Presented at AAPM2015 in the "Quantifying QA" session. A decision tree model was built using features extracted from DICOM-RT (radiation therapy treatment plans) to predict where MLC errors systematically occur.
The errors were then incorporated into the planned MLC leaf positions, and these positions were shown to be a better representation of the actual delivery than were the planned positions.
Utilizing the predicted positions raises gamma passing rates, and also shows that DVH curves presented in conventional treatment planning systems are not accurately representing the dose that linear accelerators are physically capable of delivering.
Call Girls Indiranagar Just Call 👗 9155563397 👗 Top Class Call Girl Service B...
AAPM2015 - Statistical Learning to Predict MLC Errors
1. A Statistical Learning
Approach to the
Accurate Prediction
of MLC Errors During
VMAT Delivery
Joel Carlson
Jong Min Park
So-Yeon Park
Jong In Park
Yunseok Choi
Sung-Joon Ye
2. MLCs move in complex
ways
Prostate Plan: Low complexity H&N plan: High complexity
** ~50x speed **
3. Complex movements lead to errors in leaf
positions
How can we quantify these
errors? Planned Delivered
Goal:
Create a more realistic representation in the TPS of
where the MLC leaves will be upon delivery
How can we predict these
errors?
How do these errors impact
dose delivery accuracy?
4. How can we quantify these
errors?
How can we predict these
errors?
How do these errors impact
dose delivery accuracy?
5. Quantifying the difference
between planning and
delivery
74 H&N or Prostate VMAT plans from 3 institutions
Dicom RT
Planned
Positions
DYNALOG
Delivered
Positions
Errors
(Prediction Target)
Difference
6. How can we quantify these
errors?
How can we predict these
errors?
How do these errors impact
dose delivery accuracy?
Goal:
Create a more realistic representation in the TPS of
where the MLC leaves will be upon delivery
7. We first extracted a rich
feature set from the DICOM-RT
plan files
Using only information available before plan delivery
~150 features* quantifying the MLC leaf motion
*list available, just ask!
8. Using a validation set we
chose the best features
All Features
Build
Model
Vary
Features
Error*
Minimized?
Predictions
No
Final Model
Yes
Predictions
Report
Statistics
Training Plans
(N = 3)
Validation Plans
(N = 6)
Testing Plans
(N = 65)
*Root Mean Squared Error
13. How can we quantify these
errors?
How can we predict these
errors?
How do these errors impact
dose delivery accuracy?
Goal:
Create a more realistic representation in the TPS of
where the MLC leaves will be upon delivery
16. There exist errors between planned and delivered
MLC positions
These errors are predictable at the planning stage
Utilizing predicted positions:
• Increases gamma passing rates
• Leads to a more realistic representation of where
the leaves will be upon delivery
In conclusion
17. Explore differences in patient DVHs
• In progress
Integrate predictions into TPS
• Will give planners a better view of what will be
delivered
Publish fully reproducible code and data
Future work
18. A Statistical Learning
Approach to the
Accurate Prediction
of MLC Errors During
VMAT Delivery
Joel Carlson
Jong Min Park
So-Yeon Park
Jong In Park
Yunseok Choi
Sung-Joon Ye
Thank you for
listening!
21. • Numerical Values:
• Error Magnitude
• MLC Index
• Width and Mass of leaf
• Positions
• ±5 CPs
• ±5 CPs of both adjacent
MLCs
• Velocities
• ±5 CPs
• ±5 CPs of both adjacent
MLCs
• Accelerations
• ±5 CPs
• ±5 CPs of both adjacent
MLCs
All Features
• Categorical
• Whether the MLC was
previously at rest, coming
to a stop, moving before
and after, single CP
movement
• Whether adjacent MLCs
were both moving in the
same direction, both
opposite, same/opposite,
or at rest
• Moving towards (push) or
away (pull) from the
isocenter
• The CP at which the error
occurred
22. Cubist
• “…is a rule-based
model where a tree
is grown, and each
of the terminal
leaves contain
regression models.
These models are
based on the
predictors in
previous splits.”
Editor's Notes
Today, I’m going to tell you why your TPS is not accurately telling you the dose your linac is physically capable of delivering, and how we can start taking steps towards fixing that problem
In VMAT plans we modulate several parameters.
One of these being MLC movements. Plans for different sites have mlc movements of varying complexity. As we can see
Particularly in complex plans
MLC errors though to be dominant error causing mechanism
Therefore, one statement that we can make is that. This raises several questions
Position,
Velocity, Momentum, Acceleration,
Direction of movement, At rest/Moving,
Previously at rest, Stopping,
Mass, Width,
Direction of adjacent leaves, Velocity of adjacent leaves,
Acceleration of adjacent leaves,
etc…
Candidate for deletion
Inst 2: Larger proportion of prostate plans…even moving MLC leaves are moving slower?
Approx. > half of CPs the leaf is moving in HN plans
Approx 25% moving in the course of prostate plan
~6 mins
From institution 1 only
3%/3mm : 1.75%
2%/2mm: 3.3%
1%/2mm: 3.5%
All significant at the, p < 0.05