Equipment QA - SBRT Singapore

Report
6. CLINICAL
IMPLEMENTATION AND
SBRT QUALITY
ASSURANCE
Patient Specific QA
Equipment specific QA
In vivo Dosimetry
TG-142 and TG-101 guidelines
Process assessment
Clinical challenges
Jeffrey Barber, Medical Physicist
IAEA RAS6065, Singapore Dec 2012
2
Useful References
• AAPM TG-101 Report: SBRT
• AAPM TG-142 Report: Medical Linac QA
• AAPM TG-179 Report: CT-based IGRT QA
0.5mm
gantry
locus
2mm
image
reg
2mm
couch
locus
3% dose
delivery
0.5mm
kV-MV
1mm
laser loc
10mm
target
respiratory
motion
2mm
immob
movement
2mm
contouring
variation
Quality Assurance
• Physicists should check individual parameters and
combined processes
• If you check everything in isolation, how do you
know what you are doing at the end
• TG-142 and TG-101 are guidelines. Lots of advice
on how to do things, how to investigate and how
to develop local protocol
• The future TG-100 proposes a different approach
QA Approach
• Perks et al (2012) IJROBP 83 p1324
• Fault Mode Effects Analysis (FMEA)
• Process Engineering concept used
to focus QA efforts on most
practical problems
1.
Map your processes (flowchart,
tree, etc)
2.
Give any foreseeable fault a
weighted score
• likelihood of Occurrence
• Severity of fault
• likelihood of being Detected
3.
Then add QA processes to
address the potential faults, with
most effort focused on highest
scores
QA Approach
8
QA Approach
• FMEA promises to increase the efficiency and
effectiveness of the testing required
• But FMEA takes a lot of resources and time to set
up
• Current guidelines are effective, if intensive
• Quality Assurance can be categorised as:
• Equipment QA
• Patient-specific QA
9
EQUIPMENT QA
Equipment QA
• TG-142 Daily
Equipment QA
• TG-142 Monthly
Equipment QA
• TG-142 Annual (1)
Equipment QA
• TG-142 Annual (2)
Equipment QA
• TG-142 MLC
Equipment QA
• TG-142 Imaging (1)
Equipment QA
• TG-142 Imaging (2)
17
Equipment QA
• ASTRO
Equipment QA
• TG-101
Equipment QA
• TG-101
Equipment QA
• TG-101
Equipment QA
• TG-101
Equipment QA – kV/MV coincidence
Room Lasers
Imaging Isocentre
Radiation Isocentre
Equipment QA – kV/MV coincidence
Room Lasers
Imaging Isocentre
Radiation Isocentre
Equipment QA – kV/MV coincidence
• Winston-Lutz type tests check centre points
Equipment QA – kV/MV coincidence
Sharpe et al, Med. Phys. 33, 136-144, 2006
Equipment QA – kV/MV coincidence
• Elekta: Planar images are
uncorrected. Flexmap
offset saved in DICOM
header. 3D reconstructions
include the correction.
• Varian: Flex is included in
robotic arm so each image
is corrected.
• If flex needs calibrating, it
will be visible in the
reconstructed images
Bissonnette
Equipment QA – Daily Checks
• Daily IGRT QA
Set up phantom with known offset
2. Image, register, check offset is right
3. Correct couch, re-image, check residual error
4. Visually inspect the new phantom position
1.
28
Equipment QA – Image Quality
Rings
Streaks
Capping
Motion
29
Equipment QA – Image Quality
• Most important Image Quality parameter is
spatial accuracy and scaling
30
Equipment QA – Image Quality
• Most important Image Quality parameter is
spatial accuracy and scaling
Machine QA – MLC Accuracy
• Using Picket Fence and Garden Fence beams
• Film
• EPID
• Array Device
• Analysis is the hard part
• How good is your eye?
• How good is your image processing?
• Lots of commercial solutions available
Machine QA – MLC Accuracy
33
PATIENT-SPECIFIC QA
Patient Specific QA
high doses
+ small volumes
+ complex beam arrangements
+ moving structures
= need for patient-specific QA
• Verify Dose
• Verify 3D Distribution
Patient Specific QA
• Verify Dose
• Copy plan to phantom, recalculate, deliver to
chamber
• Chamber measurements ≤ 3% from planned dose
• Array devices and film can be calibrated to dose
Patient Specific QA
• Verify Distribution
• Array devices (MapCheck, ArcCheck, Matrixx,
Octavius, Delta4, etc.)
• Film
• Gel?
• Use Record/Verify “QA Mode” deliver at true
gantry angles.
• Analyse beams individually and as whole
fraction.
Patient-Specific QA (Pre-Tx)
• Using the Delta4 phantom we get psuedo-3D distribution of
points across the plan volume
• Two 2D planes of diodes form a cross
• Real plan > copy to phantom CT, recalc > measure > analyse
• Results are highly reproducible
Delta4 Results
Delta4 Results
• Halo distribution
• TPS pumping dose in the
non-lateral-equilibrium
regions
• Absolute dose max
~200% patient
prescription
• Difference of dose
absorption between high
and low density mediums
Delta4 Results
• Very similar results when measurements are repeated on same
day and different day  reproducible delivery by MLC
• Very similar results when measurements are repeated on
different linacs  well matched and stable linacs
• Where to set tolerance for pass/fail?
Avg γ Pass
3mm DTA
2mm DTA
1mm DTA
Dose Diff 3%
100.0%
99.5%
93.1%
Dose Diff 2%
99.5%
96.0%
88.9%
Dose Diff 1%
97.9%
95.5%
77.5%
More QA Equipment
Tomas Kron, Peter MacCallum Cancer Centre
Patient-Specific QA (Post-Tx)
• Phantom measurements check one delivery,
one time.
• Linac log files can be used to check actual
treatment delivery mechanical parameters
• Combine this with IGRT and dose
reconstruction/accumulation is possible
Patient-Specific QA (Post-Tx)
• Elekta does not have dynalogs
• But a record of mechanical parameters is sent to Mosaiq after delivery
• A report can be generated and compared to the DICOM-RTPlan
In vivo Dosimetry
• TLD
• OSLD
• Diodes
• MOSFETS
• Radiochromic film squares
• “Ex vivo” Dosimetry
• Transit Dosimetry via EPID
• Per fraction beam fluence measurements
• Recommend checking in field and out of field
45
In vivo Dosimetry
46
PROCESS REVIEW
Process Evaluation
Process Evaluation
• MARGINPTV = 2.5Σ + 0.7σ
• Σ – st dev of sys errors
• σ – st dev of random errors
• 2.5 and 0.7 come from 90% and 95% confidence intervals for
Gaussian distributions, respectively.
• This margin has the 95% isodose line cover the CTV in 90% of
patients
• Systematic errors contribute more than random errors to
uncertainty
• 4DCT and IGRT should remove systematic error and reduce
random error
Process Evaluation
Van Herk 2012
Process Evaluation
For a single patient:
• Systematic Error =
mean offset
• Random Error =
standard deviation
Chris Fox, Peter MacCallum
Process Evaluation
Little Sigma
Big Sigma
For a population of
patients:
• Systematic Error =
standard deviation of
individual mean errors
• Random Error =
Root-Mean-Sum of
individual random
errors
Chris Fox, Peter MacCallum
52
Process Evaluation
• You can only collect statistics on what you
image.
• If you want to know how accurate your IGRT is,
you need another image after any couch shift
53
THANK YOU
Tomas Kron
Simon Downes
Sean White
54

similar documents