See discussions, stats, and author profiles for this publication at: https://www.researchgate.

net/publication/324241080

QALMA: A computational toolkit for the analysis of quality protocols for

medical linear accelerators in radiation therapy

Article · January 2018

DOI: 10.1016/j.softx.2018.03.003

CITATION READS

1 184

3 authors:

Mushfiqur Rahman Yu Lei

Florida Atlantic University University of Nebraska Medical Center
6 PUBLICATIONS   1 CITATION    64 PUBLICATIONS   252 CITATIONS   

SEE PROFILE SEE PROFILE

Georgios Kalantzis
21st Century Oncology Fort Lauderdale
34 PUBLICATIONS   153 CITATIONS   

SEE PROFILE

Some of the authors of this publication are also working on these related projects:

Small Field dosimetry in Cyberknife View project

Automatic Volumetric Modulated Arc Therapy (VMAT) Planning on Hippocampal Avoidance Whole-Brain Radiation Therapy View project

All content following this page was uploaded by Georgios Kalantzis on 24 April 2018.

The user has requested enhancement of the downloaded file.

 SoftwareX 7 (2018) 101–106

Contents lists available at ScienceDirect

SoftwareX
journal homepage: www.elsevier.com/locate/softx

Original software publication

QALMA: A computational toolkit for the analysis of quality protocols

for medical linear accelerators in radiation therapy
Md Mushfiqur Rahman a , Yu Lei b , Georgios Kalantzis a, *
a
Medical Physics Program, Department of Physics, Florida Atlantic University, Boca Raton, FL, 33431, USA
b
Department of Radiation Oncology, University of Nebraska Medical Center, Omaha, NE, 68198, USA

article info a b s t r a c t
Article history: Quality Assurance (QA) for medical linear accelerator (linac) is one of the primary concerns in external
Received 17 November 2017 beam radiation Therapy. Continued advancements in clinical accelerators and computer control tech-
Received in revised form 12 February 2018 nology make the QA procedures more complex and time consuming which often, adequate software
Accepted 8 March 2018
accompanied with specific phantoms is required. To ameliorate that matter, we introduce QALMA (Quality
Assurance for Linac with MATLAB), a MALAB toolkit which aims to simplify the quantitative analysis of
Keywords:
QA for linac which includes Star-Shot analysis, Picket Fence test, Winston–Lutz test, Multileaf Collimator
Star shot
Picket fence (MLC) log file analysis and verification of light & radiation field coincidence test.
Winston–Lutz © 2018 The Authors. Published by Elsevier B.V. This is an open access article under the CC BY license
Light radiation coincidence (http://creativecommons.org/licenses/by/4.0/).
Log file analysis
Quality assurance

Code metadata

Current Code version V1

Permanent link to code/repository used of this code version https://github.com/ElsevierSoftwareX/SOFTX-D-17-00089
Legal Code License GPLv3
Code Versioning system used git
Software Code Language used MATLAB
Compilation requirements, Operating environments & dependencies Image Processing Toolbox, Signal Processing Toolbox, Symbolic Math
Toolbox
If available Link to developer documentation/manual https://github.com/mrmushfiq/qalma/wiki
Support email for questions 2mushfiq@gmail.com

(IMRT), stereotactic radiosurgery (SRS) and stereotactic body ra-

diotherapy (SBRT). TG 142 did not only replace the TG 40 [2], but
1. Motivation and significance
also filled the gap between TG 40 and TG 100 [3].
Linac QA tests quite often involve several radiation field pat-
Quality Assurance (QA) procedures play a vital role in all terns, and corresponding beam measurement techniques which
aspects of external radiation therapy with a linear accelerator are specifically designed for diverse verifications, such as MLC
(Linac). In particular, the technical characteristics of QA procedures checks [4], star shot analysis [5–7], Winston–Lutz test [8–10] and
are usually well established through several protocols, codes of light and radiation field coincidence [11]. MLC positioning checks
practice and recommendations from the American Association of can be performed with the Picket Fence test [12,13] and Dynalog
Physicists in Medicine (AAPM) [1]. The AAPM Task Group (TG) file analysis for Varian Linacs [4,14], the mechanical rotational
142 [1] was published in 2009, to recommend QA techniques for alignment of the gantry, collimators, MLC and treatment couch
newer technologies such as intensity modulated radiation therapy position is checked by the Star Shot analysis, and the Winston–
Lutz (W–L) test is performed to verify that the radiation isocen-
ter coincides with the mechanical isocenter in a Linac system.
* Corresponding author. The recommendations for QA of a megavoltage radiation therapy
E-mail address: gkalan@gmail.com (G. Kalantzis). equipment also require the coincidence of the light and radiation.

https://doi.org/10.1016/j.softx.2018.03.003
2352-7110/© 2018 The Authors. Published by Elsevier B.V. This is an open access article under the CC BY license (http://creativecommons.org/licenses/by/4.0/).
102 M.M. Rahman et al. / SoftwareX 7 (2018) 101–106

To perform the QA tests, there are several commercial software Finally, the center and the radius of the minimum circle are
available from different vendors such as DoseLabTM or PIPSproTM calculated by Eqs. (8) and (9),
and several authors have published QA procedures using the soft- (
a b
)
ware and phantom provided by the Linac manufacturers [15,16]. (x0 , y0 ) = − , − (8)
2 2
However, the cost-effectiveness and customizability have always {√ }
been an issue since commercial software do not always provide r = max (xi − x0 )2 + (yi − y0 )2 (9)
enough flexibility to the medical physicist to adjust and/or modify
any QA procedure according to the needs of each facility.
In this work, we present QALMA (Quality Assurance for Linac 2.2. Picket fence
with Matlab), a user-friendly Matlab based computational toolkit
to perform Quality Assurance tests on Medical Linear Accelerators. Once the Picket fence DICOM image [13] is loaded the user
QALMA can perform quantitative analysis of various QA tests such is prompted to select the pickets required to be analyzed. The
as: Picket Fence, DynaLog analysis, Star Shot analysis, Winston– coordinates are then stored as initial reference points, which can be
Lutz test, Light and Radiation field coincidence and also generates used to find the location of the Bank A and Bank B of the collimator
reports. (Fig. 2(a)).
Firstly, a Wiener adaptive noise removal filter [17] is applied to
2. Software framework obtain smooth profiles (Fig. 2 (b) and (c)) as shown in the flowchart
(Fig. 1). Positions of the banks 1–5 (Fig. 2(a)) are then used to
extract the intensity profiles perpendicular to the movement of the
QALMA is designed as a MATLAB graphical user interface (GUI).
leaves (Fig. 2(c)). The centers of each leaf pairs can be found at the
Methods for the three most important modules are in Fig. 1. In this
minima of the profile. The individual leaf positions can be obtained
section, we briefly discuss the methods used for each module.
by averaging the two consecutive minima [13].
The intensity profiles across the pickets are extracted at the
2.1. Star shot position obtained in the previous step for each collimator leaf. The
image used for this report has five minima (Fig. 2(b)) due to the
Once the start-shot [5–7] image is loaded, it is converted to five pickets (Fig. 2(a)) in each profile. Subsequently each profile
grayscale and an adaptive filter is employed for image noise re- was inverted and the baseline was subtracted, followed by a cubic
duction. Wiener filter was chosen as default due to its ability to spline interpolation. Ten intervals were considered between each
minimize the mean square error between the estimated random pixel to interpolate the intensity profile (Fig. 2(b)). Full Width Half
process and the desired process [17]. Then the star-shot flowchart Maximum (FWHM) was determined by extracting the positions of
(Fig. 1) is followed to determine the co-ordinates of the brunches two points at the 50% of the local pick of the profile. These two
and to find the intersection between the beams. Considering an points determine the edges of the leaf pairs (Fig. 2(a)).
image having only two beams or four brunches, assume the de-
termined co-ordinates for the spokes are (x1, y1), (x2, y2), (x3, y3) 2.3. Winston–Lutz
and (x4, y4). Then the corresponding slopes (a1, a2) and intercepts
The W–L phantom consists of a small metallic ball of size 6 mm
(b1, b2) for the two beam center lines can be found using the
attached to the end of a rod and aligned to the isocenter, which is
following equations:
placed at the target and is fixed to the treatment table. Once the
DICOM image is loaded The QALMA Winston–Lutz module utilizes
y3 − y1 morphological filtering techniques since the main challenge is to
a1 = ; b1 = y1 − a1 x1 (1)
x3 − x1 separate the radiation field and the metal ball as two different
y4 − y2 objects, so that their centers can be determined separately. To
a2 = ; b2 = y2 − a2 x2 (2)
x4 − x2 achieve that, once the W–L image (Fig. 3) is loaded, it is converted
to a grayscale one and subsequently is filtered through an adaptive
The intersections can be simply found by solving for x and y,
noise removal filter. In the next step, it is passed through two
a1 x + b1 = a2 x + b2 (3) morphological filters: the top-hat and the bottom-hat [8,18].
The top-hat filter was implemented using MATLAB’s ‘imtophat’
y = a1 x + b1 (4)
function to obtain the center of the ball (Fig. 3: Filter I). The bottom-
where (x, y) is our intersection point of the two star lines. hat filter was implemented using MATLAB’s ‘imbothat’ function
The final step is to use all the intersections to fit a circle by mini- to obtain the center of the radiation field (Fig. 3: Filter II). Sub-
mizing certain objective function. Let us consider the equation of a sequently, the two separate images from the two different filters
circle (Eq. (5)), where a, b, c are undetermined coefficients. Then are converted to binary images and the center of the ball (circular
(Eq. (6)) defines the summarization of the squares of distances white object) and the radiation field (rectangular white object) is
between the circumference and all intersections (xi , yi ) [6]. obtained by deploying the MATLAB’s ‘regionprops’ function.

x2 + y2 + ax + by + c = 0 (5) 2.4. Log file analysis

∑
Q (a, b, c ) = 2 2
(x + y + ax + by + c) 2
(6)
The Varian Trilogy and Varian Clinac iX models both generate
n
two separate files for A and B MLC banks and assigns unique file
We consider Eq. (6) as the function of optimization and let it is names to the generated Dynalog files as a record of the Dynamic
three partial differentials equal to zero Multi-Leaf Collimator (DMLC) delivery details recorded at every
50 ms for a dynamic treatment [19,20]. By analyzing these log
∂Q ∂Q ∂Q
= = =0 (7) files, the positional error of each individual leaf can be determined.
∂a ∂b ∂c QALMA provides a simple GUI to verify the result of the Picket
The solution of (Eq. (7)) is obtained by simply solving three Fence test with a log file analysis for Varian Trilogy or Varian Clinac
linear equations simultaneously for a, b and c. iX.
 M.M. Rahman et al. / SoftwareX 7 (2018) 101–106 103

Fig. 1. Flowcharts for picket fence, star-shot and Winston–Lutz.

Fig. 2. (a) Picket Fence window, (b) Intensity profile of across the pickets at central leaf position, (c) Intensity profile perpendicular to the movement of the leaves at position
1 (see the labels 1–5 on top of the picket fence image).

The sample Dynalog files provided with QALMA are generated were used to calculate the planned gap. However, to generate result
by Clinac 21EX system, which is equipped with 120 leaves MLC. comparable to the Picket Fence test we only find the gaps when the
Once the two log files for Bank A and Bank B are loaded, four leaves stop moving which can be achieved by finding the peaks on
matrices are created which store the planned and actual position the leaf position profile. Finally, the errors are found by calculating
of each leaf in the two banks. The actual positions of bank A and B the difference between planned and actual gaps. Actual gaps can
are compared to calculate the actual gap and the planned positions be used to verify the Picket Fence results.
104 M.M. Rahman et al. / SoftwareX 7 (2018) 101–106

Fig. 3. Winston–Lutz module.

2.5. Light-radiation coincidence which are Resolution of the film or the imager and the Magni-
fication Factor. The resolution can be found in vendor’s manual.
To perform Light Radiation field coincidence test with QALMA The magnification factor can be calculated using the following
cost effectively we have developed an in-house phantom which equation:
was developed using a graph paper glued to the top of a cork board
Source to Imager Distance (SID)
and board pins pinned at the four corners of a 10 × 10 cm2 square M .F . = (11)
to indicated the light field corners. Another pin was used to mark Source Axis Distance (SAD)
the center of the light field. For Star-Shot QA the Magnification factor can left with default
The first step is adjustment of the contrast and the sensitivity value ‘1’ as the test is performed at the isocenter where SAD is
after it is gone through the adaptive noise removal filtering pro- equal to SID. However, for Picket Fence, Winston–Lutz and Log File
cess [17]. Once the markers, for our case the pins are detected, their Analysis SAD and SID are usually set at 100 cm and 150 cm which
coordinates are connected to form a square which indicates the sets the Magnification factor as 1.5.
radiation field. However, the user can also switch to the manual Finally, the user can add comment and click Analyze and gen-
mode and click on the markers to get their coordinates. Next, we
erate test result. The user can add more images one by one and
extract four profiles from the four edges of the square radiation
follow the same procedure. The results are shown in a table. At the
field using the coordinates determined earlier. The full width half
end of the tests the user can generate a report by clicking the Report
maximum (FWHM) of these profiles determines each of the edges
button. We recommend the users to check our manual for detailed
of the radiation field. Finally, we quantify the similarity between
step by step instruction of each module.
light field and radiation field using Dice coefficient. The Dice coeffi-
cient between two sets A and B ranges from 0 to 1 and is expressed
as: 4. Impact
|A ∩ B|
Dice (A, B) = 2 (10) QALMA provides a cost-effective solution to medical physicists
|A| + |B|
with the precision of the commercial software and customizability
included in one single package. Our Picket Fence result was verified
3. Illustrative example
with PipsProTM , Star-shot was verified with FilmQATM pro and
Winston–Lutz was verified with DoseLabTM . Leaf gap observed in
QALMA requires MATLAB installed on a computer with Sig-
the picket fence test was confirmed again using DynaLog analysis.
nal processing toolbox, Image processing toolbox and Symbolic
Finally, for the Light-radiation we employed the Dice Coefficient
math toolbox. The software can be downloaded for free from
our github repository http://github.com/mrmushfiq/qalma. Then as a quantitative metric even if it could be verified by simply visual
open MATLAB and set path with subfolders by navigating to the inspection.
downloaded ‘qalma’ folder. The user can run QALMA by typing QALMA is not developed for any specific linac. Users can follow
qalma in the command line window. Fig. 4 is a visual flow chart the manual to customize it for their facility. This makes QALMA
depicting a typical workflow with QALMA. Picket Fence, Winston– very useful, especially for small clinics on budget. It also features a
Lutz and Light Radiation modules requires the image to be in the very friendly Graphical User Interface which makes it user-friendly
DICOM format and Star-shot module requires it to be JPEG, PNG or without knowledge of programming. In addition, it could be valu-
TIF. able to an academic institution since it employs a rich variety of
Once the image is loaded, check the boxes which correspond numerical techniques such as non-linear data interpolation, image
to the image. For example, on the Star-Shot menu the user smoothing and filtering, adaptive noise reduction, as well as mor-
must choose whether the image belongs to the Gantry, Couch or phological filtering, contrast adjustment which makes beneficial
Collimator. Next, the user must plug in the configuration values for students and researchers.
 M.M. Rahman et al. / SoftwareX 7 (2018) 101–106 105

Fig. 4. QALMA work flow.

5. Conclusion [5] Winkler P, Bergmann H, Stuecklschweiger G, Babic S, Mcniven A, Battista J,

Skworcow P, Mills JA, Haas OCL, Depuydt T, Penne R, Verellen D, Hrbacek J.
Computer-aided analysis of star shot films for high-accuracy radiation therapy
QALMA provides a convenient and cost-effective way to treatment units. Phys Med Biol 2012;57(10):2997.
perform five Quality Assurance tests: Star Shot, Picket Fence, [6] Ke L, Xin B, Lixin H, Xin Z. Study on star shot analysis for physics quality
Winston–Lutz, LogFile analysis and Light & Radiation field Co- assurance in radiotherapy. Appl Mech Mater 2014;542:1313–8.
[7] Castro I, Martı JA. A procedure to determine the radiation isocenter size in a
incidence. It is not only an advanced software with a friendly
linear accelerator. Med Phys 2004;31(6).
graphical user interface but also a customizable platform. Users [8] Ravindran PB. A study of Winston –Lutz test on two different electronic portal
with basic knowledge of MATLAB will be able to customize QALMA imaging devices and with low energy imaging. Australas Phys Eng Sci Med
as required for their facilities. Future plans include the inclusion of 2016;39(3):677–85.
[9] Gao J, Liu X. Off-isocenter Winston-Lutz test for stereotactic radiosurgery/
the Flatness/Symmetry test, VMAT, CBCT and kV/MV tests [21–23].
stereotactic body radiotherapy. Int J Med Phys Clin Eng Radiat Oncol 2016:
We also plan to include a database to track machine QA results. 5(2):154–61.
Work along these lines is already in progress. [10] Winston KR, Lutz W. Linear accelerator as a neurosurgical tool for stereotactic
radiosurgery. Neurosurgery 1988;22(3):454.
[11] Njeh CF, Caroprese B, Desai P. A simple quality assurance test tool for the
Appendix A. Supplementary data visual verification of light and radiation field congruent using electronic portal
images device and computed radiography a simple quality assurance test tool
for the visual verification of light and radiation. Radiat Oncol 2012;7(1):49.
Supplementary material related to this article can be found [12] Christophides D, Davies A, Fleckney M. Automatic detection of MLC relative
online at https://doi.org/10.1016/j.softx.2018.03.003. position errors for VMAT using the EPID-based picket fence test. Phys Med Biol
2016;61(23):8340.
[13] Agnew A. Monitoring daily MLC positional errors using trajectory log files and
References EPID measurements for IMRT and VMAT. Phys Med Biol 2014;59(9).
[14] Lee L, Mao W, Xing L. The use of epid-measured leaf sequence files for imrt
[1] Klein EE, Hanley J, Bayouth J, Yin F-F, Simon W, Dresser S, Serago C, Aguirre F, dose reconstruction in adaptive radiation therapy. Med Phys 2008;35(11).
Ma L, Arjomandy B, Liu C, Sandin C, Holmes T. Task Group 142 report: Quality [15] Clivio A, Nicolini G, Vanetti E, Fogliata A, Cozzi L. Commissioning and early ex-
assurance of medical acceleratorsa). Med Phys 2009;36(9):4197–212. perience with a new-generation low-energy linear accelerator with advanced
[2] Kutcher GJ, Coia L, Gillin M, Hanson WF, Leibel S, Morton RJ, Palta JR, Purdy JA, delivery and imaging functionalities. Radiat Oncol 2011;6(1):129.
Reinstein LE, Svensson GK, Weller M, Wingfield L. Reprinted, Comprehensive [16] Clivio A, Vanetti E, Rose S, Nicolini G, Belosi MF, Cozzi L, Baltes C, Fogliata A.
QA for radiation oncology published for the by the American Institute of Evaluation of the machine performance check application for truebeam linac.
Physics comprehensive QA For. Med Phys 1994;21(4). Radiat Oncol 2015;10(1):97.
[3] Saiful Huq M, Fraass BA, Dunscombe PB, Gibbons JP, Ibbott GS, Mundt A, Mutic [17] Ogawa H, filter Projection, filter Wiener. Projection filter Wiener filter and
S, Palta JR, Rath F, Thomadsen BR, Williamson JF, Yorke ED. The report of Task Karhunen-Loève subspaces in digital image restoration. J Math Anal Appl
Group 100 of the AAPM: Application of risk analysis methods to radiation 1986;114(1):37–51.
therapy quality management. Med Phys 2015;43(7):4209–62. [18] Efford N. Digital Image Processing: A Practical Introduction Using Java, 2000.
[4] Eckhause T, Ritter T, Demarco J, Farrey K, Popple R, Perez M, Booth JT. [19] Varian Medical Systems, DynaLog File Viewer reference guide, DynaLog File
Automating linear accelerator quality assurance. Med Phys 2015;42(10). Viewer Ref. Guid. vol. Palo Alto, 2007.
106 M.M. Rahman et al. / SoftwareX 7 (2018) 101–106

[20] Varian Medical Systems, Millennium mlc system and maintenance guide, [22] Mao W, Lee L, Xing L. Development of a QA phantom and automated analysis
Varian Medical Systems, Tech. Rep. 2013. Millenn. mlc Syst. Maint. Guid. vol. tool for geometric quality assurance of on-board MV and kV x-ray imaging
Tech. Rep, 2013. systems. Med Phys 2008;35(4):1497–506.
[21] Du W, Yang J, Chang E, Luo D, McAleer M, Shiu A, Martel M. SU-GG-J-74: [23] Liu B, Adamson J, Rodrigues A, Zhou F, Yin F-F, Wu Q. A novel technique for
Congruence of cone-beam CT image center with the radiation isocenter of a VMAT QA with EPID in cine mode on a Varian TrueBeam linac. Phys Med Biol
linear accelerator. Med Phys 2010;37(6):3162. 2013;58:6683–700.

View publication stats