IMRT Planning Objectives with Corvus

1
IMRT Planning Objectives with Corvus

• Walter Grant III, Ph.D.
• Baylor College of Medicine
• The Methodist Hospital
• Houston, Texas

2
Some Initial Comments

• Corvus was designed by a physician, not a
  physicist.
• His intention was that physicians would run their
  own plans.
• So when you ask, Why did they do it that way?,
  remember the above.

3
Some Initial Comments

• That physician is Dr. Mark Carol, a
  neurosurgeon.
• As the 1st clinical user of the NOMOS system, I
  had the luxury and privilege of learning from
  the master.

4
What I will to do today

• I will share some of the initial philosophical
  changes that occurred in our transition from
  forward planning to inverse planning.
• Ill discuss the changes from Corvus 1.0 (Peacock
  Plan) to Corvus 5.0 in planning philosophy.
• Ill review the intentions of the Tissue Types
  and some approaches to getting better plans, if
  possible.

5
Philosophy

• Why am I doing IMRT at all?
• To preserve some normal structure(s).
• Spinal cord, optic chiasm, etc.
• Maybe a previously irradiated area
• Brain mets
• Its never really just to deliver a conformal
  dose to a target.
• Thats easy if there is no critical structure
  close or the dose is low.

6
Philosophy

• For me, that was the break through in
  understanding how to approach inverse planning.
• The BCM mantra became, Protect the normal
  structure as much as you need and then put as
  much dose in the target as you can.
• And it is the inverse of what I was trained to do
  for many years.

7
Philosophy

• This was occurring in 1993 and it took only a
  couple of plans to identify where we were weak.
• We didnt know the biological limits of partial
  volume irradiation of structures.
• Its 2004 and we still dont.

8
Philosophy

• So if you are struggling with planning, try
  keeping these 2 principles in mind

• Protect the structure first and then dose the
  target.
• We dont really know partial volume tolerances,
  so the best plan will be subjective, not
  objective for now.

9
Philosophy

• If the best plan is subjective, how will the
  physician decide that hes got the best plan?

• The same way he always has --- when its time to
  treat the patient.
• Youll run plans that change a few percent here,
  a few percent there, just to give the physician
  peace of mind.
• We ask our physicians, Am I treating the patient
  or am I treating the doctor? and they are very
  honest about it.

10
Philosophy

• So how can I help the physician?

• I do it by considering RTP as a boundary problem.
• We run an Efficient Plan that basically covers
  the target and blasts the structures.
• Remember this as I use information from this plan
  in obtaining my clinical plan.
• We run a Conformal Plan that protects the
  structures at the expense of under dosing the
  target.

11
Philosophy

• So how can I help the physician?

• This helps the physician recognize what his
  tradeoffs are going to be.
• And it is usually when he shares his desires and
  fears.

12
A Quick Wrap

• I have shared some planning experience and
  philosophy with you.
• I believe these are generally applicable, not
  Corvus specific.

13
Peacock Plan

• The original Peacock Plan (later called Corvus
  1.0) had features that some of you may not have
  experienced, so I want to cover the significant
  ones.
• It was not a Partial Volume Planner.
• You entered a Goal Dose and controlled the
  importance with a weight from 0 to 2.
• Or you lowered/raised the Goal Dose.
• Tissue had a weight control.

14
Peacock Plan Rx Panel
15
Peacock Plan

• The cost function was a simple quadratic
  expression whose steepness was controlled by the
  weight
• Cost W ?(Goal Dose - Dosei)2

16
Corvus 2.0 and up

• Dr. Carol believed that the only way to gather
  the partial volume information was to plan with
  it, thereby encouraging us to record the results.
• So a PVP panel was introduced with Corvus 2.0
• But so was a new concept of optimization and
  Goal control.

17
Corvus 2.0 and up

• Goal or Limit
• under or over
• Minimum
• Maximum

• PV planning requires the creation of a DVH

18
Corvus 2.0 and up

• The Cost Function is no longer a quadratic on one
  point.

• Its control of 3 points and 4 areas.
• The use of area optimization creates a planning
  rule that Ill point out later.

19
Corvus 2.0 and up
20
Corvus 2.0 and up

• The shape of the Cost Function is a combination
  of linear and exponential components that were
  selected by Dr. Carol.

• And since he wants physicians to plan, the
  combinations are determined by the selection of a
  Tissue Type and not by the choice of a
  weight.
• This is not the way a physicist wants to do
  things.

21
Corvus 2.0 and up

• Since Dr. Carol is a neurosurgeon, his training
  is to drive dose out of the normal tissue into
  the target à la SRS.

• Corvus controls that with the normal structure
  called Tissue which is all the non-Target
  tissue.

22
Corvus 2.0 and up

• In Corvus 1.0, he thought Tissue should always
  have a weight 2, but to treat malignant tumors
  we wanted more homogeneous dose distributions, so
  I would set values close to 0.02 just to
  aggravate him.

• One had to be careful that the maximum dose was
  in the Target because Tissue insures that!
• He got even with me in later versions as Tissue
  now cant be turned off.
• And we found ways to take advantage of that.

23
Cleanup Time

• When you approach Corvus planning, you do not
  want to think like a physicist.
• Dont worry about what is under the hood, just
  learn to enjoy driving the car.
• But do recognize, as a physicist, that there are
  some traps that you should avoid.
• Ill try to point them out next.

24
Corvus 5.0 Planning

• We have a large database of clinical outcome for
  mainly 3 sites.
• Cranial, HN and Prostate
• The goal is to try to match what has given
  clinically acceptable outcome
• Doses
• MUs

25
Corvus 5.0 Planning

• As information for the non-Corvus users, there
  are 4 issues to address in planning

• Complete the PVP
• Pick an optimizer
• Apply uncertainties for setup error and organ
  motion.
• Place your beams.

26
Corvus 5.0 Rx Panel
27
Picking Types
28
Prescription Instructions
29
Corvus 5.0 Optimizer
30
Corvus 5.0 Optimizer
31
Corvus 5.0 Optimizers

• It is Simulated Annealing
• Optimizes both pencil beam intensities and
  monitor units (MU).
• Uses all intensity levels and constrains to
  discrete levels at end.
• Applies a Gradient Descent optimization at end.
• Will always give the best plan.
• Will always require the most MU.
• Use the Efficiency Slider Bar

• Continuous

32
Corvus 5.0 Optimizers

• It is Simulated Annealing
• Optimizes discrete pencil beam intensities.
• Since MU settings are not optimized, all MU
  settings are equal.
• Applies a Gradient Descent optimization at end.
• Usually as good as Continuous for most plans.
• Will always require less MU.
• As much as 1/3 has been seen compared to
  Continuous.
• We use it as default in 5.0 for MIMiC delivery.

• Discrete

33
Corvus 5.0 Optimizers

• It is Simulated Annealing
• Optimizes MU settings only.
• Basically no modulation so just a conformal plan.

• Continuous (MU Only)

34
Corvus 5.0 Optimizers

• Begins by optimizing MU Only
• Then optimizes beamlets using a gradient descent
  algorithm.
• Probably will produce more homogeneous results
  requiring less MU.
• Good for simple plans (prostate cranial), but
  poor for complex plans (HN).
• We use it as the default for our MLC delivery.

• Smooth Downhill Search

35
Corvus 5.0 Immobilization
36
Corvus 5.0 Beams
37
Now Comes the Hard Part

• At least its hard for me because I have to share
  11 years of clinical inverse planning experience
  in about 11 more minutes.
• I remind you that there are only clinically
  acceptable plans, not correct plans.

38
Now Comes the Hard Part

• My goal is not to get the best plan first.

• Remember, I dont believe the physician knows
  that plan or shares that with me initially.
• My goal is to know what to do on the 2nd
  iteration in case the physician isnt satisfied.

39
How I Approach 5.0 Planning

• I am going to cover the art of putting numbers
  in the Corvus Rx Panel.
• Im going to discuss how I prepare myself to get
  the next iteration.
• Im going to remind you now that Ill think like
  a doctor because thats the way Dr. Carol
  designed the system.

40
Corvus 5.0 Rx Panel
41
How I Approach 5.0 Planning

• Fill in the panel with true clinical numbers.
• This will help you build a foundation for
  expansion to other disease sites.
• Run an Efficient Plan
• Set all Structures as Reference Tissue Type.
• If you fail to dose the Target, then something
  is wrong.
• Check Anatomy
• Check entries in Rx Panel.
• For MIMiC it could be a required isocenter shift.

42
How I Approach 5.0 Planning

• If you do dose the target, then see what
  structures meet their goals.

• Most distal structures can meet goals because
  Tissue is designed to drive the dose out of
  normal tissue into the target.
• I leave those as Reference Structures so that the
  optimizer doesnt spend time minimizing the Cost
  Function on non-important structures.
• Gotcha Mark!

43
How I Approach 5.0 Planning

• You can run a Conformal plan by loosening the
  limits for the targets.

• Lower minimum dose.
• 1020 volume under.
• Reference Target is useful only if there is more
  than one target.
• Doesnt help much except to see how many MUs are
  required to treat.

44
How I Approach 5.0 Planning

• When entering the numbers, follow the rules in
  the Prescription Information window.

• Structure Limits should be set independent of
  Structure Type.
• The dose limits should be clinical values.
• If 50 of the parotid gland can exceed 22 Gy,
  then put those numbers in.

45
How I Approach 5.0 Planning

• If the target is more important than the
  structure, set the Structure Maximum dose to
  Target Goal dose.
• If the structure is more important than the
  target, set the Target Minimum dose to the
  Structure Maximum dose.
• If the structure is critical, set the Target
  Minimum dose to the Structure Limit dose.

• Help the system understand where you want it to
  put the steep dose gradient.

46
How I Approach 5.0 Planning
And one that I still have to fight with my staff,

• Never set Structure Minimum to 0.
• Remember that Corvus is optimizing areas as well
  as control points.
• If I dont know any better, I set the minimum
  about 50 of the Limit and adjust when I see the
  Statistics Panel.

47
So Lets Get an Acceptable Plan

• If we can put all these steps together, then we
  should be able to get an acceptable plan.
• Here are the iterations we teach at Baylor.

48
So Lets Get an Acceptable Plan

• Step 1
• Insure that the anatomy is correct and that there
  are no stray voxels.

49
Where did these come from?

• Anatomy review

50
So Lets Get an Acceptable Plan

• Step 2
• Enter the desired target dose into the Rx panel.

• Follow the rules at the bottom of the screen

51
So Lets Get an Acceptable Plan

• Step 3
• Enter appropriate values for structures into the
  Rx panel.

• Follow the rules at the bottom of the screen.

• If 20 over is OK, put 20 not 10.

52
So Lets Get an Acceptable Plan

• Step 4
• Adjust values to be consistent.

• If a structure and target touch, make the minimum
  for the target and the maximum for the structure
  the same value.

53
So Lets Get an Acceptable Plan

• Step 5
• Assign Tissue Types.

• Start with Basic unless its a SRS case.

• This defines a baseline for further iterations.

54
So Lets Get an Acceptable Plan

• Step 6
• Adjust types to be consistent.

• i.e., dont mix Basic with Homogenous

• Some structures are not players and can be set
  to Reference.

• Some structures (like lenses) cant be controlled
  and should be set to Reference.

55
So Lets Get an Acceptable Plan

• Step 7
• Use Growth on Immobilizer and Uncertainty.

• Are you creating a PTV or are you trying you make
  the statistics for the CTV look good?

56
So Lets Get an Acceptable Plan

• Select Original Dose slices for calculation
• Use 10 Intensity levels
• Pick an optimizer
• Is the plan simple or complex?
• Calculate the plan.

• Step 8
• Enter Beam Geometry.

57
So Lets Get an Acceptable Plan

• Step 9
• Evaluate the plan.

• If information was entered correctly, then the
  results should be acceptable.

• If not.

58
So Lets Get an Acceptable Plan

• Step 10
• Check limits and goals again.

• If OK, then .

59
So Lets Get an Acceptable Plan

• Step 11
• Adjust Tissue Types.

• If a structure has a dose that is not met, set it
  to Critical.

• If a structure has its limits met but they dont
  need to be met, change it to Expendable.

• If you want a more homogeneous dose, select that
  but be aware that nearby structures will suffer.

60
So Lets Get an Acceptable Plan

• Step 12
• Try another optimizer.

• Maybe you need more modulation.

61
So Lets Get an Acceptable Plan

• Step 13
• Adjust Dose Goals as a last resort.

• Instead of decreasing percentages to get better
  results, you really need to increase them, i.e.,
  if changing a Structure to Critical failed to
  protect it, then you may have to increase the
  Below for the Target.

62
So Lets Get an Acceptable Plan

• That should do it.
• We usually do no more than 2 or 3 plans.

63
Thats a Wrap

• Thanks to all of you who sat here this afternoon.
  
• I hope I shared some useful information.
• You can download a copy at
• http//www.thegrants.us/wg3acmp2004.pdf
• Thanks to Mike Mills for inviting me to give this
  talk.

64
And a Special Thanks to

• Our first grandchild, Jennifer Bethany Powell,
  whose reluctance to join this world created the
  opportunity for me to get this presentation
  finished on schedule.

65
Beth Powell