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


     

This Article
Right arrow Abstract Freely available
Right arrow Figures Only
Right arrow Full Text (PDF)
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 Joe, B. N.
Right arrow Articles by Bozik, M. E.
Right arrow Search for Related Content
PubMed
Right arrow PubMed Citation
Right arrow Articles by Joe, B. N.
Right arrow Articles by Bozik, M. E.
(Radiology. 1999;212:811-816.)
© RSNA, 1999

Computer Applications

Brain Tumor Volume Measurement: Comparison of Manual and Semiautomated Methods1

Bonnie N. Joe, MD, PhD, Melanie B. Fukui, MD, Carolyn Cidis Meltzer, MD, Qing-shou Huang, MS, Roger S. Day, ScD, Phil J. Greer, BS and Michael E. Bozik, MD

1 From the Departments of Radiology, Division of Neuroradiology (B.N.J., M.B.F., C.C.M., P.J.G.), Psychiatry (C.C.M.), Neurosurgery (M.E.B.), Neurology (M.E.B.), and Medicine (M.E.B.), and the Pittsburgh Cancer Institute (Q.H., R.S.D.), University of Pittsburgh Medical Center, Pittsburgh, Pa. Received May 29, 1998; revision requested July 22; revision received November 11; accepted March 29, 1999. Address reprint requests to M.B.F., Department of Radiology, Presbyterian University Hospital, Rm D-132, 200 Lothrop St, Pittsburgh, PA 15213-2582 (e-mail: fukuimb@radserv.arad.upmc.edu).


   Abstract
TOP
 Abstract
Introduction
 MATERIALS AND METHODS
 RESULTS
 DISCUSSION
 References
 
PURPOSE: To compare the reliability of two approaches to measuring enhancing brain tumor volumes—the conventional manual trace method and a threshold-based, semiautomated computer software method.

MATERIALS AND METHODS: Two operators rated contrast material–enhanced, T1-weighted axial magnetic resonance (MR) image data sets from 16 patients aged 21–71 years with high-grade gliomas. Each MR data set was rated twice by using manual tracing and twice by using the semiautomated method. The semiautomated measurement method involved a thresholding algorithm based on mixture modeling. The data collection time for each method was recorded. Reliability was measured by using inter- and intraoperator agreement indexes.

RESULTS: Mean intraoperator agreement indexes (± SD) were 0.90 ± 0.09 (operator 1) and 0.83 ± 0.15 (operator 2) for the manual trace method and 0.83 ± 0.17 (operator 1) and 0.84 ± 0.16 (operator 2) for the semiautomated measurement method. The mean interoperator agreement was 0.85 ± 0.14 for the manual method and 0.82 ± 0.18 for the semiautomated method. The semiautomated method was faster than the manual trace method by an average of 4.6 minutes per patient.

CONCLUSION: The semiautomated computer method of measuring tumor volume was faster than the manual trace method. Semiautomated computer approaches offer an alternative to manual tracing for measuring serial tumor volumes in patients with high-grade brain neoplasms.

Index terms: Brain neoplasms, 10.363, 10.3634 • Brain neoplasms, MR, 10.12141, 10.12143 • Magnetic resonance (MR), volume measurement, 10.12141, 10.12143 • Technology assessment


   Introduction
TOP
 Abstract
 Introduction
MATERIALS AND METHODS
 RESULTS
 DISCUSSION
 References
 
Brain tumor volumes measured from diameters on computed tomographic (CT) and magnetic resonance (MR) images have been commonly used in the evaluation of intracerebral tumors and have implications for patient prognosis and treatment. Tracing the enhancing margin could potentially provide a more representative measurement of tumor size and has been used in clinical trials to assess treatment response (1). At many institutions, brain tumor volume measurements are obtained by the MR technologist, who manually traces the enhancing tissue on each image by using widely available imaging software. This approach is potentially subjective and prone to large variations in intra- and interoperator performance. Such variations affect the validity of serial measurements used in the planning of therapy, because a single technologist does not necessarily perform all tumor volume measurements in a given patient. Variability in serial volume measurements and inconsistency between observers have been noted by others (2–5) in studies involving tumor volume measurement methods used to follow tumor response to treatment or to plan treatment. Although previous studies (2,3) have involved limited evaluations of the reliability of tumor volumetrics, detailed comparisons of alternative semiautomated methods versus the conventional manual trace method have not been performed. To our knowledge, the reliability of manual tracing has not been considered in the more extensive studies (6,7) of inter- and intraoperator reliability of tumor volume measurement methods.

In this study, we quantitatively compared two methods of measuring enhancing tumor volumes from MR image data—the conventional manual trace method and a threshold-based semiautomated computer segmentation method. To compare the inter-and intraoperator reliability of the two methods, two operators performed tumor volume determinations twice with each method in the same subjects. Our hypotheses were that (a) the manual trace method would be more user dependent—that is, have low interoperator reliability, and (b) the semiautomated method would provide more reproducible measurements—that is, have high intraoperator reliability.


   MATERIALS AND METHODS
TOP
 Abstract
 Introduction
 MATERIALS AND METHODS
RESULTS
 DISCUSSION
 References
 
Patient Population
The study group consisted of 16 adult patients (13 men, three women; mean age, 53 years; age range, 22–71 years) with histologically proved, unresected high-grade gliomas, who were randomly selected from an ongoing treatment protocol conducted by the University of Pittsburgh Medical Center Neuro-Oncology Service. The inclusion criterion was the presence of an enhancing tumor on more than one MR image obtained by using routine imaging parameters.

MR Image Acquisition
From images obtained by using the routine clinical imaging protocol performed with a Signa 1.5-T imaging unit (GE Medical Systems, Milwaukee, Wis), we selected the contrast material–enhanced (0.1 mmol/kg gadoteridol [Prohance; Bracco Diagnostics, Princeton, NJ] or gadopentetate dimeglumine [Magnevist; Berlex Laboratories, Wayne, NJ]), T1-weighted axial MR series for the tumor volume measurements. The following parameters were used to obtain the T1-weighted images: 450–650/16–35 (repetition time range msec/echo time range msec), 256 x 192 matrix, 24-cm field of view, 5-mm section thickness with a 1-mm intersection gap, and one signal acquired.

Tumor Volume Measurement
Two nonradiologists (B.N.J., who was a medical student at the time of the study, and P.J.G., a research assistant in our positron emission tomography [PET] center) independently rated each MR image twice by using each of the two measurement methods—that is, manual tracing and semiautomated segmentation. Operator 1 followed the order ABBA, whereas operator 2 followed the order ABAB, where method A was the semiautomated segmentation method and method B was the manual trace method. The images obtained in all 16 subjects were rated in one batch for each manual or semiautomated measurement trial. The rationale for using the two nonradiologists in this study was that tumor volume measurements are typically performed by MR technologists rather than radiologists in our radiology department. To compare inter- and intraoperator reliability, we needed two individuals who were available to perform the repetitive measurements. Because our technologists were not available to obtain these measurements, we chose a medical student and a PET research assistant, because their familiarity with imaging approached that of an imaging technologist but not that of a radiologist.

The image processing time for both methods was recorded by both operators during the second trial. In each patient, the total processing time included two components—the time required to generate tumor contours and the time required to load and save image data.

For both methods, only the enhancing portions of the tumor were included in the volume calculation. The nonenhancing portions within the tumor (ie, areas of tumor necrosis) were traced, and their volumes were excluded from the total tumor volume.

Manual trace method.—Tumor volume measurements were determined by manually outlining, with a track ball, the enhancing portion of the mass on each image by using standard software on the imager console (GE Medical Systems). Nonenhancing regions within the tumor also were outlined and subtracted so that only the enhancing tumor area was measured. An example of manual tracing is shown in Figure 1. The area measurements were automatically calculated and multiplied by the MR image section thickness plus the intersection gap to calculate a per-section tumor volume. The total tumor volume was obtained by summing the volume calculations for all sections. Each image, along with its traced contours, was saved by using the "screen save" function on the console and then stored on optical disks.



View larger version (146K):
[in this window]
[in a new window]
  		Figure 1a. (a, b) Axial T1-weighted MR images (repetition time msec/echo time msec, 466/11) demonstrate the manual trace method in two areas of tumor enhancement (arrows in a) in the left temporal lobe of a patient with glioblastoma multiforme. In b, the contours generated by manually tracing the enhancing portions of the tumor are shown.

 


View larger version (146K):
[in this window]
[in a new window]
  		Figure 1b. (a, b) Axial T1-weighted MR images (repetition time msec/echo time msec, 466/11) demonstrate the manual trace method in two areas of tumor enhancement (arrows in a) in the left temporal lobe of a patient with glioblastoma multiforme. In b, the contours generated by manually tracing the enhancing portions of the tumor are shown.

 
Semiautomated segmentation method.—For the semiautomated segmentation method, threshold-based segmentation software (Morph; Silicon Graphics, Mountain View, Calif) was used. This software, which has not yet been released for general use (Thulborn K., personal communication, September 1998) (8), was operated on an SGI workstation (Silicon Graphics). For each image, the operator selected a point just within the enhancing border of the mass (Fig 2a) and a point just outside the enhancing border of the mass (Fig 2b). These two points defined the threshold used to generate the contour surrounding the enhancing area (Fig 2c). Similarly, nonenhancing regions within the tumor were excluded by choosing a point inside the nonenhancing (necrotic) region (Fig 2d) and a point outside the nonenhancing region (Fig 2e) to generate the contour surrounding the nonenhancing area. Figure 2f shows the result after all enhancing areas have been segmented. As with the manual trace method, the calculated tumor volume reflects the enhancing area only. A similar method, which involves the same mixture modeling algorithm for contour derivation, was previously validated for volume measurements in a phantom, with errors of 4%–6% (9). With the semiautomated method, the contours were linked to the original images, and only the contour information plus the linking information was stored.



View larger version (196K):
[in this window]
[in a new window]
  		Figure 2a. (a-d) Axial T1-weighted MR images (466/11) obtained in the same patient and same image section as those in Figure 1 illustrate the semiautomated computer tracing method. (a-c) First, a point just inside the enhancing border (blue dot at tip of arrow in a) is selected. (b, c) Next, a point just outside the enhancing border (yellow dot at tip of arrow in b) is selected. In c, the enhancing area is automatically outlined (shown in red). (d, e) By selecting a point inside the nonenhancing (necrotic) region (blue dot at tip of arrow in d) and a point outside the nonenhancing region (yellow dot at tip of arrow in e), the contour surrounding the nonenhancing area (outlined in red in e) is generated and excluded from the final tumor volume measurement. (In a-e, all dots have been increased from 1 pixel to 4 pixels in width for clarity of reproduction.) (f) The final result after all enhancing areas have been segmented is shown.

 


View larger version (191K):
[in this window]
[in a new window]
  		Figure 2b. (a-d) Axial T1-weighted MR images (466/11) obtained in the same patient and same image section as those in Figure 1 illustrate the semiautomated computer tracing method. (a-c) First, a point just inside the enhancing border (blue dot at tip of arrow in a) is selected. (b, c) Next, a point just outside the enhancing border (yellow dot at tip of arrow in b) is selected. In c, the enhancing area is automatically outlined (shown in red). (d, e) By selecting a point inside the nonenhancing (necrotic) region (blue dot at tip of arrow in d) and a point outside the nonenhancing region (yellow dot at tip of arrow in e), the contour surrounding the nonenhancing area (outlined in red in e) is generated and excluded from the final tumor volume measurement. (In a-e, all dots have been increased from 1 pixel to 4 pixels in width for clarity of reproduction.) (f) The final result after all enhancing areas have been segmented is shown.

 


View larger version (193K):
[in this window]
[in a new window]
  		Figure 2c. (a-d) Axial T1-weighted MR images (466/11) obtained in the same patient and same image section as those in Figure 1 illustrate the semiautomated computer tracing method. (a-c) First, a point just inside the enhancing border (blue dot at tip of arrow in a) is selected. (b, c) Next, a point just outside the enhancing border (yellow dot at tip of arrow in b) is selected. In c, the enhancing area is automatically outlined (shown in red). (d, e) By selecting a point inside the nonenhancing (necrotic) region (blue dot at tip of arrow in d) and a point outside the nonenhancing region (yellow dot at tip of arrow in e), the contour surrounding the nonenhancing area (outlined in red in e) is generated and excluded from the final tumor volume measurement. (In a-e, all dots have been increased from 1 pixel to 4 pixels in width for clarity of reproduction.) (f) The final result after all enhancing areas have been segmented is shown.

 


View larger version (195K):
[in this window]
[in a new window]
  		Figure 2d. (a-d) Axial T1-weighted MR images (466/11) obtained in the same patient and same image section as those in Figure 1 illustrate the semiautomated computer tracing method. (a-c) First, a point just inside the enhancing border (blue dot at tip of arrow in a) is selected. (b, c) Next, a point just outside the enhancing border (yellow dot at tip of arrow in b) is selected. In c, the enhancing area is automatically outlined (shown in red). (d, e) By selecting a point inside the nonenhancing (necrotic) region (blue dot at tip of arrow in d) and a point outside the nonenhancing region (yellow dot at tip of arrow in e), the contour surrounding the nonenhancing area (outlined in red in e) is generated and excluded from the final tumor volume measurement. (In a-e, all dots have been increased from 1 pixel to 4 pixels in width for clarity of reproduction.) (f) The final result after all enhancing areas have been segmented is shown.

 


View larger version (191K):
[in this window]
[in a new window]
  		Figure 2e. (a-d) Axial T1-weighted MR images (466/11) obtained in the same patient and same image section as those in Figure 1 illustrate the semiautomated computer tracing method. (a-c) First, a point just inside the enhancing border (blue dot at tip of arrow in a) is selected. (b, c) Next, a point just outside the enhancing border (yellow dot at tip of arrow in b) is selected. In c, the enhancing area is automatically outlined (shown in red). (d, e) By selecting a point inside the nonenhancing (necrotic) region (blue dot at tip of arrow in d) and a point outside the nonenhancing region (yellow dot at tip of arrow in e), the contour surrounding the nonenhancing area (outlined in red in e) is generated and excluded from the final tumor volume measurement. (In a-e, all dots have been increased from 1 pixel to 4 pixels in width for clarity of reproduction.) (f) The final result after all enhancing areas have been segmented is shown.

 


View larger version (185K):
[in this window]
[in a new window]
  		Figure 2f. (a-d) Axial T1-weighted MR images (466/11) obtained in the same patient and same image section as those in Figure 1 illustrate the semiautomated computer tracing method. (a-c) First, a point just inside the enhancing border (blue dot at tip of arrow in a) is selected. (b, c) Next, a point just outside the enhancing border (yellow dot at tip of arrow in b) is selected. In c, the enhancing area is automatically outlined (shown in red). (d, e) By selecting a point inside the nonenhancing (necrotic) region (blue dot at tip of arrow in d) and a point outside the nonenhancing region (yellow dot at tip of arrow in e), the contour surrounding the nonenhancing area (outlined in red in e) is generated and excluded from the final tumor volume measurement. (In a-e, all dots have been increased from 1 pixel to 4 pixels in width for clarity of reproduction.) (f) The final result after all enhancing areas have been segmented is shown.

 
Each operator underwent a training and validation exercise before beginning the tumor volume measurements. This involved using the semiautomated method to measure the volume of a cadaverous brain hemisphere that was imaged with a volumetric spoiled gradient-recalled-echo sequence (25/5, 40° flip angle, one signal acquired, 256 x 192 matrix, 1.5-mm section thickness, 24-cm field of view). Both operators obtained measurements within 2% of the actual brain volume (as measured by the displacement of water by a brain fixed in a preservative substance) before proceeding with the tumor volume measurement study (8).

Statistical Considerations and Data Analyses
To compare the intra- and interoperator reliability of the two measurement methods, we used the following agreement index, AI, equation (10):

For interoperator agreement calculations, xa was the measurement obtained by operator 1 and xb, the measurement obtained by operator 2 with the same technique on the same image. For intraoperator agreement calculations, xa was the measurement made during the first trial, and xb, the measurement made during the second trial by the same operator with the same technique on the same image.

Intraoperator and interoperator agreement indexes were calculated for each image, because the tumor contours were generated on a section-by-section basis and represented independent observations. Making the agreement comparisons on a section-by-section basis also increased the statistical power of the study, because the tumors in the 16 patients spanned about 100 sections in total and consequently provided about 100 comparison data points.

An agreement index was calculated for each instance in which both volume measurements (ie, xa and xb) were greater than 0. (There were instances at the extreme inferior and superior tumor margins on the axial images in which very small contours [generally representing 2%–5% of the total tumor volume] were generated by one observer and not the other. The results of statistical analyses of these instances by means of the McNemar test showed no significant difference between the semiautomated method and manual trace method and no significant difference between operators in terms of the tendency to find tumor in a particular section.) A value of 1.0 indicated perfect agreement, and a value of 0 indicated no agreement. The measurements made by each operator during the second trial were used in the interoperator agreement calculations. A paired Student t test was used to evaluate differences in interoperator agreement between the semiautomated and manual trace methods. Intraoperator reliability was similarly assessed by evaluating the intraoperator agreement between operators and between methods. Statistical significance was determined by a P value of less than .05. A paired Student t test was also used to look for systematic bias in volume measurements between the two approaches.


   RESULTS
TOP
 Abstract
 Introduction
 MATERIALS AND METHODS
 RESULTS
DISCUSSION
 References
 
Tumor Volume Measurements
A total of 384 axial images obtained in the 16 patients were evaluated; approximately 100 of these images demonstrated enhancing tumor. Tumor enhancement was seen on approximately four to 10 (mean, 6.2) images per patient. The range of tumor volume measurements is presented in Table 1. In 49 (77%) of 64 volume measurements, the volume measurement obtained with the semiautomated method was lower (mean difference ± SD, 12.1% ± 11.6) than that obtained with the manual method (P = .001).


View this table:
[in this window]
[in a new window]
  		TABLE 1. Range of Tumor Volume Measurements per Patient
 
Intraoperator Agreement
The intraoperator agreement indexes for the two operators who used the two tumor volume measurement methods are summarized in Table 2. The difference in mean intraoperator agreement between the manual and semiautomated methods was significant for operator 1 (P = .003) but not for operator 2 (P = .74). Scatterplots of the intraoperator agreement indexes for the manual trace method versus the semiautomated segmentation method illustrate the superior reliability of operator 1 with the manual trace method compared with the reliability of this operator with the semiautomated method (Fig 3a). In contrast, for operator 2, there was no significant difference in intraoperator reliability between the two methods (Fig 3b). There was a significant difference in intraoperator reliability between operators when they used the manual trace method (P = .001), whereas there was no significant difference in intraoperator reliability between operators when they used the semiautomated method (P = .60).


View this table:
[in this window]
[in a new window]
  		TABLE 2. Mean Intraoperator Agreement Indexes
 


View larger version (20K):
[in this window]
[in a new window]
  		Figure 3a. Scatterplots of intraoperator agreement indexes for the manual versus semiautomated segmentation method. The dotted line is the line of equality, along which the agreement indexes for the manual and semiautomated methods are equal. Each data point represents one image. (a) Agreement indexes for operator 1. Note that there are more points clustered toward the manual axis; this indicates that operator 1 had better reliability with the manual trace method. (b) Agreement indexes for operator 2. In this case, the points are evenly distributed on both sides of the equality line; this indicates that operator 2 had comparable reliability with the two measurement methods.

 


View larger version (20K):
[in this window]
[in a new window]
  		Figure 3b. Scatterplots of intraoperator agreement indexes for the manual versus semiautomated segmentation method. The dotted line is the line of equality, along which the agreement indexes for the manual and semiautomated methods are equal. Each data point represents one image. (a) Agreement indexes for operator 1. Note that there are more points clustered toward the manual axis; this indicates that operator 1 had better reliability with the manual trace method. (b) Agreement indexes for operator 2. In this case, the points are evenly distributed on both sides of the equality line; this indicates that operator 2 had comparable reliability with the two measurement methods.

 
Interoperator Agreement
There was no significant difference (P = .20) in interoperator reliability between the two tumor volume measurement methods (Fig 4). The mean interoperator agreement index (± SD) was 0.85 ± 0.14 for the manual trace method and 0.82 ± 0.18 for the semiautomated segmentation method.



View larger version (20K):
[in this window]
[in a new window]
  		Figure 4. Scatterplot of interoperator agreement indexes for the manual versus semiautomated segmentation method. The agreement indexes for the manual and semiautomated methods are equal along the dotted line. The points are evenly distributed on both sides of the equality line; this indicates similar interoperator reliability between the two measurement methods.

 
Time Considerations
The mean time required to generate contours and the total tumor measurement time—that is, the time required to generate contours plus the time required to load and save image data—by using the two measurement methods are presented in Table 3. There was no significant difference in the average time required to generate contours between the manual trace method and the semiautomated segmentation method (P = .24). However, a significant difference in the total processing time between the two methods was found; the semiautomated method required an average of 4.6 minutes less per patient than did the manual trace method (P = .001). A significant difference in the total processing time per image section also was found between the two methods; the semiautomated method required an average of 0.6 minutes less per image than did the manual trace method (P = .008).


View this table:
[in this window]
[in a new window]
  		TABLE 3. Mean Tumor Volume Measurement Times
 

   DISCUSSION
TOP
 Abstract
 Introduction
 MATERIALS AND METHODS
 RESULTS
 DISCUSSION
References
 
In a study conducted by the Brain Tumor Cooperative Group (11), the tumor size at CT, as measured by using diameters, was determined to be of prognostic importance, independent of other known prognostic variables. In addition, Blankenberg et al (12) suggested that MR imaging– or CT-determined volumetric tumor doubling time is superior to histologic grade in predicting the survival of patients with intracranial astrocytomas. Currently, MR imaging is considered to be superior to CT in determining the extent of tumor infiltration, although histologic evidence of malignancy may extend beyond the margin of enhancement (13–15). In addition, surgery, steroids, and radiation therapy may substantially alter the enhancement characteristics of the tumor bed (16,17). Despite this, the extent of tumor enhancement at MR imaging or CT is routinely used to guide radiation therapy, and recent attention has been given to improving the accuracy of in vivo brain tumor volume measurements that are obtained with MR imaging and used in patient treatment and prognostic determinations (1,3,7,18).

In our study of the reliability of brain tumor volume measurements, we quantitatively compared the conventional manual trace method with a semiautomated computer segmentation method. We hypothesized that the semiautomated method would provide more reproducible measurements—that is, have high inter- and intraoperator reliability—whereas the manual trace method would be more user dependent and have low interoperator reliability. As expected, we found no significant difference in intraoperator reliability between operator 1 and 2 (mean agreement index, 0.83) when they used the semiautomated measurement method. In addition, interoperator reliability (agreement index, 0.82) was comparable to intraoperator reliability when the semiautomated method was used. However, the manual trace method was more user dependent in that operator 1 had better intraoperator reliability (agreement index, 0.90) than did operator 2 (agreement index, 0.83). When we compared the interoperator reliability of the two methods, we found no significant difference between the manual and semiautomated measurement methods (agreement index, 0.85 vs 0.82, respectively).

There was a highly significant difference in absolute volume measurements between the two methods; the semiautomated method resulted in lower volume measurements than did the manual tracing method 77% (49 of 64 measurements) of the time. We postulate that this may have resulted from the ability of the semiautomated method to provide a more detailed outline of the irregular edges of the tumor, whereas the manual trace method tends to approximate the tumor border with a smooth curve. Because the true histologic extent of these tumors is not delineated with contrast enhancement, it is impossible to determine which of the tumor volume measurement methods is more accurate (13–15). Given the systematic differences, all serial volume measurements that are typically used to follow tumor response to treatment should be performed with the same method.

In some studies (19), attempts to correlate survival with volume of abnormal tissue at T2-weighted imaging have been made. The semiautomated method used in this study would be impractical to apply to T2-weighted images because of the lack of contrast and the inability to discern a complete border between the tumor-related signal intensity abnormality and the cerebrospinal fluid. However, the semiautomated method potentially could be used for volume measurement of abnormal signal intensity at T2-weighted imaging with fluid-attenuated inversion recovery sequences, because the contrast between tumor and cerebrospinal fluid would be increased and thus allow appropriate segmentation. Morph software is currently being used to measure white matter signal intensity abnormalities on fluid-attenuated inversion recovery images in an ongoing study (C.C.M., personal communication, September 1998).

The semiautomated method required less time to generate tumor volume measurements than did the manual trace method. This difference reflects the decreased time required to load and save image data with the semiautomated measurement method. In addition, the semiautomated method required less storage space than did the manual trace method.

The differences between the two measurement methods that were difficult to quantify included qualitative "ease of use" factors. Both operators subjectively believed that the manual trace method was highly labor intensive and required more concentration to do well than did the semiautomated method. Because editing functions were not available with the manual trace method, an image had to be completely retraced if a mistake was made. Also, the outputs of the manual trace method were area measurements, which then had to be multiplied by the section thickness to generate volume measurements. The semiautomated method enabled tumor volumes to be calculated automatically and allowed basic editing functions.

The tumor volume measurements calculated with standard manual tracing and a semiautomated computer program were comparable in interoperator reliability, but the semiautomated computer program was faster than manual tracing. The reliability of MR imaging–determined tumor volume measurements potentially may be improved with decreased reliance on subjective image interpretation. Although the semiautomated method used in this study still requires an operator to choose the points that determine the threshold value used to outline the area of enhancement, generation of the tumor contour is based on consistent objective criteria. Several automated methods of MR image segmentation based on texture analysis (20), clustering techniques (eg, unsupervised fuzzy c-means) (21,22), or vector decomposition and probability techniques (23) have been reported on. These alternative approaches to MR image segmentation remain under development and retain problems with misclassification. Ultimately, by improving the reliability of MR imaging–based tumor volume measurements, these types of analyses may be used to follow tumor response to therapy and thus potentially influence patient care and treatment planning.


   Acknowledgments
 
We thank Keith Thulborn for supplying the Morph software before its completion.


   Footnotes
 
Author contributions: Guarantors of integrity of entire study, B.N.J., M.B.F., C.C.M.; study concepts, B.N.J., M.B.F., C.C.M., R.S.D., M.E.B.; study design, B.N.J., M.B.F., C.C.M., R.S.D.; definition of intellectual content, B.N.J., M.B.F., C.C.M.; literature research, B.N.J., M.B.F., C.C.M.; clinical studies, M.E.B.; experimental studies, M.B.F., B.N.J., C.C.M., P.J.G.; data acquisition, M.B.F., B.N.J., P.J.G., C.C.M.; data analysis, B.N.J., M.B.F., C.C.M., P.J.G., Q.H., R.S.D.; statistical analysis, B.N.J., Q.H., R.S.D.; manuscript preparation, B.N.J.; manuscript editing, B.N.J., M.B.F., C.C.M.; manuscript review, all authors.


   References
TOP
 Abstract
 Introduction
 MATERIALS AND METHODS
 RESULTS
 DISCUSSION
 References
 

  1. Grossman S, Wharam M, Sheidler V, et al. Phase II study of continuous infusion carmustine and cisplatin followed by cranial irradiation in adults with newly diagnosed high-grade astrocytoma. J Clin Oncol 1997; 15:2596-2603.[Abstract/Free Full Text]
  2. Zijlstra EJ, Taphoorn MJ, Barkhof F, Hoogenraad FG, Heimans JJ, Valk J. Radiotherapy response of cerebral metastases quantified by serial MR imaging. J Neurooncol 1994; 21:171-176.[Medline]
  3. Ten Haken RK, Thornton AF, Jr, Sandler HM, et al. A quantitative assessment of the addition of MRI to CT-based, 3-D treatment planning of brain tumors. Radiother Oncol 1992; 25:121-133.[Medline]
  4. Pan DH, Guo WY, Chung WY, Shiau CY, Liu RS, Lee LS. Early effects of Gamma Knife surgery on malignant and benign intracranial tumors. Stereotact Funct Neurosurg 1995; 64(suppl 1):19-31.
  5. Clarke LP, Velthuizen RP, Camacho MA, et al. MRI segmentation: methods and applications. Magn Reson Imaging 1995; 13:343-368.[Medline]
  6. Vaidyanathan M, Clarke LP, Hall LO, et al. Monitoring brain tumor response to therapy using MRI segmentation. Magn Reson Imaging 1997; 15:323-334.[Medline]
  7. Vaidyanathan M, Clarke LP, Velthuizen RP, et al. Comparison of supervised MRI segmentation methods for tumor volume determination during therapy. Magn Reson Imaging 1995; 13:719-728.[Medline]
  8. Uttecht S, Betancourt C, Thulborn K. Evaluation of semi-automated border recognition software from quantitative brain morphometry and topology from high-resolution MRI (abstr) In: Proceedings of the Sixth Meeting of the International Society for Magnetic Resonance in Medicine. Berkeley, Calif: International Society for Magnetic Resonance in Medicine, 1998; 2075.
  9. Johnson LA, Pearlman JD, Miller CA, Young TI, Thulborn KR. MR quantification of cerebral ventricular volume using a semiautomated algorithm. AJNR 1993; 14:1373-1478.[Abstract]
  10. Filippi M, Horsfield MA, Bressi S, et al. Intra- and inter-observer agreement of brain MRI lesion volume measurements in multiple sclerosis: a comparison of techniques. Brain 1995; 118:1593-1600.[Abstract/Free Full Text]
  11. Wood JR, Green SB, Shapiro WR. The prognostic importance of tumor size in malignant gliomas: a computed tomographic scan study by the Brain Tumor Cooperative Group. J Clin Oncol 1988; 6:338-343.[Abstract]
  12. Blankenberg FG, Teplitz RL, Ellis W, et al. The influence of volumetric tumor doubling time, DNA ploidy, and histologic grade on the survival of patients with intracranial astrocytomas. AJNR 1995; 16:1001-1012.[Abstract]
  13. Greene GM, Hitchon PW, Schelper RL, Yuh W, Dyste GN. Diagnostic yield in CT-guided stereotactic biopsy of gliomas. J Neurosurg 1989; 71:494-497.[Medline]
  14. Johnson PC, Hunt SJ, Drayer BP. Human cerebral gliomas: correlation of postmortem MR imaging and neuropathologic findings. Radiology 1989; 170:211-217.[Abstract/Free Full Text]
  15. Earnest IV. Cerebral astrocytomas: histopathologic correlation of MR and CT contrast enhancement with stereotactic biopsy. Radiology 1988; 166:823-827.[Abstract/Free Full Text]
  16. Hatam A, Bergstrom M, Yu ZY, Granholm L, Berggren BM. Effect of dexamethasone treatment on volume and contrast enhancement of intracranial neoplasms. J Comput Assist Tomogr 1983; 7:295-300.[Medline]
  17. Jeffries BF, Kishore PRS, Singh KS, Ghatak NR, Krempa J. Contrast enhancement in the postoperative brain. Radiology 1981; 139:409-413.[Abstract/Free Full Text]
  18. Thornton Jr AF, Sandler HM, Ten Haken RK, et al. The clinical utility of magnetic resonance imaging in 3-dimensional treatment planning of brain neoplasms. Int J Radiat Oncol Biol Phys 1992; 24:767-775.[Medline]
  19. Brunberg JA, Eldevik OP, Junck L, et al. Gliomas: correlation of MR imaging findings, patient age and pathologic grade with duration of patient survival (abstr) In: Proceedings of the 34th Annual Meeting of the American Society of Neuroradiology. Seattle, Wash: American Society of Neuroradiology, 1996; 151.
  20. Schad LR, Bluml S, Zuna I. MR tissue characterization of intracranial tumors by means of texture analysis. Magn Reson Imaging 1993; 11:889-896.[Medline]
  21. Velthuizen RP, Clarke LP, Phuphanich S, et al. Unsupervised measurement of brain tumor volume on MR images. JMRI 1995; 5:594-605.
  22. Phillips WE, Velthuizen RP, Phuphanich S, Hall LO, Clarke LP, Silbiger ML. Application of fuzzy c-means segmentation technique for tissue differentiation in MR images of a hemorrhagic glioblastoma multiforme. Magn Reson Imaging 1995; 13:277-290.[Medline]
  23. Kao YH, Sorenson JA, Bahn MM, Winkler SS. Dual-echo MRI segmentation using vector decomposition and probability techniques: a two-tissue model. Magn Reson Med 1994; 32:342-357.[Medline]




This Article
Right arrow Abstract Freely available
Right arrow Figures Only
Right arrow Full Text (PDF)
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 Joe, B. N.
Right arrow Articles by Bozik, M. E.
Right arrow Search for Related Content
PubMed
Right arrow PubMed Citation
Right arrow Articles by Joe, B. N.
Right arrow Articles by Bozik, M. E.

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