Commissioning inverse planning and leaf sequence routines

1
Commissioning inverse planning and leaf sequence
routines

• A. W. Beavis PhD SRCS
• Principal Physicist,
• Hull and East Yorkshire Hospitals (NHS) Trust
• (Hon) Senior Research Fellow,
• University of Hull

2
THANK YOU!

• Prof. Dan Low (Mallinckrodt)
• Prof. Art Boyer/ Lei Xing (Stanford)
• Cal Huntzinger (Varian)
• Prof. Rock Mackie (U Wisconsin)
• Bruce Curran (NOMOS Corp)
• CMS family

3
Inverse planning algorithms

• Dose optimisation algorithm
• Computes the required beams
• Leaf Sequencing algorithm
• Computes the segments
• Leaf positions and associated MUs

4
Talk based on our experience

• First IMRT treatment was January 2002.
• Using CMS XiO (nee FOCUS) and Varian 600CD
  (120MLC)
• Training
• project leader over several years at Stanford,
  Mallinckrodt, ..
• Junior Physicists Varian course (Cal
  Huntzinger), AWB training
• Dosimetrists/ Radiographers in-house course

5
IMRT planning

• Inverse planning algorithm
• Computes the optimal beams to produce the desired
  distribution
• Leaf Sequencing algorithm
• Aim to review a few issues that require extra
  commissioning work in addition to that for
  conventional 3D-CRT modelling

6
Prescription

• Two beam cartoon!

• give target (yellow) 100 dose
• give maximum of 50 to critical organ (red)
• give rest of body maximum of 20 dose
• generally choose the number and direction of
  beams to use

Planning Target Volume PTV
Organ at risk OAR
7
Compare the (first pass) calculated doses to
those required at each point in the patient
dose 1 calcd lt dose required
beam a
a1
a3
dose 12 calcd gt dose required
a2
a4
b1
b2
b3
b4
beam b
8
Suggest changes to beamlet weights (e.g. a1 b1)
based on the difference computed at dose point (1)
increase a1 and b1
beam a
a1
a3
decrease a4, possibly maintain b3 (for
contributions to points 9 10)
a2
a4
b1
beam b
b2
b3
b4
changes are computed from dose differences seen
dose pts 9 and 10
9
Recompute the doses at all points, recompare to
desired distribution, make further adjustments to
(a1, a2, a3, a4, b1, b2, b3, b4) and . repeat
until dose calcd - dose req 0
dose 1 calcd dose required
beam a
a1
a3
dose 2 calcd dose required
a2
a4
b1
b2
b3
b4
beam b
10
Implication on the dose calc

• Must use a convolution type algorithm
• None others at least in our system
• SIITP algorithm (currently)
• Calc the dose to open field
• Vary the weights of the beamlets
• Produce the required fluence maps
• ? leaf sequencer

11
Commissioning issues

• If you use solid wedges then re-tune the model
  removing any compromised in open beam dosimetry
  that the wedges may have demanded
• Not going to use wedges!
• More about education and understanding the
  implementation in your system

12
Delivery by MLC Leaf Sequence

• operation to convert required Intensity map to
  deliverable field using an MLC
• consider the 2-D surface to be a set of 1-D
  independent profiles

Leaf pair considered here
13
1-D sequence generation

• typical to consider dividing the intensity
  profile into steps of equal widths (1cm) each
  step representing intensity levels at fixed
  increments.
• can then use a step and shoot or sliding window
  to deliver this

• Bortfeld-Boyer algorithm

14
Segment 1

• leaf A 5 cm
• leaf B -3 cm
• deliver 10 MU

100
90
80
70
60
50
40
30
20
Opening in leaf pair
10
0
deposited Dose/ Intensity
15
Segment 2

• leaf A 5 cm
• leaf B 0 cm
• deliver 10 MU

100
90
80
70
60
50
40
30
20
10
0
16
Segment 3

• leaf A 5 cm
• leaf B 0 cm
• deliver 10 MU

100
90
80
70
60
50
40
30
20
10
0
17
Segment 4

• leaf A 4 cm
• leaf B 0 cm
• deliver 10 MU

100
90
80
70
60
50
40
30
20
10
0
18
Segment 5

• leaf A 4 cm
• leaf B 1 cm
• deliver 10 MU

100
90
80
70
60
50
40
30
20
10
0
19
Segment 6

• leaf A 4 cm
• leaf B 1 cm
• deliver 10 MU

100
90
80
70
60
50
40
30
20
10
0
20
Segment 7

• leaf A 4 cm
• leaf B 1 cm
• deliver 10 MU

100
90
80
70
60
50
40
30
20
10
0
21
Segment 8

• leaf A 2 cm
• leaf B 1 cm
• deliver 10 MU

100
90
80
70
60
50
40
30
20
10
0
22
Segment 9

• leaf A 2 cm
• leaf B 4 cm
• deliver 10 MU

100
90
80
70
60
50
40
30
20
10
0
23
Segment 10

• leaf A 1 cm
• leaf B 4 cm
• deliver 10 MU

100
90
80
70
60
50
40
30
20
10
0
24
Segment 11

• leaf A 1 cm
• leaf B 5 cm
• deliver 10 MU

100
90
80
70
60
50
40
30
20
10
0
25
Leaf sequencer

• Repeat the algorithm for each leaf pair and for
  each beam
• Collate 1-D solutions to create 2-D solution
• Done to minimise leaf motion
• Minimise tongue and grove leakage
• Following this step, perform a FORWARD
  calculation based on the set of segments

26
Implications smaller fields

• Run a convolution calculation that adds up the
  dose due to each segment
• Segments tend to be smaller than the portal!!
• Maybe much smaller
• Convolution model should be tuned for achieving
  accuracy at smaller field sizes
• Can do this at expense of larger fields

27
Fine tuning MLC-defined small fields

• No modelling tools available yet
• So, needed to plan MLC-defined fields
• 3x3 MLC field 15x15 collimator
• Compare to measured data
• Overlay or electronically
• Vary Sigma to improve
• Didnt need to do anything

28
Implications smaller fields

• May need to model fields down to 1 cm x 1 cm
• Need to provide output factors and scatter
  factors for 1x1, 2x2, 3x3
• Need to ensure the model is optimised to produce
  accurate small field PDDs and profiles
• Need to measure these for verification purposes
• Not usually considered necessary for conventional
  treatments!!

29
Implications head scatter component, Sc

• For very small fields can have problems with the
  dose contribution due to the Sc fall-off

30
Implications head scatter component

• We have a scaling factor that improves the dose
  accuracy of very small fields
• Set it empirically with a simple experiment
• Create beam with 2x2 segment
• Delete all other segments and vary parameter to
  find agreement between dose/MU and measured value

31
Segment removal
In our system can remove very small segments

• Can set the minimum segment area wish to accept
• Can prune segments once they have been generated

32
Reduction of segments?

• Plan 1 used 5 intensity levels
• Beam on time 4.35 mins
• Plan 2 used 10 intensity levels
• Beam on time 7 mins
• Plan 3 used 10 but deleted segments less than 1
  cm2 and recalculated
• Beam on time 5.5 mins
• So, difference 1 (2,3)
• 5 OR 10 intensity levels
• Difference between 23
• /- segments with small spatial contribution

Study was done in version 3.0 new sequencing
algorithm in 3.1 reduces these times
33
Implications leaf transmission

• many contributions to field due to leaf
  transmission only
• i.e. contribution from closed part of segment
• So may need to optimise the MLC leaf transmission
  
• We didnt need to do anything

34
Rounded Leaf End effect.

• Varian MLC leaf design
• Get differential transmission through leaf

• generally the MLC is calibrated to the light
  field
• Should correct the leaf positions defined by the
  leaf sequencer to reduce any leakage due to RLE

Imax
35
FOCUS IMRT parameters

• RLE offset
• Shot films
• Puts leaf ends at x and y cm ? gap xy
• Put leaf ends at x-D and y-D ? gap xy 2D
• Found offset, D 0.5 mm was optimal (6X)

EPID D 0 mm
EPID D gt 0 mm
EPID D gtgt 0 mm
36
A few words about our routine QA process

• Ion Chamber dose/MU verification at sensible
  places
• Checks of the intensity maps
• Checks of the dose distribution

37
How can we check we get the dose distribution the
plan gave us?
38
CMSFOCUS QA tools

• Introduced by NOMOS in Peacock
• we can take the beams generated for the patient
• apply them to a simple (cubic) phantom
• Defined by user
• Recalculate distribution

39

• Put a film between the slabs of the phantom and
  irradiate it!
• Can extract the 2D slice distribution and
  compare to the film

40
DoseProfile/ output tool

• IN PLAN DISPLAY MODE
• Select axial/ sagital or coronal planes of
  interest
• Coronal plane under 5cm depth

1) MUdose verification

• Can produce ASCII files of 1-D dose profiles
• These are x/y planes through ion chamber position
  to identify uncertainties in measurement

ASCII file
41
Film Dosimetry methods
Curve done for each batch of film

• Kodak EDR2 film (Linear to 5 Gy, saturates at
  7Gy)
• Multidata scanning densitometer (dump data in
  ASCII format)
• MATLABown analysis/ graphics software
• To achieve good consistent results, pay attention
  to
• 1) Processing (we use one in Radiology good
  QC!)
• 2) Calibrate film according to measurement!!
• Beavis et al. Implementation of the Varian EDW
  into a commercial RTP system. Physics in Medicine
  and Biology 41 1691-1704 (1996)

42
2) Individual beam verification

• In FOCUS use the DOSE_PROFILE tool to generate a
  2-D dose plane at 5cm deep in perspex (PMMA)
  block
• Identify the maximum dose in the 2-D matrix and
  compute the MU required to give 4.5 Gy to this
  point
• Calc MU required to give 1.5, 3, 4.5 Gy from a 10
  x 10 field
• Take three such films and use a simple 3 pt.
  Calibration curve to compare to previous graphs

Film registered to Linac CAX
5 cm Px
Film

• Measuring individual intensity/dose map

43
Film cassette and registration

• Made a cassette for taking dose map films
• Has a centralising cross scribed on with holes
  for a pin-punch on its axes
• Can uniquely locate the beams CAX on processed
  film
• So can accurately position the film read scanning
  and identify co-ordinates

44
DoseProfile/ output tool

• Can produce ASCII files of 2-D dose maps
• Select axial/ sagital or coronal planes of
  interest
• Coronal plane under 5cm depth

ASCII file
45
Fig A comparison of absolute dose measured from
film and that computed by the CMS system.
Isodoses shown 0.5, 1.0, , 4.5Gy
46
3) Composite dose/ dose plane verification

• Compression straps used to hold slab phantom
  together
• Multiple films can be inserted between the slabs
• Set isocentre to centre of phantom

47
Film Calibration methods

• In FOCUS use the QAPLAN tool to take the set of
  beams (Patients) and apply them to a simple cubic
  phantom.
• Compute the MUs required to deliver a maximum of
  4.5 Gy to the film.
• Calc MU required to give 4.5 Gy to dmax for a 10
  x 10 field (95 cm TSD), also create an ASCII file
  via DOSE_PROFILE of the depth dose profile to
  use to calibrate a depth dose film

Film

• Measuring dose distribution/ 2-D dose plane

48
Computation and ASCII output of dose planes
within IMRT Phantom
Trans-axial plane viewed
ASCII file
49
4.5Gy (to maximum point) then 0.25 Gy decrements.
absolute dose comparison
50
Conclusion

• We have implemented the IMRT option of the CMS
  planning system
• We are using it routinely!
• The implementation was made easier by a good
  understanding of the algorithms and their
  implementation

51
Thanks to

• My governors
• My colleagues
• My mentors and friends in the IMRT world ?

Grey Owl A Great British Export