kobayashi I, Taguchi Y, Ohta Y, Yanagimachi N. A dynamic analysis of chest wall motions with MRI in healthy young subjects. Respirology 2000;5:19–25.[CrossRef][Medline]
  11. Napadow VJ, Mai V, Bankier A, Gilbert RJ, Edelman R, Chen Q. Determination of regional pulmonary parenchymal strain during normal respiration using spin inversion tagged magnetization MRI. J Magn Reson Imaging 2001;13:467–474.[CrossRef][Medline]
  12. Plathow C, Fink C, Ley S, et al. Measurement of diaphragmatic length during the breathing cycle by dynamic MRI: comparison between healthy adults and patients with an intrathoracic tumor. Eur Radiol 2004;14:1392–1399.[Medline]
  13. Plathow C, Ley S, Fink C, et al. Evaluation of chest motion and volumetry during the breathing cycle by dynamic MRI in healthy subjects: comparison with pulmonary function tests. Invest Radiol 2004;39:202–209.[CrossRef][Medline]
  14. Plathow C, Ley S, Fink C, et al. Analysis of intrathoracic tumor mobility during the whole breathing cycle by dynamic MRI. Int J Radiat Oncol Biol Phys 2004;59:952–959.[CrossRef][Medline]
  15. Sodickson DK, Manning WJ. Simultaneous acquisition of spatial harmonics (SMASH): fast imaging with radiofrequency coil arrays. Magn Reson Med 1997;38:591–603.[Medline]
  16. Pruessmann KP, Weiger M, Scheidegger MB, Boesiger P. SENSE: sensitivity encoding for fast MRI. Magn Reson Med 1999;42:952–962.[CrossRef][Medline]
  17. Griswold MA, Jakob PM, Heidemann RM, et al. Generalized autocalibrating partially parallel acquisition (GRAPPA). Magn Reson Med 2002;47:1202–1210.[CrossRef][Medline]
  18. Fink C, Ley S, Kroeker R, Requardt M, Kauczor HU, Bock M. Time-resolved contrast-enhanced three-dimensional magnetic resonance angiography of the chest: combination of parallel imaging with view sharing (TREAT). Invest Radiol 2005;40:40–48.[Medline]
  19. Mitra S, Plank LD, Knight GS, Hill GL. In vivo measurement of total body chlorine using the 8.57 MeV prompt de-excitation following thermal neutron capture. Phys Med Biol 1993;38:161–172.[CrossRef][Medline]
  20. Watanabe H, Nakata H, Egashira K, Nakamura K. Spiral volumetric CT as a routine technique for thoracic imaging. J Thorac Imaging 1993;8:316–320.[Medline]
  21. Heimann T, Thorn M, Kunert T, Meinzer HP. New methods for leak detection and contour correction in seeded region growing segmentation: international archives of photogrammetry and remote sensing, part B. Presented at the 20th International Society for Photogrammetry and Remote Sensing Congress, Istanbul, Turkey, July 12–23, 2004.
  22. Soille P. Morphological image analysis. 2nd ed. Berlin, Germany: Springer, 2002.
  23. Kuhnigk JM, Dicken V, Bornemann L, Wormanns D, Krass S, Peitgen HO. Fast automated segmentation and reproducible volumetry of pulmonary metastases in CT-scans for therapy monitoring. Presented at Seventh International Medical Image Computing and Computer-Assisted Intervention Conference, Saint-Malo, France, September 26–29, 2004.
  24. Yankelevitz DF, Reeves AP, Kostis WJ, Zhao B, Henschke CI. Small pulmonary nodules: volumetrically determined growth rates based on CT evaluation. Radiology 2000;217:251–256.[Abstract/Free Full Text]
  25. Wormanns D, Beyer F, Diederich S, Ludwig K, Heindel W. Diagnostic performance of a commercially available computer-aided diagnosis system for automatic detection of pulmonary nodules: comparison with single and double reading. Rofo 2004;176:953–958.[Medline]
  26. Thorn M, Kremer M, Heimann T, et al. Accurate volume measurement for liver surgery: in vivo evaluation with a pig model. Presented at Computer Assisted Radiology and Surgery (CARS 2004) 18th International Congress and Exhibition, Chicago, Ill, June 23–26, 2004.
  27. Gottschalk S, Lin MC, Manocha D. OBB-tree: a hierarchical structure for rapid interference detection. Presented at 23rd Annual Conference on Computer Graphics and Interactive Techniques SIGGRAPH '96, New Orleans, La, August 4–9, 1996.
  28. Preim B, Tietjen C, Spindler W, Peitgen HO. Integration of measurement tools in medical visualizations. IEEE Visualization 2002: 21–28.
  29. Lahanas M, Kemmerer T, Milickovic N, Karouzakis K, Baltas D, Zamboglou N. Optimized bounding boxes for three-dimensional treatment planning in brachytherapy. Med Phys 2000;27:2333–2342.[CrossRef][Medline]
  30. Yankelevitz DF, Henschke CI. Does 2-year stability imply that pulmonary nodules are benign? AJR Am J Roentgenol 1997;168:325–328.[Free Full Text]
  31. Madore B, Pelc NJ. SMASH and SENSE: experimental and numerical comparisons. Magn Reson Med 2001;45:1103–1111.[CrossRef][Medline]
  32. Plathow C, Schoebinger M, Fink C, et al. Evaluation of lung volumetry using dynamic three-dimensional magnetic resonance imaging. Invest Radiol 2005;40:173–178.[CrossRef][Medline]
  33. Bello F, Colchester AC, Roll SA. A geometry- and intensity-based partial volume correction for MRI volumetric studies. Comput Med Imaging Graph 1998;22:123–132.[CrossRef][Medline]
  34. Jung BA, Hennig J, Scheffler K. Single-breathhold 3D-trueFISP cine cardiac imaging. Magn Reson Med 2002;48:921–925.[CrossRef][Medline]
  35. Henschke CI, Yankelevitz DF, Mirtcheva R, McGuiness G, McCauley D, Miettinien OS. CT screening for lung cancer: frequency and significance of part-solid and nonsolid nodules. AJR Am J Roentgenol 2002;178:1053–1057.[Abstract/Free Full Text]




This Article
Right arrow Abstract Freely available
Right arrow Figures Only
Right arrow Full Text (PDF)
Right arrow All Versions of this Article:
2401050727v1
240/2/537    most recent
Right arrow Submit a response
Right arrow Alert me when this article is cited
Right arrow Alert me when eLetters are posted
Right arrow Alert me if a correction is posted
Right arrow Citation Map
Services
Right arrow Email this article to a friend
Right arrow Similar articles in this journal
Right arrow Similar articles in PubMed
Right arrow Alert me to new issues of the journal
Right arrow Download to citation manager
Right arrow reprints & permissions
Citing Articles
Right arrow Citing Articles via Google Scholar
Google Scholar
Right arrow Articles by Plathow, C.
Right arrow Articles by Kauczor, H.-U.
Right arrow Search for Related Content
PubMed
Right arrow PubMed Citation
Right arrow Articles by Plathow, C.
Right arrow Articles by Kauczor, H.-U.

HOME HELP FEEDBACK SUBSCRIPTIONS ARCHIVE SEARCH TABLE OF CONTENTS
RADIOLOGY RADIOGRAPHICS RSNA JOURNALS ONLINE