Lung Cancer Detection Using Convolutional Neural Network
PinnacleCommissioning
1. Pinnacle Treatment Planning Commissioning in the
Department of Radiation Oncology at the University of
Arizona
Lars Ewell
6/20/05
Outline
I) Introduction
II) Measurements
A) Equipment Used
B) Data Acquisition
C) Data Location/Inventory
III) Beam Modeling
A) Pinnacle 7.0
B) Pinnacle 7.4
IV) Hardware
V) Model Verification
A) Profiles and Depth Dose Curves
B) Anomalies
C) Comparison with Theraplan
VI) Conclusion
Summary: The process of commissioning the Treatment
Planning System, Pinnacle, in the Department of Radiation
Oncology at the University of Arizona is described. Different
aspects of the commissioning process are considered, from initial
data acquisition, to patient treatment.
2. Introduction
Of paramount importance in the administration of radiation in the field of
Radiation Oncology is the Treatment Planning System (TPS) used to predict radiation
distributions and plan treatments. Here in the Department of Radiation Oncology at the
University of Arizona Medical Center, a new TPS, Pinnacle (Philips ADAC1
) was
recently purchased, installed and commissioned.
It is the intent of this document to describe this commissioning process, as well as
provide reference information. A copy of this commissioning report can currently be
found on the ‘H-drive’ under H:commonPhysicsPinnacleDocuments.
Measurements
At present, Pinnacle is being used primarily for patient treatment on two linacs in
the department: The Siemens MD2.2 and the Elekta SLi. In the future, it will be
commissioned for use in brachytherapy, but that is beyond the scope of this report. In
view of this, the only data considered in this report will be that obtained from these two
machines.
Data Acquisition
In order to model the radiation therapy beams, a series of data were obtained
during the months of May, June and July, 2004. Joe Granados (jag5@email.arizona.edu)
and Russell Hamilton (rjh@email.arizona.edu) are the people that were primarily
responsible for obtaining these data. They consist mainly of depth dose scans, profiles at
multiple depths and field sizes, photon output factors and electron output factors. The
requirements for Pinnacle version 7.0, on which these scans were based, is detailed in the
‘Beam Data Collection Guide’, a copy of which is located in
H:commonPhysicsPinnacleDocuments .
Equipment Used
The data were acquired using a ‘Wellhofer’ scanning tank, along with ion
chambers and diodes. The software used was OmniPro 6.0 (see the OmniPro/Welhoffer
scanning computer currently located in the southwest corner of the BrainLab planning
room).
1
See http://www.medical.philips.com/us/
3. Data Location/Processing
The raw data are located on the shared drive under
H:commonPhysicsPinnacleElektaPhotonsScans for the Elekta photon scans and
H:commonPhysicsPinnacleSiemens IIPhotonsScans for the Siemens photon scans
and in similar places for electron data.
In order to conform with the data format outlined in the above mentioned data
collection guide, these raw data were ‘processed’ as follows: 1) Renormalized to 100% on
central axis (profile scans) or d_max (depth dose scans). 2) Smoothed using the Least
Squares Smoothing Function (8.0mm Mean Value Region) and Linear Interpolation
(0.2mm step width). 3) Centered using 50% of CAX value (profile scans). 4) Symmetrized
using 1.0mm resolution (profile scans).
After the data have been processed, they are saved in ASCII format, and then
finally converted to the exact format as specified in the reference guide. In order to
facilitate this final conversion, a program was written (Grandos) as a macro in an Excel
spreadsheet. This spreadsheet is titled ‘Macro for OmniPro.xls’ and a copy of it is located
in H:commonPhysicsPinnacleDocuments .
Two Excel spreadsheets that contain listings of all of the data that has been taken in
order to model the Elekta and Siemens are called ‘Pinnacle Elekta Data.xls’ and ‘Pinnacle
Siemens Data.xls’ and copies are located in
H:commonPhysicsPinnacleElektaDocuments and
H:commonPhysicsPinnacleSiemens IIDocuments respectively.
Beam Modeling
As indicated in the purchase contract for Pinnacle, Philips was responsible for
modeling the beam on both of the therapy machines. To accommodate this, a series of
emails were sent to Philips starting on 10/19/04 that contained the above mentioned data
as attachments.
Pinnacle 7.0
Rather than wait for the release of version 7.4, it was decided to model one
machine with the existing version of Pinnacle, 7.0. The Elekta SLi had its’ data
processed first, so that this machine was modeled in version 7.0. The first finished
Pinnacle 7.0 Elekta SLi photon beam model was received on 11/23/04.
Pinnacle 7.4
Pinnacle 7.4 has substantial differences with regards to version 7.0, and requires
additional modeling. The requirements needed to model this version of Pinnacle are
outlined in the ‘Pinnacle Physics Reference Guide, Release 7.4’, a copy of which is
4. located in H:commonPhysicsPinnacleDocuments. The Siemens machine was modeled
in 7.4 from the beginning and the first finished Pinnacle 7.4 Siemens photon beam model
was received on 3/17/04.
As indicated in the 7.4 reference guide, some additional scans are needed in order
to model the curved leaf edges on the MLC in the Elekta. These are ‘MLC-only’ scans,
where the secondary diaphragm is retracted and scans are obtained in which only the
MLC leaves define the field in one dimension. These additional scans were taken during
5/2005. They are listed in the Excel spreadsheet (Pinnacle Elekta Data.xls) and the data
were sent to Philips on 5/26/05.
Hardware
The various different pieces of hardware used in running the Pinnacle TPS
(workstations, printers, etc.) were received in the department the week of 10/19/04 and
installed by David May (dave.may@philips.com) the following week. In Table 1, the
names, locations and IP addresses of the five different pinnacle workstations are
described.
Table 1: Hardware Characteristics
As can be seen in the table, all of the workstations are on the radiology subnet,
198. There is connectivity between these stations and the 170 subnet via ‘Reflection’
software installed on a number of different PCs in the department (P3MD).
Model Verification
Upon receipt of the beam model from Philips, a number of predictions were tested
against data, to see how close the agreement was. As a first check, each of the profiles and
depth doses that were used to model the beam, were checked to see that the Pinnacle
Hardware Description Location Name IP Address
SunFire 250 Server Hallway/Dosimetry pinnserv 198.60.162.230
SunBlade 2000 Planning Station Hallway/Dosimetry pinnblade2 198.60.162.232
SunBlade 2000 Planning Station Dosimetry pinnblade1 198.60.162.231
SunBlade 2000 Planning Station Dosimetry pinnblade3 198.60.162.233
SunBlade 2000 Contouring Station CT Sim pinnacq1 198.60.162.234
5. prediction met the ‘Van Dyk’ criteria regarding acceptable deviations2
. In Table 2, some
of these criteria are displayed.
Table 2: Van Dyk Criteria2
Photon Beam Criteria
Central Ray (Build-up Excluded) 2%
High Dose Region – Small Dose Gradients 3%
Large Dose Region (>30%/cm) 4mm
Low Dose Region – Small Dose Gradients 3%
Since these depth dose curves and profiles were used to actually model the beam, the fact
that they met the Van Dyk criteria was not considered a sufficient check of the beam
model. As a more independent check, a number of scans not used to model the beam
were checked against the Van Dyk criteria. These scans included 13x6, 6x13 and 7x7 cm
fields at different depths. Although the 7x7cm field passed, the non-square fields were
discovered to be in error, as discussed in the ‘Anomaly’ section below.
Profiles and Depth Dose Curves
In addition to comparison with Van Dyk, a number of additional checks were
conducted for the initial beam model, Elekta 7.0. For example, the mean percent error
near the center of the profiles was plotted as a function of field size, for different depths
and scan directions. The mean percent error was computed as: Mean % Error =
(Computed Dose – Measure Dose)/(CAX Dose) x100. In Figures 1 and 2, the Mean %
Error and the Mean Square % Error are plotted as a function of field size for different
depths and scan directions for 6MV.
An Excel spreadsheet that contains these error data, along with plots is titled
‘Pinnacle_Elekta_Error.xls’. A copy of it can be found in
H:commonPhysicsPinnacleElektaDocuments .
2
Van Dyk, J., R.B. Barnett, J.E. Cygler, and P.C. Shragge. 1993.
Commissioning and quality assurance of treatment planning computers.
International Journal of Radiation Oncology, Biology and Physics
26(2):261-273.
6. Mean %Error - Center
-1.00
-0.50
0.00
0.50
1.00
1.50
0 10 20 30 40
Square Side (cm)
Mean%Error
Depth = 1.6cm - y scan
Depth = 5.0cm - y scan
Depth = 10.0cm - y scan
Depth = 20.0cm - y scan
Depth = 1.6cm - x scan
Depth = 5.0cm - x scan
Depth = 10.0cm - x scan
Depth = 20.0cm - x scan
Figure 1: Mean % Error for 6MV Elekta Data.
Mean Square % Error - Center
0.00
0.20
0.40
0.60
0.80
1.00
1.20
1.40
1.60
1.80
0 10 20 30 40
Square Side (cm)
MeanSquare%Error
Depth = 1.6cm - y scan
Depth = 5.0cm - y scan
Depth = 10.0cm - y scan
Depth = 20.0cm - y scan
Depth = 1.6cm - x scan
Depth = 5.0 cm - x scan
Depth = 10.0cm - x scan
Depth = 20.0cm - x scan
Figure 2: Mean Square % Error for 6MV Elekta Data.
Anomalies 7.0 - Elekta
As indicated above, some irregularities were discovered upon investigation of the
initial Elekta 7.0 beam model: 1) Investigation of the non-square fields revealed that the
x and y coordinates of these fields were switched. 2) It was also discovered that the
depth dose curves for the small (1x1cm) IMRT fields did not meet the Van Dyk criteria.
These facts were relayed back to the beam modelers at Philips.
The beam was remodeled with the x and y coordinates reversed. In addition, a ‘split’ in
the model was performed so that for fields smaller than 4x4cm, a smaller grid size (2mm
7. as opposed to 4mm) was chosen. This smaller grid improved the accuracy with which
small fields were predicted so that the IMRT fields did in fact pass the Van Dyk criteria.
Anomalies 7.4 – Siemens
There were also some anomalies found with the Siemens 7.4 beam model: 1) The
allowed wedge orientations included heel in and heel out, in conflict with the actual
allowed wedge orientation on this machine. It was not possible to toggle these
orientations off under the ‘Machine Editor’ panel obtained from the ‘edit’ button in the
‘Photon Physics Tool’. 2) The photon output factor was found to be in error by ~5%.
This was discovered during evaluation of comparison patient plans.
Regarding the first anomaly, Philips is remodeling the beam so that these
disallowed wedge orientations can be toggled off. Regarding the second anomaly, this
was changed to match tabulated depth dose data.
Comparison with Theraplan
Once these above anomalies were understood, the beam models could be
officially ‘commissioned’ in Pinnacle. This involves signing the ‘Commissioned by’
statement in the ‘Commission Machine’ panel, obtained by pushing the ‘Commission . . ‘
button in the ‘Photon Physics Tool’. Prior to this, no actual planning could be done.
After the machines were commissioned, a final check on the TPS was a
comparison of patients planned on both Pinnacle, and the TPS that preceded it,
Theraplan. It was decided that ten comparison plans should be completed for each
machine and agreement checked. These plans were chosen to represent a variety of
different disease sites, and the number of MUs was forced to be the same under both
TPSs. A copy of the Excel spreadsheet titled ‘Pinnacle vs Theraplan.xls’ that contains
these comparisons can be found here H:commonPhysicsPinnacle. The plans compared
favorably, with dose differences generally below 5%.
After the beam models passed this final test, they were released to the clinic for
patient treatment. Regarding the wedge anomaly discovered in the Siemens 7.4 model:
Dosimetry and physics staff were made aware of the potential problem and instructed to
pay close attention to the wedge orientation, so as to make sure that heel in and out are
avoided on all of the Siemens plans. This model will be replaced upon receipt and
commissioning of the updated/revised model from Philips.
Conclusion
The Commissioning of a TPS is an important and necessary step in order for it to
be used safely, and effectively to treat patients in the Department of Radiation Oncology.
It is, however, not sufficient. A continued degree of vigilance is required in order for
anomalies to be detected, and potentially hazardous situations to be anticipated. An
efficient way for vigilance to be maintained is that a maximum number of people become
8. familiar with this new TPS. If this happens, different perspectives can increase
awareness so that unforeseen problems can be forecast and avoided.