AUTOMATIC ORGAN DELINEATION OF COMPUTED

TOMOGRAPHY IMAGES FOR RADIOTHERAPY PLANNING

IN PROSTATE CANCER

An Overview

Celeste Marques Oliveira

1,2,3

and Pedro Pereira Rodrigues

1,4

Faculty of Medicine of the University of Porto, Al. Prof. Hernâni Monteiro, 4200-319, Porto, Portugal

Faculty of Sciences of the University of Porto, Rua do Campo Alegre, s/n, 4169-007, Porto, Portugal

Department of Radiotherapy, Portuguese Institute of Oncology Francisco Gentil

Rua Dr. António Bernardino de Almeida, 4200-072, Porto, Portugal

LIAAD - INESC Porto, L.A. & CINTESIS - Center for Research in Health Technologies and Information Systems

University of Porto, Porto, Portugal

Keywords: Radiotherapy planning, Prostatic neoplasms, Computer-assisted methods, Tomography X-ray computed,

Image processing.

Abstract: Prostate cancer is a common cancer worldwide and a leading cause of death. Radiotherapy is usually the

first-line treatment for patients with slow-growing cancer that is confined to the prostate. In Radiation

Therapy Planning (RTP), the recognition and outlining of clinical volumes in computed tomography (CT)

images are one of the most time-consuming steps carried out by human experts. The aim of this review is to

identify and summarize evidence of the use of automatic organ delineation of CT images for radiotherapy

planning in prostate cancer. From the literature search, a total of seven studies, reported between 1994 and

2009, were selected. We associate the selected studies in order to compare results, in spite of their

differences in methodology and outcome evaluators. Most of the studies conclude that the automatic

approach is faster, while having equivalent accuracy to manual method. Concerning the observer’s

variability, automatic segmentation reaches significant gains in reproducibility. As future directions, it is

recommended the improvement of the segmentation algorithms in the delineation of problematic soft tissues

and future validation studies with large scale trials and possible studies of meta-analysis in the specific

problems.

1 INTRODUCTION

Prostate cancer is a common cancer worldwide and a

leading cause of death. According World Health

Organization is accounting for about 250,000 new

cases annually. Radiotherapy is usually the first-line

treatment for patients with slow-growing cancer that

is confined to the prostate. It represents a curative

treatment option in these patients (Boehmer, D. et

al., 2006) and the three-dimensional (3D) conformal

radiotherapy is being increasingly applied since it

may result in improved targeting of the prostate and

significant sparing of normal tissues.

In the Radiation Therapy Planning (RTP), the

recognition and outlining of clinical target volume

and adjacent organs at risk, in Computed

Tomography (CT) images, are one of the most time-

consuming steps carried out routinely by human

experts (Huyskens, D.P. et al., 2009, Haas, B. et al.,

2008). It is only by displaying these that the

dosimeters can devise an optimal plan to the

prescribed dose while minimizing radiation of

adjacent non-target tissues thereby maximizing the

therapeutic gain of treatment (Neal, A.J. et al.,

1994). Usually, they outline the boundaries of the

structure by a process of continuous contour drawing

on most (or all) slices of CT image set using a

computerized Treatment Planning System (TPS).

This is a laborious and subjective task ultimately

dependent on the clinician´s expert eye, which is

also prone to inconsistency and variability (Mcbain,

C.A. et al., 2008, Pekar, V. et al., 2004).

482

Marques Oliveira C. and Pereira Rodrigues P..

AUTOMATIC ORGAN DELINEATION OF COMPUTED TOMOGRAPHY IMAGES FOR RADIOTHERAPY PLANNING IN PROSTATE CANCER - An

Overview.

DOI: 10.5220/0003127904820485

In Proceedings of the International Conference on Health Informatics (HEALTHINF-2011), pages 482-485

ISBN: 978-989-8425-34-8

 2011 SCITEPRESS (Science and Technology Publications, Lda.)

TPS for irradiation of malignant neoplasm

requires CT images due to the similar physical

behavior of the radiation used for imaging and for

treatment. These images are limited of contrast for

the structures of interest, ruling out many

approaches to segmentation for this application

(Quicken, M., 2000). The emergence of imaging and

adaptive radiotherapy is producing amounts of

image data whose manual delineation slice by slice

has become infeasible (Boehmer, D. et al., 2006).

The need of efficient and robust segmentation tools

is even increasing (Li, T. et al., 2006). Segmentation

is an image processing term literally implying the

breaking down of an image into smaller parts. In the

context of RTP this comprises the contouring of

important volumes. At the present there is no

universally accepted segmentation method that is

proven to work on a large representative image

database (Bueno, G. et al., 2001).

Despite all the advances in imaging for RTP,

some anatomical regions remain indistinct and it is

very difficult to delineate. This is usually because of

an inability to differentiate the region of interest,

from the adjacent structures of similar grey-scale

signal density, for e.g. the base of the prostate gland

at its interface with the base of the bladder (Mcbain,

C.A. et al., 2008). Bladder, rectum and femoral

heads should be delineated in the TPS to achieve a

protection against high dosage of radiation trough a

selection of the optimal beam orientations by

visualizing 3D reconstruction (Mazonakis, M. et al.,

2001).

For a better understanding of the automatic

segmentation used in TPS see the authors

(Freedman, D. et al., 2005, Quicken, M., 2000).

The aim of this article is to review, to identify

and summarize evidence from scientific studies to

obtain an overview about the use of automatic organ

delineation of CT images for RTP in prostate cancer.

2 METHODOLOGY

The comprehensive literature search of this review

was performed by using PubMed®. The final search

was executed on December 31, 2009 using the query

presented in Figure 1. The selection over the data

collection was performed by one reviewer that

examined the related articles.

This strategy was based on these following steps:

apply the query (n=25), exclude the articles

inadequate by reading the title and the abstract

(n=21); review the chosen articles (n=4), add some

relevant articles from related articles (n=2), and add

some article adequate from the review references of

the previous articles (n=1).

Figure 1: MeSH terms and search query used in the

methodology.

The inclusion criteria required that studies had a

clinical evaluation of the automatic (semi-automatic)

organ delineation and had used CT-based

radiotherapy planning for prostate cancer patients.

The exclusion criteria eliminate studies that had a

clinical evaluation of the automatic organ

delineation based only in Magnetic Resonance (MR)

image, or ultrasound-guided brachytherapy, or

multimodality images (combination of CT, positron

emission tomography or MR). Articles that

presented automatic localization of the prostate for

image-guided radiotherapy on cone-beam CT scans

or on megavoltage computed tomography images

were not considered in this review.

3 FINDINGS

A total of seven studies were selected from the

scientific literature. The studies were reported

between 1994 and 2009. Three of them were

published by British Journal of Radiology. Some of

these studies were a result of a work between

radiotherapy departments from different institutions

or countries.

All of the studies are clinical validations that

compare automatic segmentation with manual

tracing of pelvic organs. Most of them, had

presented quantitative and qualitative evaluations,

and nearly all studies had as object of study the

quality assessment of the automatic segmentation in

AUTOMATIC ORGAN DELINEATION OF COMPUTED TOMOGRAPHY IMAGES FOR RADIOTHERAPY

PLANNING IN PROSTATE CANCER - An Overview

483

terms of expert inter/intra variability. In the selected

articles, it was identified the technique or algorithm

or the software used in the automatic segmentation.

To this evaluation, the observers that participated in

the studies were clinical experts like Radiation

Oncologist (n=5), Dosimeter (n=1), Physicist (n=4),

Radiographer (n=1) and Oncologist (n=1). In the

studies, the segmented pelvic organs were: the

prostate (n=5), the seminal vesicles (n=3), the

bladder (n=6), the rectum (n=7) and the femoral

heads (n=4). Five studies used the same patients for

both methods of segmentation (automatic and

manual).

The table 1 shows a systematization of the

outcomes resulting from the selected studies with

different evaluators and respective metrics.

4 DISCUSSION

This review attempts to associate several studies in

order to compare the outcomes, despite the

differences in their methodologies, concerning

segmentation techniques and statistical methods. In

the segmentation approach, the studies evaluated the

following techniques: region growing, deformable,

morphological, automatic segmentation software and

auto-segmentation algorithm. Regarding the

samples, the studies differ in: CT data (number of

sets and slices, slice thickness, image resolution)

conditions of CT acquisition (administration of

contrast, bowel gas, bladder filling), organs

segmented, user’s interaction, etc. The outcomes

were measured trough different evaluators:

efficiency (given by the respective segmentation

times with mean volume/area and standard

deviations of the organs segmented or distance

Table 1: Systematization of most outcomes, presenting metrics used to evaluate the manual tracing and the automatic

segmentation (volume and time mean, overlapping ratio, Hausdorff distance) and metrics to compare directly the both

segmentation methods (percentage of relative error, percentage of agreement, band area difference in terms of average

maximum, median difference of volume and time, deviation mean, and overall rating in terms of percentage or levels).

V- Volume mean/ T- Time mean/ O- Overlapping ratio/ HD- Hausdorff Distance/ RE- Relative Error/ AG- Agreement/ BD- Band area

Difference (average maximum)/ VD- median Volume Difference/ TD- median Time Difference/D-Deviation mean / E- excellent/ G- Good/

A- Acceptable/ NA- Not Acceptable

Study

Manual tracing Automatic Segmentation Comparative Evaluators

cm3

cm2

cm3

Overall Rating ( %)

E G A NA

Huyskens

Prostate 60 53 33 75 9 45 30 15

Bladder 250 242 9 91 36 42 12 9

Rectum

Femoral

122

Hass

Prostate 0.81 

Bladder 1.02 

Rectum 2.64 

Femoral 2.58



McBain

Prostate 61 480 54 120 3.1 360

Vesicles 23 120 17 120 6 0

Bladder 174 720 171 300 7.7 420

Rectum 69 480 54 120 5.6 360

Pekar

Bladder 1.5

Rectum 1.6

Right femur 0.9

Bueno

Vesicles 9.1 1.1 0.9 1.5 72

Bladder 0.9 1.5 0.9 1.2 92

Rectum 1.0 0.8 0.7 1.7 90

Mazonakis

738 504

Prostate 46 47

Bladder 305 302

Rectum 118 110

Neal

Prostate 194 219

Vesicles 108 123

Bladder 242 198

Rectum

Femur

197

1136

162

204

HEALTHINF 2011 - International Conference on Health Informatics

484

measurement between the two methods),

reproducibility (intra/inter variability given by the

coefficient of variation values or relative percentage

of agreement among the experts), accuracy (mean

segmentation error), sensitivity (True Positive Rate),

and specificity (True Negative Rate).

The most part of the studies, conclude that

automatic segmentation closely reproduce manual

contours with no significant volume difference and

with a significant time difference. The automatic

approach is faster with equivalent accuracy to

manual method. In concern with the observer’s

variability, automatic segmentation reaches

significant gains in reproducibility. This may be

attributed to the reduced user interaction required for

efficient segmentation of the organs.

In general, the studies used small samples,

except the Hass et al, and one of the studies, Neal et

al, contrary of the general outcomes, concluded that

the time taken to segment automatically the prostate

was superior compared with the manual tracing.

This fact could be explained due to the algorithm

limitations in the soft tissue segmentation.

In the automatic segmentation, the limitations

found in the studies were: underestimation of

prostate, distinguish the base of bladder from

prostate, segment the real boundary of rectum and

separation of the rectum from seminal vesicles.

All studies have emphasized the potential of the

automatic approach to improve radiotherapy

planning conditions. In contrast to manual slice

delineation, organ segmentation can be done within

a few minutes with no significant mean

segmentation error. However, some problematic

contours of soft tissues have to be corrected

interactively.

In general, many published approaches in image

segmentation are validated on a small set of test

images and few methods in the domain of automated

organ segmentation for RTP have been

quantitatively validated so far.

As future directions, it is recommended the

improvement of the segmentation algorithms in the

delineation of problematic soft tissues and future

validation studies with large scale trials.

Furthermore, a thorough systematic review

aiming at a study of meta-analysis is required to

critically access the differences between automatic

and manual segmentation, especially for prostate

cancer.

REFERENCES

Boehmer, D., Maingon, P., et al. 2006. Guidelines for

primary radiotherapy of patients with prostate cancer.

Radiotherapy and Oncology, 79, 259-269.

Bueno, G., Fisher, M., et al. 2001. Automatic

segmentation of clinical structures for RTP:

Evaluation of a morphological approach. Proceedings

of Medical Image Understanding and Analysis (MIUA

'01), 73-76.

Freedman, D., Radke, R., et al. 2005. Model-based

segmentation of medical imagery by matching

distributions. IEEE Transactions on Medical Imaging,

24, 281-292.

Haas, B., Coradi, T., et al. 2008. Automatic segmentation

of thoracic and pelvic CT images for radiotherapy

planning using implicit anatomic knowledge and

organ-specific segmentation strategies. Journal

Physics in Medicine and Biology, 53, 1751-1771.

Huyskens, D. P., Maingon, P., et al. 2009. A qualitative

and a quantitative analysis of an auto-segmentation

module for prostate cancer. Radiotherapy and

Oncology, 90, 337-345.

Li, T., Xing, L., et al. 2006. Four-dimensional cone-beam

computed tomography using an on-board imager.

Medical Physics, 33, 3825-3833.

Mazonakis, M., Damilakis, J., et al. 2001. Image

segmentation in treatment planning for prostate cancer

using the region growing technique. British Journal of

Radiology, 74, 243-248.

Mcbain, C. A., Moore, C. J., et al. 2008. Early clinical

evaluation of a novel three-dimensional structure

delineation software tool (SCULPTER) for

radiotherapy treatment planning. British Journal of

Radiology, 81, 643-652.

Neal, A. J., Sivewright, G., et al. 1994. Technical note:

Evaluation of a region growing algorithml for

segmenting pelvic computed tomography images

during radiotherapy planning. BJR, 67, 392-395.

Pekar, V., Mcnutt, T., et al. 2004. Automated model-based

organ delineation for radiotherapy planning in

prostatic region. International Journal of Radiation

Oncology Biology Physics, 60, 973-980.

Quicken, M. 2000. Generation and Application of

Statistical Shape Models for the Segmentation of

Abdominal Organs in Radiotherapy Planning. Doctor

of Technical Sciences, Swiss Federal Institute of

Technology.

AUTOMATIC ORGAN DELINEATION OF COMPUTED TOMOGRAPHY IMAGES FOR RADIOTHERAPY

PLANNING IN PROSTATE CANCER - An Overview

485