423
Asian Journal of Healthy and Science
p-ISSN: 2980-4302
e-ISSN: 2980-4310
Vol. 2 No. 10 October 2023
AUTOMATIC CONTOURING ANALYSIS OF LUNG CANCER FOR
RADIOTHERAPY RADIATION PLANNING FROM CT-SIMULATOR
IMAGE
1
Putri Pradita Nuramalia, ²Suryono, ³Edy Susanto
1,3
Poltekkes Kementerian Kesehatan Semarang, Indonesia
2
Universitas Diponegoro Semarang, Indonesia
Email : putri.pradita15@gmail.com
Abstract
Accurate determination of tumor and radiotherapy target volume is crucial to prevent
local and regional failure in lung cancer. Radiotherapy planning considers critical
normal tissue structures or Organs at Risk (OAR). While manual contouring is the
gold standard in radiotherapy planning, it is still susceptible to intra- and inter-
observer variation. A digital image processing system capable of automatically and
semi-automatically determining tumor targets and OARs through computer
programming can assist in the contouring process. The study aims to create a digital
contouring computer program to aid in planning radiotherapy irradiation for lung
cancer cases, guided by CT-Simulator image guidance. The research involved
contouring CT-Simulator images of lung cancer cases using automatic segmentation,
semi-automatic segmentation, and a combination of automatic, semi-automatic, and
manual segmentation. Contour accuracy was assessed by a radiation oncologist.
Automatic segmentation showed high accuracy for lungs (>95%), moderate for
tumor targets (72.22%) and the heart (73.33%), but low accuracy for the esophagus
(34.44%) and spinal cord (36.67%). Semi-automatic segmentation achieved high
accuracy for the esophagus (96.67%) and spinal cord (95.56%), and moderate
accuracy for tumor targets (73.33%), heart (66.67%), right lung (66.25%), and left
lung (72.94%). The combination of automatic, semi-automatic, and manual
segmentation resulted in high accuracy for tumor targeting and OARs (>95%).
Automatic or semi-automatic segmentation using location and gray level
classification methods for lung cancer cases did not produce optimal results.
However, a contouring program combining automatic, semi-automatic, and manual
segmentation proved more effective, balancing human interaction and automation in
the lung cancer contouring process. This digital contouring program offers valuable
support for radiation oncologists, potentially leading to improved treatment
outcomes for lung cancer patients.
Keywords: automatic contouring; lung cancer; radiotherapy
424
INTRODUCTION
Radiotherapy is considered one of the most effective cancer treatment methods.
It involves the utilization of ionizing radiation, which is high-energy radiation
capable of displacing electrons from atoms and molecules. The primary goal of
radiotherapy is to destroy cancer cells and inhibit their growth (Aselmaa et al., 2017;
Marcus, 2020; Sebaaly et al., 2019). CT-Scan is essential in radiation planning or
radiotherapy treatment planning, as it offers high geometric accuracy in locating
tumors and surrounding tissues and organs at risk, enabling precise identification.
Additionally, CT-Scan provides electron density information maps for various
tissues, which are used for dose calculation in the treatment planning system (TPS)
(Davis et al., 2017). Local irradiation, such as radiotherapy, relies heavily on the
accuracy of the imaging procedure to precisely identify the target tissue. This
accuracy serves as the foundation for successful irradiation while minimizing
potential complications (Sterzing et al., 2011).
Until recently, contour segmentation was predominantly a manual task
performed by radiation oncologists, and radiation technology was considered the
gold standard (Kerenhapukh, 2022; Sudarsa, 2019). However, contouring can be
time-consuming, and published data demonstrate a high level of inter-observer
variability (Cacicedo et al., 2019; Silva et al., 2018). Automated or semi-automated
contouring tools are increasingly being employed in tumor delineation for
radiotherapy. While fully automatic contouring tools have not yet achieved optimal
efficiency, semi-automatic contouring tools prove to be more effective as they strike
a balance between human interaction and automation (Aselmaa et al., 2017).
RESEARCH METHODS
The patient data used in this study were CT-Simulator lung images from lung
cancer patients in DICOM format. Target volume and risk organ volume (OAR)
must be precisely contoured to produce an accurate treatment plan (Nelms et al.,
2012). The OARs included are the heart, esophagus, spinal cord, left lung, and right
lung (Men et al., 2020).
This research is an experimental study which was tested qualitatively (Lee et
al., 2013). The research sample (lung CT-Simulator images from lung cancer
patients) applied three types of segmentation (automatic segmentation, semi-
automatic segmentation and a combination of automatic, semi-automatic, and
manual segmentation). The three types of segmentation were compared for their
accuracy values rated by radiation oncologists. Radiation oncologists rated of
accuracy on a predetermined scale. After assessing the accuracy of contouring, the
data is then processed with SPSS (Santoso, 2017).
The contouring accuracy data for the three types of segmentation are averaged.
In addition, the data was calculated using a non-parametric test. Kruskal-Wallis test
followed by Tamhane's post hoc analysis.
RESULTS AND DISCUSSION
In this study, the experiment was conducted using MATLAB software to
process the image. Contouring was performed using three types of segmentation:
automatic segmentation, semi-automatic segmentation, and a combination of
automatic, semi-automatic, and manual segmentation (Hidayati et al., 2017). The
425
automatic segmentation utilized the K-Nearest Neighbor method and thresholding,
while the semi-automatic segmentation employed the active contour method and the
manual method of Region of Interest (ROI) for manual segmentation. The study
results are depicted in Figure 1.
(a) (b)
(c) (d)
Figure 1. Original image (a), automatic segmentation (b), semi-automatic
segmentation (c), combination of automatic, semi-automatic and manual
segmentation (d).
The contouring results obtained from the segmented image are subsequently
assessed for accuracy by radiation oncologists concerning the tumor target and OAR,
as shown in Table 1. The results of the Kruskal-Wallis test are presented in Table 2,
and the outcomes of Tamhane's post hoc test can be observed in Table 3.
Table1. Average Contouring Accuracy Score
Otomatis
Semi-Otomatis
Combination
Tumor Target
72,22
73,33
97,78
Heart
73,33
66,67
93,33
Esophagus
34,44
96,67
97,78
Spinal Cord
36,67
95,56
97,78
Right Lung
97,5
66,25
98,75
Left Lung
98,82
72,94
98,82
Table 2. Kruskal-Wallis Test Results
Variable
p-value
Tumor Target
0,000
Heart
0,225
Esophagus
0,000
Spinal Cord
0,000
426
Variable
p-value
Right Lung
0,000
Left Lung
0,000
Table 3. Mann-Whitney Test Results
Variable
p-value
Otomatis and Semi-Otomatis
Tumor Target
0,673
Heart
0,700
Esophagus
0,000
Spinal Cord
0,000
Right Lung
0,000
Left Lung
0,000
Otomatis and Combination
Tumor Target
0,001
Heart
0,400
Esophagus
0,000
Spinal Cord
0,000
Right Lung
0,780
Left Lung
1,000
Semi-Otomatis and Combination
Tumor Target
0,000
Heart
0,100
Esophagus
0,791
Spinal Cord
0,584
Right Lung
0,000
Left Lung
0,000
The average accuracy grade of the automatic segmentation contouring results
indicates a low score for the esophagus and spinal cord. However, both the tumor
and heart targets have not received a moderate score. High scores are only observed
in both right and left lungs.
In the semi-automatic segmentation method, high scores are obtained for the
esophagus and spinal cord, while the target values for tumor, heart, right lung, and
left lung still receive moderate scores. It should be noted that the esophagus and
spinal cord are small-sized organs, whereas the tumor, heart, and lungs are larger
organs.
The combination of automatic, semi-automatic, and manual segmentation
represents an improvement over relying solely on automatic and semi-automatic
segmentation methods. This editing process enhances the accuracy and yields high-
quality results in contoured areas. While fully automated tools may still lack
efficiency, utilizing tools that strike a balance between human interaction and
automation can prove to be more effective(Aselmaa et al., 2017).
In this study, the development of a program using fully automatic and semi-
automatic segmentation has shown to be less effective due to various factors, such as
gray level inconsistencies leading to incorrect contouring of certain organs. However,
427
the contouring program's effectiveness significantly improves when a combination of
automatic, semi-automatic, and manual segmentation methods is utilized, finding a
balance between human interaction and automation. This enables users to make
additional adjustments to contours based on other considerations beyond the CT-
Simulator image. The incorporation of this process results in accurate contouring
according to the ICRU guidelines.
From the results of the Kruskal-Wallis test were < 0.05 except for the heart, it
was concluded that there was at least a difference between the two groups. The next
test is post hoc using Mann-Whitney to find out between groups that have differences
(Sopiyudin, 2014). The results of the Mann-Whitney test, it can be ascertained that
there is no difference in the results according to contouring on variables with p-vaue
> 0.001. The variable with p-value < 0.001 indicates a difference in the two groups.
CONCLUSION
Fully automatic segmentation with location classification methods and fully
semi-automatic segmentation with methods that use a closed curve model, which
widens or narrows at the same degree of gray level in the CT-Simulator image of
lung cancer cases, have not achieved maximum results. The combination of
automatic, semi-automatic, and manual segmentation can increase the accuracy of
contouring due to the editing process to improve the contouring results that are less
accurate than automatic or semi-automatic segmentation in lung cancer cases.
Contouring programs are more effective as they balance human interaction and
automation. With this balance between human interaction and automation, the
approach minimizes time, reduces inter-observer variability, increases contour
accuracy, and enables adjustments to contour areas according to the doctor's
preferences based on various considerations.
BIBLIOGRAPHY
Aselmaa, A. Laprie, A. (2017). The influence of automation on tumor contouring.
Cognition, Technology and Work, 19(4), 795–808.
https://doi.org/10.1007/s10111-017-0436-0
Cacicedo, J. Dahele, M. (2019). Systematic review of educational interventions to
improve contouring in radiotherapy. Radiotherapy and Oncology, 144, 86–92.
https://doi.org/10.1016/j.radonc.2019.11.004
Davis, A. T. Nisbet, A. (2017). Can CT scan protocols used for radiotherapy
treatment planning be adjusted to optimize image quality and patient dose? A
systematic review. British Journal of Radiology, 90(1076).
https://doi.org/10.1259/bjr.20160406
Hidayati, I. N.Danoedoro, P. (2017). Pemetaan Lahan Terbangun Perkotaan
Menggunakan Pendekatan NDBI dan Segmentasi Semi-Automatik.
Kerenhapukh, Y. K. (2022). Perancangan Interior Rumah Sakit Dr. Sardjito Khusus
Kanker Payudara Dengan Pendekatan Art Therapy Di Yogyakarta.
Universitas Komputer Indonesia.
Lee, H. Kim, J. (2013). Why do people share their context information on Social
Network Services? A qualitative study and an experimental study on users’
behavior of balancing perceived benefit and risk. International Journal of
Human-Computer Studies, 71(9), 862–877.
428
Marcus, K. J. (2020). Radiation therapy. In Encyclopædia Britannica. Encyclopædia
Britannica.
Men, K. Xiao, Y. (2020). Automated Quality Assurance of OAR Contouring for
Lung Cancer Based on Segmentation With Deep Active Learning. Frontiers in
Oncology, 10(July), 1–7. https://doi.org/10.3389/fonc.2020.00986
Nelms, B. E. Wheeler, J. (2012). Variations in the contouring of organs at risk: Test
case from a patient with oropharyngeal cancer. International Journal of
Radiation Oncology Biology Physics, 82(1), 368–378.
https://doi.org/10.1016/j.ijrobp.2010.10.019
Santoso, S. (2017). Menguasai statistik dengan SPSS 24. Elex Media Komputindo.
Sebaaly, A. … Akl, T. (2019). Radiotherapy Treatment Planning System Simulation
Lung Cancer Application. IEEE International Multidisciplinary Conference on
Engineering Technology, IMCET 2018, 1–6.
https://doi.org/10.1109/IMCET.2018.8603054
Silva, B. A. Paixao, F. (2018). Femural Head Autosegmentation for 3D
Radiotherapy Planning: Preliminary Results. IEEE Transactions On Medical
Imaging.
Sopiyudin, M. (2014). Statistik untuk Kedokteran dan Kesehatan Deskriptif,
Bivariat, dan Multivariat, Dilengkapi Aplikasi Menggunakan SPSS (6th ed.).
Epidimologi Indonesia.
Sterzing, F. Debus, J. (2011). Image-Guided Radiotherapy. Deutsches Arzteblatt,
108(16), 274–280. https://doi.org/10.3238/arztebl.2011.0274
Sudarsa, I. W. (2019). BUKU AJAR BEDAH ONKOLOGI: Mata Kuliah BDH 202
Program Studi Ilmu Bedah Tingkat Bedah Dasar. Airlangga University Press.
Copyright holders:
Putri Pradita Nuramalia, Suryono, Edy Susanto (2023)
First publication right:
AJHS - Asian Journal of Healthy and Science
This article is licensed under a Creative Commons Attribution-ShareAlike 4.0
International