British Journal of Radiology 74 (2001),243-249 © 2001 The British Institute of Radiology
Image segmentation in treatment planning for prostate cancer using the region growing technique
M Mazonakis, MSc1,
J Damilakis, MSc, PhD2,
H Varveris, MD, PhD3,
P Prassopoulos, MD, PhD4 and
N Gourtsoyiannis, MD, PhD4
1 Department of Medical Physics, University Hospital of Iraklion, PO Box 1352
2 Departments of Medical Physics
3 Radiotherapy
4 Radiology, Faculty of Medicine, University of Crete, Stavrakia, 711 10 Iraklion, Crete, Greece
Correspondence: Dr J Damilakis
 |
Abstract
|
|---|
The purpose of this study was to evaluate the performance of a region growing technique for segmenting prostate, bladder and rectum in CT images of prostate cancer patients. Prostate, bladder and rectum were segmented in all CT images of 10 patients using the region growing technique and manual tracing. Volumes of the above organs computed with the region growing technique were compared with those from manually traced images on a slice-by-slice basis. Measurement reproducibility of both segmentation techniques was evaluated using the data obtained from four independent observers. The region growing technique was 1.5 times faster than manual tracing. There was no statistical difference between the slice volumes of prostate, bladder and rectum obtained by the two segmentation techniques (p>0.05, paired Student's t-test). Correlation between slice volumes of all organs of interest provided both by region growing and by manual tracing was very good (prostate r2=0.84; bladder r2=0.93; rectum r2=0.85). An overall reasonable agreement was found between the two segmentation techniques. The intraobserver and interobserver variations for prostate, bladder and rectum volume segmentation were found to be lower with the region growing technique than with manual tracing. The suggested semi-automatic technique allows the possibility of generating accurate and reproducible segmentation of prostate, bladder and rectum from CT data with great saving in labour.
|
Introduction
|
|---|
Three-dimensional (3D) conformal radiotherapy is being increasingly applied in patients with prostate cancer since it may result in improved targeting of the prostate and significant sparing of normal tissues [1–3]. Patient data are currently acquired as a series of two-dimensional images [4, 5] using CT. An essential part of the conformal treatment planning procedure is to segment target volume and organs-at-risk in all CT images. Bladder and rectum are considered as the organs-at-risk that should be protected against high dosage of radiation during treatment of prostate cancer. 3D reconstruction of the segmented contours allows selection of the optimal beam orientations by visualizing the 3D geometry of the prostate and its relationship with adjacent bladder and rectum.
Image segmentation during treatment planning for prostate cancer is currently generated using the manual tracing technique [6–9]. However, manual segmentation of the target volume and all organs-at-risk on each CT section is a time-consuming procedure, especially as the number of CT images required is usually large [3]. The aim of the present study is to investigate the use of a region growing technique for segmentation of prostate, bladder and rectum volumes in patients with prostate carcinoma.
|
Patients and methods
|
|---|
CT data acquisition
Ten patients with prostate cancer underwent treatment planning CT prior to radiotherapy. Contiguous 5 mm thick axial slices were obtained using a Philips Tomoscan LX CT scanner (Philips, Eindhoven, The Netherlands). A 40 cm reconstruction field-of-view was used in all examinations, which corresponded to 0.78 mmx0.78 mm resolution for a 512x512 matrix size. To simulate the daily treatment situation, patients were scanned in the supine position. They were also instructed to refrain from micturating or drinking any fluid for 1 h prior to the examination to ensure that their bladders were full. 50 ml of dilute contrast medium was administered through a catheter before each CT scan [10].
Image segmentation
All CT scans were transferred to a SUN Sparc 5 computer workstation (Sun Microsystems, Mountain View, CA) for image segmentation using the Analyze software (Mayo Foundation, Rochester, MN). The semi-automatic region growing technique was adopted to segment prostate, bladder and rectum on CT images. The region growing technique connects all pixels within a specified threshold range about a selected seed pixel, and then draws a trace around the connected region. A reference CT image in which prostate, bladder and rectum are sectioned was chosen. Three seed pixels were interactively selected within the interior of the prostate, bladder and rectum to start the segmentation. The seeds were positioned close to the geometrical centre of the structures of interest. A slider was moved to define three independent threshold ranges for prostate, bladder and rectum segmentation. The above threshold ranges were adjusted to the different Hounsfield numbers that characterized each structure of interest. As region growing took place for each particular structure, pixels falling outside the specified threshold range were considered as boundary pixels. Growth of the regions was extended automatically into all slices in which the structures of interest appeared. The results were then checked in a slice-by-slice order. In case the obtained region in a slice differed from the real boundaries of the organ to be segmented, the segmentation process may be reiterated in that particular slice by altering the threshold range. The region growing technique might also include undesired portions of adjacent structures in the segmented region. Segmentation leakage mainly occurs whenever two structures of similar Hounsfield numbers are adjacent. The user may limit the growth of the region by drawing a region encompassing the organ of interest. The above region should be drawn as close as possible to the boundaries of the organ where the segmentation leakage appears. This manually drawn region was used as a hard constraint for the region growing process.
To evaluate the performance of the region growing technique, the ten sets of CT images were also segmented by manually tracing prostate, bladder and rectum. Manual tracing is the conventional technique for segmenting target volume and organs-at-risk during conformal treatment planning. Segmentation of the three organs of interest using both techniques was performed by a senior radiotherapist.
Volumetric measurements
Prostate, bladder and rectum volumes obtained from the region growing technique were compared with those determined by manual segmentation on a slice-by-slice basis. All volume measurements were generated using the Analyze software. First, the slice volume of prostate, bladder and rectum was measured in each single segmented CT image by counting the number of pixels enclosed by the contours of the above organs, and taking into account the magnification factor known from CT scale and slice thickness. Subsequently, the total volume for each organ of interest was calculated by summing the above slice volumes.
Measurement reproducibility
The senior radiotherapist repeated the segmentation of prostate, bladder and rectum from the complete CT data set of 10 patients using the region growing technique and manual tracing twice more with each technique. There was a 2-week delay between the first and second segmentation. The delay between the second and third segmentation was at least 1 week. The slice volume of the above contours was measured each time. Therefore, each single slice volume was calculated three times using the region growing technique and three times using manual tracing. Intraobserver coefficient of variation values for prostate, bladder and rectum segmentation using the region growing technique and manual tracing were obtained from the consecutive slice volume measurements.
A second radiotherapist and three medical physicists were asked to independently segment prostate, bladder and rectum on the contiguous CT slices of all 10 patients. The above observers were well trained in treatment planning of prostate cancer. Each observer segmented the three organs of interest once with the region growing technique and once using manual tracing. Slice volume measurements of prostate, bladder and rectum obtained from the four independent observers were used to evaluate interobserver variation for both segmentation techniques.
Statistical analysis
A two-tailed paired Student's t-test was used to determine the statistical differences between the slice volumes of prostate, bladder and rectum obtained from manual tracing and from the region growing technique on a slice-by-slice basis. Linear regression analysis was also performed to determine the relationship among slice volumes calculated from the region growing technique and those calculated by manually tracing for each of the organs of interest. Average slice volumes of prostate, bladder andrectum obtained from the three separate delineations performed by the senior radiotherapist using both segmentation techniques were used for the statistical analysis and for the calculation of total organ volume. For both t-test and regression analysis, a p-value below 0.05 was considered as statistically significant.
|
Results
|
|---|
Figure 1
illustrates typical results obtained using the region growing technique for the segmentation of prostate, bladder and rectum volumes. The time for the segmentation of the above organs by applying the region growing technique ranged from 7.0–10.9 min, with an average time of 8.4 min. The above segmentation time includes the required time to correct or repeat organ delineation whenever the results were not satisfactory. The respective segmentation times using manual tracing varied from 8.4–14.8 min, with an average time of 12.3 min.
View larger version (84K):
[in this window]
[in a new window]
|
Figure 1. Prostate, bladder and rectum segmentation by applying the region growing technique in a pelvic CT image.
|
|
Total prostate, bladder and rectum volumes using the region growing technique and manual tracing are presented in Table 1. The mean ratios±standard deviation of prostate, bladder and rectum total volume derived from manual tracing over the corresponding volume derived from the region growing technique were 0.98±0.08, 1.01±0.03 and 0.98±0.05, respectively. The above figures provide strong evidence regarding agreement between the region growing technique and manual tracing. The paired Student's t-test showed that there was no statistical difference between the slice volumes of prostate, bladder and rectum obtained from the region growing technique and those from manual tracing. The two-tailed p-value was equal to 0.18, 0.43 and 0.20 for prostate, bladder and rectum statistical analysis, respectively. Linear regression analysis showed a strong relationship between organ slice volumes provided by both segmentation techniques (prostate, y=0.98x+0.18, p<0.0001, r2=0.84, Figure 2; bladder, y=0.99x+0.12, p<0.0001, r2=0.93, Figure 3; rectum, y=0.91x+0.74, p<0.0001, r2=0.85, Figure 4). The maximum local distances between prostate, bladder and rectum contours obtained from the two segmentation techniques were 13 mm, 9 mm and 10 mm, respectively. The most common differences among the contours obtained from the region growing technique and manual tracing were in the regions around the seminal vesicles and between prostate and rectum. No considerable differences were observed at the prostate–bladder interface owing to the use of intravesical contrast medium.
View this table:
[in this window]
[in a new window]
|
Table 1. Prostate, bladder and rectum volume estimations obtained from manual tracing and from the region growing technique
|
|
View larger version (16K):
[in this window]
[in a new window]
|
Figure 2. Linear regression analysis of prostate slice volume obtained with the region growing technique and with manual tracing for 84 CT slices.
|
|
View larger version (18K):
[in this window]
[in a new window]
|
Figure 3. Linear regression analysis of bladder slice volume obtained with the region growing technique and with manual tracing for 147 CT slices.
|
|
View larger version (19K):
[in this window]
[in a new window]
|
Figure 4. Linear regression analysis of rectum slice volume obtained with the region growing technique and with manual tracing for 153 CT slices.
|
|
Bladder and rectum could be delineated automatically without any correction in four patients, and one patient, respectively. For prostate segmentation, corrections were applied in all cases. Segmentation leakage was observed in 36%, 20% and 28% of the total number of slices containing prostate, bladder and rectum, respectively. Regarding bladder and rectum, segmentation leakage was mainly due to the difficulty of the region growing technique to connect the bladder or rectum walls with the fillings of these organs. For that reason, the range of threshold values used for bladder or rectum delineation was increased and subsequently resulted in segmentation leakage. For prostate segmentation, leakage was due to the poor contrast between prostate volume and the surrounding soft tissues (Figure 5). To constrain the extent of the above regions, a limit was drawn around the organs of interest and the region growing technique was started again. Moreover, application of a new threshold range was needed occasionally to include disconnected regions in the final contours of prostate. The above correction was made in 10% of the total number of slices imaging prostate. The adjusted threshold values for prostate segmentation varied from patient to patient. Analysis of the complete set of CT images showed that prostate, bladder and rectum appeared on a mean of 8 (range 7–9), 15 (13–17) and 15 (13–18) slices, respectively. Therefore, for a typical CT examination only 4, 3 and 4 slices containing prostate, bladder and rectum, respectively, should be corrected.
View larger version (82K):
[in this window]
[in a new window]
|
Figure 5. Prostate delineation using the region growing technique results in segmentation leakage into the adjacent soft tissues. The leakage may be constrained by drawing a region around the prostate.
|
|
Measurements of intraobserver and interobserver variation are shown in Tables 2 and 3, respectively. The region growing technique provided more reproducible segmentation of prostate, bladder and rectum than the conventional technique of manual tracing.
View this table:
[in this window]
[in a new window]
|
Table 2. Intraobserver coefficient of variation values (%) for prostate, bladder and rectum volume segmentation using the region growing technique and manual tracing
|
|
View this table:
[in this window]
[in a new window]
|
Table 3. Interobserver coefficient of variation values (%) for prostate, bladder and rectum volume segmentation using the region growing technique and manual tracing
|
|
|
Discussion
|
|---|
MRI of the pelvic region provides better contrast between soft tissues compared with CT. Moreover, MRI is considered to be superior to non-contrast CT in the definition of prostate volume [11, 12]. However, the use of MRI for treatment planning in patients with prostate cancer is very limited. This is owing to the lack of tissue density information for the correction of inhomogeneities used in dose calculation as well as to the presence of intrinsic system-related and object-induced image MR distortions [13]. Reported experience has also suggested that non-contrast CT may lead to underestimation of prostate extent and to significant underdosage [10]. Administration of intravesical contrast medium is required to distinguish the base of bladder from prostate. The above leads to an enhancement of prostate visualization and definition of tumour volume. This was the reason for routinely performing contrast CT in patients with prostate cancer and using these scans as treatment planning data.
The region growing segmentation technique was evaluated by comparing it with the manual tracing method in 10 prostate cancer patients. Differences in prostate, bladder and rectum slicevolume calculations obtained by the two segmentation techniques were not statistically significant. An excellent correlation was found between bladder slice volumes obtained from the region growing technique and from manual tracing. Correlation coefficients between the two segmentation techniques for both prostate and rectum slice volume calculations were lower than that for bladder slice volumes. This was owing to the unclear depiction of prostate edges that is often presented in CT images and to the more intricate morphology of rectum in relation with that of bladder [14]. The above may cause difficulties in determining the real boundaries of prostate and rectum.
Several studies have investigated the reproducibility of the manual tracing technique in defining the target volume and organs-at-risk during treatment planning of prostate cancer[6–9]. However, studies comparing the reproducibility of manual tracing with the region growing technique or with other segmentation techniques have been lacking from the literature. The current study showed that the intraobserver and interobserver variation was superior with the region growing technique compared with manual tracing for prostate, bladder and rectum volume segmentation from CT data. This may be attributed to the reduced user interaction required for efficient segmentation of the above organs using the semi-automatic technique. There was still intraobserver variation even when using the region growing technique owing to the different threshold ranges used and to the different seed positions selected during semi-automatic segmentation.
Reproducibility measurements revealed that the coefficient of variation values for prostate segmentation were considerably higher than those for bladder and rectum, irrespective of the selected segmentation technique. We consider that prostate segmentation is a more difficult task compared with that for bladder and rectum, requiring user perception of prostate boundary, experience and sufficient knowledge of the regional anatomy.
Only a few reports have discussed the use of the region growing technique during radiation treatment planning. Siverwright and Elliott [15]applied the region growing technique for segmenting all therapy-relevant volumes in brain CT and MR images. Neal et al [16] used the region growing technique for segmenting pelvic CT images. They reported that the semi-automatic technique has several limitations regarding the segmentation of soft tissues and that manual tracing may be often considered as the method of choice. Moreover, they found that the time required for the segmentation of soft tissues in the pelvic region using the region growing technique is not significantly different from the time needed using manual tracing. The current study revealed that the semi-automatic technique can reduce segmentation time by a factor of 1.5, providing reliable contouring of soft tissues. This difference may be attributed to both the intravesical contrast medium introduced before CT examination for the accurate determination of the prostate and bladder boundaries and to the reduction of segmentation leakage by limiting the growth of the region.
|
Conclusion
|
|---|
In this study, the region growing technique was compared with the conventional technique of manual tracing and appeared to provide several advantages for segmenting prostate, bladder and rectum from contrast CT data. The semi-automatic technique was more reproducible and faster than manual delineation. An overall reasonable agreement was found between the two segmentation methods, indicating that region growing is an accurate and reliable technique. We consider that the management of prostate cancer patients can potentially benefit from the adoption of the region growing technique as the standard method for image segmentation during the treatment planning process.
Received for publication May 15, 2000.
Revision received August 31, 2000.
Accepted for publication September 25, 2000.
|
References
|
|---|
-
Dearnaley DP, Khoo VS, Norman AR, Meyer L, Nahum A, Tait D, et al. Comparison of radiation side-effects of conformal and conventional radiotherapy in prostate cancer: a randomised trial. Lancet 1999;353:267–72.[Medline]
-
Leibel SA, Zelefsky MJ, Kutcher GJ, Burman CM, Kelson S, Fuks Z. Three-dimensional conformal radiation therapy in localized carcinoma of the prostate: interim report of a phase 1 dose escalation study. J Urol 1994;152:1792–8.[Medline]
-
Perez CA, Michalski J, Ballard S, Drzymala R, Kobeissi BJ, Lockett MA, et al. Cost benefit of emerging technology in localized carcinoma of the prostate. Int J Radiat Oncol Biol Phys 1997;39:875–83.[Medline]
-
Dobbs HJ, Husband JE. The role of CT in the staging and radiotherapy treatment planning of prostatic tumours. Br J Radiol 1985;58:429–36.[Abstract]
-
Gerber RL, Smith AR, Owen J, Hanlon A, Wallace M, Hanks G. Patterns of care survey results: treatment planning for carcinoma of the prostate. Int J Radiat Oncol Biol Phys 1995;33:803–8.[Medline]
-
Cazzaniga LF, Marinoni MA, Bossi A, Bianchi E, Cagna E, Cosentino D, et al. Interphysician variability in defining the planning target volume in the irradiation of prostate and seminal vesicles. Radiother Oncol 1998;47:293–6.[Medline]
-
Fiorino C, Reni M, Bolognesi A, Cattaneo GM, Calandrino R. Intra- and inter-observer variability in contouring prostate and seminal vesicles: implications for conformal treatment planning. Radiother Oncol 1998;47:285–92.[Medline]
-
Lebesque JV, Bruce AM, Kroes AP, Touw A, Shouman T, Van Herk M. Variation in volumes, dose–volume histograms and estimated normal tissue complication probabilities of rectum and bladder during conformal radiotherapy of T3 prostate cancer. Int J Radiat Oncol Biol Phys 1995;33:1109–19.[Medline]
-
Valicenti RK, Sweet JW, Hauck WW, Hudes RS, Lee T, Dicker AP, et al. Variation of clinical target volume definition in three-dimensional conformal radiation therapy for prostate cancer. Int J Radiat Oncol Biol Phys 1999;44:931–5.[Medline]
-
Sharma R, Duclos M, Chuba PJ, Shamsa F, Forman JD. Enhancement of prostate tumor volume definition: a three-dimensional dosimetric evaluation. Int J Radiat Oncol Biol Phys 1997;38:575–82.[Medline]
-
Khoo VS, Padhani AR, Tanner SF, Finnigan DJ, Leach MO, Dearnaley DP. Comparison of MRI with CT for the radiotherapy of prostate cancer: a feasibility study. Br J Radiol 1999;72:590–7.[Abstract]
-
Roach M, Failace-Akazawa P, Malfati C, Holland J, Hricak H. Prostate volumes defined by magnetic resonance imaging and computerized tomographic scans for three-dimensional conformal radiotherapy. Int J Radiat Oncol Biol Phys 1996;35:1011–8.[Medline]
-
Khoo VS, Dearnaley DP, Finnigan DJ, Padhani A, Tanner SF, Leach MO. Magnetic resonance imaging (MRI): considerations and applications in radiotherapy treatment planning. Radiother Oncol 1997;42:1–15.[Medline]
-
Mazonakis M, Damilakis J, Varveris H. Bladder and rectum volume estimations using CT and stereology. Comp Med Imag Graph 1998;22:195–201.[Medline]
-
Siverwright GJ, Elliott PJ. Interactive region and volume growing for segmenting volumes in MR and CT images. Med Inform 1994;19:71–80.
-
Neal AJ, Siverwright G, Bentley R. Evaluation of a region growing algorithm for segmenting pelvic computed tomography images during radiotherapy treatment planning. Br J Radiol 1994;67:392–5.[Abstract]
This article has been cited by other articles:
|
|
|
|
 
C A McBAIN, C J MOORE, M M L GREEN, G PRICE, J S SYKES, A AMER, V S KHOO, and P PRICE
Early clinical evaluation of a novel three-dimensional structure delineation software tool (SCULPTER) for radiotherapy treatment planning
Br. J. Radiol.,
August 1, 2008;
81(968):
643 - 652.
[Abstract]
[Full Text]
[PDF]
|
|
|
Top
Abstract
Introduction
