J. Biomedical Science and Engineering, 2009, 2, 16-19
Published Online February 2009 in SciRes. http://www.scirp.org/journal/jbise JBiSE
16
Modified technique for volumetric brain tumor
measurements
Yasser M. Salman1
1Egyptian Armed Forces, Correspondence should be addressed to Yasser M. Salman ([e-mail])
Received March 22nd, 2008; revised November 24th, 2008; accepted December 8th, 2008
ABSTRACT
Quantitative measurements of tumor response
rate in three dimensions (3D) become more re-
alistic with the use of advanced technology im-
aging during therapy, especially when the tumor
morphological changes remain subtle, irregular
and difficult to assess by clinical examination.
These quantitative measurements depend strongly
on the accuracy of the segmentations methods
used. Improvements on such methods yield to
increase the accuracy of the segmentation
process. Recently, the essential modification in
the Traditional Region Growing (T-RG) method
has been developed and a “Modified Region
Growing Method” (MRGM) has been presented
and gives more accurate boundary detection
and holes filling after segmentation. In this pa-
per, the new automatic calculation of the volu-
metric size of brain tumor has been imple-
mented based on Modified Region Growing
Method. A comparative study and statistical
analysis performed in this work show that the
modified method gives more accurate and better
performance for 3D volume measurements. The
method was tested by 7 fully investigated pa-
tients of different tumor type and shape, and
better accurate results were reported using
MRGM.
Keywords: Region Growing, Modified Region
growing, and Volumetric Brain Tumor Meas-
urements
1. INTRODUCTION
More recent studies have shown that 3D quantitative
imaging-based method of tumor size assessment using
MRI is highly accurate in determining actual tumor size
[1,2] and may be superior to clinical palpation in pre-
dicting local tumor control [3,4,5].
Manual region of interest (ROI) volumetry method is
a standard approach of 3D quantitative measurements
which is very precisely to detect tumor contours and it is
considered to be the “gold standard” because region of
interest (ROI) is segmented manually by the expert radi-
ologists. The disadvantage of this method is, it requires
intensive time because of its dependency on manual
segmentation process. Segmentation of ROI in volumet-
ric medical images is still a challenging problem, and
solutions usually have been based on either model-based
deformation of templates or intensity thresholding such
as region growing method [6,7]. Recent studies prove
that the region growing is an effective approach and less
computation intensive for image segmentation especially
for the homogenous regions [8,9,10,7,11]. The primary
disadvantage of region growing method is the partial
volume effect [12,13]. Partial volume effect limits the
accuracy of MR brain image segmentation. It blurs the
intensity distinction between tissue classes at the border
of the two tissues types, because the voxel may represent
more than one kind of tissue types [14,15]. S. Lakare [12]
et. al, developed effective modifications in region grow-
ing technique. This modification called Modified region
growing method (MRGM) used to remove the partial
volume effects and to incorporate gradient information
for more accurate boundary detection and filling holes
occurred after segmentation.
The software implemented in this paper involves the
proposed quantitative measurement of brain tumors
based on MRGM segmentation and the visualization tool
to monitor and reconstruct the brain tumor in 2D and 3D
space. For testing and validation, the proposed MRGM
method has been compared with traditional region
growing method against experts’ manual trace method,
and the statistical analysis was performed to evaluate the
proposed method against TRG and golden tracing
method by experts.
In this paper, section 2 describes the settings of MR
image acquisition, the details of patient population and
noise reduction technique as data pre-processing. In sec-
tion 3, we present the segmentation and calculation used
for assessment of the brain tumor measurements, also
this section describes the statistical and data analyses
used to evaluate and validate the proposed method. Sec-
tion 4 describes the results of brain tumor visualization
in 2D and 3D spaces, the brain volume calculation using
the proposed method compared with other method, and
the result of statistical and data analysis. Section 5 pre-
sents the merits and demerits of the proposed method
SciRes Copyright © 2009

Y. M. Salman et al. / J. Biomedical Science and Engineering 2 (2009) 16-19 17
SciRes Copyright © 2009 JBiSE
compared to each others and concludes allover the work
done in this study.
2. MATERIALS
2.1. MRI Image Acquisitions
MRI images were acquired on a 1.5T using T1-weighted
contrast images. A resolution of 256x256x60 with a
voxel resolution of 0.93mm x0.93mm with slice thick-
ness of 3mm was set.
2.2. Patient Population
The study group consists of seven patients scanned with
228 MRI axial slices with biopsy histologically proved
Glioblastoma Multiforme (GBM) and Low Grade As-
trosytoma brain tumors types.
2.3. Pre-processing
Noise presented in the image can reduce the capacity of
the region growing filter to grow large regions, or may
result in a fault edges. When faced with noisy images, it
is usually convenient to pre-process and enhance the
image by using a noise reduction filter. Gaussian
smoothing filter [16] is commonly used as an approach
for noise reduction. The size of the neighborhood mask
can be set by the user. The quality of the enhanced Gaus-
sian filtered images is much better as the contrast be-
tween tumor and surrounding tissue is high as well tu-
mors studied are of homogenous borders (regular convex
shapes).
3. METHODS
3.1. Traditional Region Growing Segmenta-
tion Method
The Traditional region growing algorithm based on ex-
traction of a connected set of pixels whose pixel intensi-
ties are consistent with the pixel statistics of a seed point.
The mean and variance across 8-connected neighbor-
hood are calculated for a seed point [16,17].
3.2. Modified Region Growing Segmentation
Methods (MRGM)
To understand the basic principles behind MRGM, we
first reviewed the S. Lakare, et al. [12]. In their work,
MRGM provided for object segmentation has been im-
plemented. This implementation allowed stable bound-
ary detection when the gradients suffer from intersection
variations and gaps.
3.3. Volume Calculation
3.3.1. Manual ROI Volumetry
Figure 1 shows the area inside the outline that was
manually segmented, labeled, calculated, and multiplied
by the MR slice thickness plus the interslice gap to cal-
culate a per-slice tumor volume. The total tumor volume
was obtained by summing the volume calculations for all
slices.
Figure 1. Manual trace method
3.3.2. TRG and MRGM Calculations for Growing
Tumor
After tumor region has been segmented using both T-RG
and MRGM segmentation techniques, the tumor volume
calculations are performed in this segmented region. To
calculate the volume of segmented tumor region, the
automatic labeling of the entire volumetric tumor region
has been done slice by slice and by calculating the total
number of pixels into the labeled regions. Areas of the
labeled region were calculated and multiplied by the MR
slice thickness plus the interslice gap to obtain a per-slice
tumor volume. The total tumor volume was then ob-
tained by summing the tumor-bearing slices.
3.4. Statistical Consideration and Data
Analysis
The comparative study has been done using T-RG,
MRGM and Manual ROI volumetry methods. Two ob-
servers (expert radiologist and a none radiologist) inde-
pendently rated each MR image twice by using manual
and the two automated measurement methods. Observers
have performed the comparison between inter and intra-
observer reliability and image processing computational
times for both methods were reported. To compare the
intra and inter-observer reliability of the three measure-
ment methods, we used the following agreement index
[21], AI as follow:
AI=1―
2/)( ba
ba
XX
XX
−
−
(1)
For inter-observer agreement calculation, Xa was the
measurement obtained by observer 1 and Xb, the meas-
urement obtained by observer 2 with the same technique
on the same case. For intra-observer agreement calcula-
tion, Xa is the measurement made during the first trial,
and Xb is the measurement made during the second trail
by the same observer with the same technique on the
same case. Intra-observer and inter-observer agreement
indexes were calculated for each image, to increase sen-
sitivity performance. A value of “1” indicates the perfect
agreement and value of “0” indicates no agreement.
4. RESULTS
Results showed that the proposed quantitative measure-

18 Y. M. Salman et al. / J. Biomedical Science and Engineering 2 (2009) 16-19
SciRes Copyright © 2009 JBiSE
ment of brain tumors based on MRGM has a higher ac-
curacy and precision against traditional region growing
method compared to expert manual computation. This
yields to better effect in the assessment of brain tumor
measurements.
4.1. PC Based Package
We improve our previous work [10] PC based software
package implemented using three programming devel-
opment environment, as VTK[18], ITK [20] and Visual
C++, to segment, visualize and calculate the tumor vol-
ume at different instants of tumor growing or shrinking.
Figure 2 shows the result of tumor segmentation using
T-RG and MRGM. Figure 3 shows the 3D reconstruc-
tion of segmented tumor region for MRGM method,
using surface reconstruction algorithm [19].
4.2. Tumor Volume Measurement Accuracy
The Relative Error (RE) for tumor volume can be calcu-
lated as Where Pq tumor volume using 3D quantitative
methods (Traditional and modified region growing
methods), Pm is tumor volume calculated using expert
RE(%)=(
m
mq
P
PP−)×100 (2)
Figure 2. (a) Results of T-RG segmentation; (b) Results of
MRGM segmentation
Figure 3. Extracted tumor in 3D space
manual tracing method. Table 1 shows calculation results
and their relative errors for the different quantitative
methods compared with the gold standard manual
method. These results had been obtained by observer 1
and Figure 4 summarizes these relative errors in chart.
Also, Table 2 shows calculation results and their rela-
tive errors for the different methods compared with the
standard manual method. These results obtained by ob-
server 2 and Figure 5 summarizes these relative errors in
chart.
4.3. Observer Agreements
The intra and inter-observer agreement indexes for the
two observers are summarized in Table 3. As shown in
the table, there are no significant differences in mean
intra and inter-observer agreement between the manual
method, traditional and modified region growing methods.
Table 1. Volumetric calculation for brain tumor in cm3 using the
different calculation methods and the relative errors for each
method compared with manual segmentation method (data rated
using first observer)
Volume in cm3 Relative Error%
Cases T-RGMRGMManual T-RG MRGM
Case#15.5876.2816.125 8.7836 2.546
Case#28.7829.84510.234 14.188 3.801
Case#311.40612.06611.234 1.5310 7.406
Case#413.68712.75512.987 5.3900 1.786
Case#514.01414.28414.123 0.7717 1.139
Case#615.22316.11616.987 10.384 5.127
Case#716.12216.69516.354 1.4186 2.0851
0
2
4
6
8
10
12
14
16
1234567
C A S E S
Relative Errors (%)
T-RG
MRGM
Figure 4. Relative errors for T-RG and MRGM compared with
manual segmentation method (data rated using first observer)
Table 2. Volumetric calculation for brain tumor in cm3 using the
different calculation methods and the relative errors for each
method compared with manual segmentation method (data rated
using second observer)
Volume in cm3 Relative Error%
Cases T-RGMRGMManual T-RG MRGM
Case#15.7636.2235.987 3.7414 3.941
Case#29.0889.63710.345 12.150 6.843
Case#310.83612.96311.897 8.9182 8.960
Case#413.96212.59712.235 14.115 2.958
Case#513.50914.10814.678 7.9643 3.883
Case#614.7316.12216.544 10.964 2.550
Case#714.83116.36316.786 11.646 2.519
(a) (b)

Y. M. Salman et al. / J. Biomedical Science and Engineering 2 (2009) 16-19 19
SciRes Copyright © 2009 JBiSE
0
2
4
6
8
10
12
14
16
1234567
C A S E S
Relative Errors (%)
T-RG
MRGA
Figure 5. Relative errors for T-RG and MRGM compared with
manual segmentation method (data rated using second observer)
Table 3. Results of intra- and inter observer agreement index
Index Observer
MRGM T-RG Manual
1 0.9899
±0.019
0.9374
±.0.058
0.9665
±0.0183
Intra-Observer
Agreement 2 0.9773
±0.026
0.9728
±0.0215
0.9744
±0.0122
Inter-Observer
Agreement 1,2 09689
±0.0342
0.9587
±0.0207
0.9559
±0.0237
5. DISCUSSION AND CONCLUSION
Recent attention has been given to improve the segmen-
tation methods in order to increase brain tumor meas-
urements accuracy [8]. In this paper the essential modifi-
cations has been implemented in the region growing al-
gorithm and presented as “Modified Region Growing
Method” (MRGM). These modifications overcome the
partial volume effect artifacts. Hence, applying MRGM
to segment brain tumors will increase the accuracy of the
volumetric measurements of the brain tumors. A volu-
metric measurement based on T-RG, MRGM and manual
segmented ROI volumetry method have been applied in
MR volumetric data total of 228 MRI slices of Glioblas-
toma Multiforme (GBM) and Low Grade Astrosytoma
brain tumors scanned from 7 patients. The results of
comparative study showed that the MRGM produces
lower relative errors than T-RG method. Also, it has a no
significant differences in inter- and intra-observer
agreement index. These results ensure that the accuracy
of the volumetric measurements of the brain tumor have
been improved using MRGM which yields to great ef-
fects in many applications in tumor prognosis and ther-
apy such as early signs of treatment failure in radiother-
apy and chemotherapy to avoid unneeded higher doses
of radiation to patient, tumor growth rate and early de-
tection of tiny changes in tumor size in which it is diffi-
cult to be detected in traditional visual metric and clini-
cal examination measurements.
ACKNOWLEDGMENT
We would like to acknowledge ISBR. They support us to obtain the
data. Data was provided by the Center for Morphometric Analysis at
Massachusetts General Hospital.
REFERENCES
[1] H. Hricak, C. G. Lacey, L. G. Sandles, Y. C. F. Chang, M. L.
Winkler, J. L. Stern. (1988) Invasive cervical carcinoma: Compari-
son of MR imaging and surgical findings, Radiology 166: 623-631.
[2] K. Hatano, Y. Sekiya, H. Araki, et al. (1999) Evaluation of the
therapeutic effect of radiotherapy on cervical cancer using magnetic
resonance imaging, Int. J. Radiat. Oncol. Biol. Phys. 45: 639-644.
[3] N. A. Mayr, W. T. C. Yuh, et al. (1997) Tumor size evaluated by
pelvic examination compared with 3-D MR quantitative analysis
in the prediction of outcome for cervical cancer, Int J Radiat.
Oncol. Biol. Phys. 39: 395-404.
[4] H. Hricak, J. Quivey, et al. (1993) Phillips T. Carcinoma of the
cervix: Predictive value of clinical and magnetic resonance
(MR) imaging assessment of prognostic factors, Int. J. Radiat.
Oncol. Biol. Phys. 27: 791-801.
[5] N. A. Mayr, E. T. Tali, et al. (1993) Cervical cancer: Application
of MR imaging in radiation therapy, Radiology 189: 601-608.
[6] M. Kass, A. Witkin, and D. Terzopoulos, (1987) “Snakes: Active
contour models,” Int. J. Computer Vision, Vol. 1, No. 4, 321-
331.
[7] R. Adams and L. Bischof, (1994) Seeded Region Growing, IEEE
Transactions on Image processing, Vol.16, No.6, 641-47.
[8] F. Vincent, H. Chong, J. Y. Zhou, B. James, K. Khoo, J.
Huang, T. K. Lim. (2004) Tongue carcinoma: Tumor volume
measurement, Int. J. Radiation Oncology Biol. Phys., Vol. 59,
No. 1, 59-66.
[9] Y. M. Salman, A. M. Badawi, M. A. Assal, S. M. Alian, (2005)
New automatic technique for tracking brain tumor responsem ,
International Conference on Biological and Medical Physics,
UAE.
[10] Salman Y. M., A. M. Badawi, (2005) Validation Techniques for
Quantitative Brain Tumor Measurements, The 27th Annual In-
ternational Conference of the IEEE Engineering in Medicine and
Biology Society, China, pp. 7048-7051.
[11] S. C. Zhu and A. Yuille, (1996) Region competition: Unifying
snakes, region growing and Bayer/MDL for multiband image
segmentation, IEEE Transactions on Pattern Analysis and Ma-
chine Intelligence, Vol. 18, No. 9, pp. 884-900.
[12] M. Sato, S. Lakare, M. Wan, and A. Kaufman, (2000) A Gradient
Magnitude Based Region Growing Algorithm For Accurate
Segmentation, In Proc. International Conference on Image Proc-
essing, Vol.3, 448-451.
[13] S. Lakare, (2000) 3D Segmentation Techniques for Medical
Volumes, Center of Visual Computer, state university of NY,
Stony Brooks.
[14] Jauhiainen, Tommi, Jarvinen, et al., (1998) MR Gradient Echo
Volumetric Analysis of Human Cardiac Casts: Focus on the
Right Ventricle, Journal of Computer Assisted Tomography,
Vol.22, No.6, 899-903.
[15] J. M. Links , L. D. Beach , B. Subramaniam, (1998) Edge com-
plexity and partial volume effects, Journal of Computer Assisted
Tomography, Vol. 22, No.3, 450-458.
[16] R. C. Gonzalez and R. E woods, (1992) Digital Image Process-
ing, Addison-Wesley, USA.
[17] Y. L. Chang, X. Li, (1994) Adaptive Image Region-Growing,
IEEE Trans. on Image Processing, Vol. 3, No. 6, 868-872.
[18] W. Schroeder, K. Martin, and B. Lorensen B, (1998) the Visuali-
zation Toolkit. New Jersey, Prentice Hall.
[19] W. E Lorensen, H. E Cline, (1987) Marching cubes: a high reso-
lution 3D surface construction algorithm, Computer Graphics
(SIGGRAPH 87 Proceedings), Vol. 21, 1693-169.
[20] L. Ibanes, W. Schroeder, L. Ng, (2003) Insight Segmentation and
Registration Toolkit (ITK) Software Guide.
[21] B. N. Joe, et al., (1999) Brain Tumor Volume Measurement:
Comparison of manual and semi automated methods, Radiology,
No.212, 811-816.