automated qa: is this the future? - ucsf cme...outline •qa in radiation oncology. •how to judge...
TRANSCRIPT
Automated QA:Is this the Future?
Olivier Morin, PhDUniversity of California San Francisco
UCSF Annual CourseMarch 23rd, 2014
Sunday, March 23, 2014
Outline
• QA in Radiation Oncology.
• How to judge the quality of a QA program?
• Automation -> road to improvements?
• Machine learning in RT. (Smart Plan Import)
• Linac QA using EPID.
• QA trend for hyprofractionation (daily Linac QA).
Sunday, March 23, 2014
Outline
• QA in Radiation Oncology.
• How to judge the quality of a QA program?
• Automation -> road to improvements?
• Machine learning in RT. (Smart Plan Import)
• Linac QA using EPID.
• QA trend for hyprofractionation (daily Linac QA).
Sunday, March 23, 2014
Outline
• QA in Radiation Oncology.
• How to judge the quality of a QA program?
• Automation road to improvements?
• Machine learning in RT. (Smart Plan Import)
• Linac QA using EPID.
• QA trend for hyprofractionation (daily Linac QA).
Sunday, March 23, 2014
Outline
• QA in Radiation Oncology.
• How to judge the quality of a QA program?
• Automation road to improvements?
• Machine learning in RT. (Smart Plan Import)
• Linac QA using EPID.
• QA trend for hyprofractionation (daily Linac QA).
Sunday, March 23, 2014
Outline
• QA in Radiation Oncology.
• How to judge the quality of a QA program?
• Automation road to improvements?
• Machine learning in RT. (Smart Plan Import)
• Linac QA using EPID.
• QA trend for hyprofractionation (daily Linac QA).
Sunday, March 23, 2014
Outline
• QA in Radiation Oncology.
• How to judge the quality of a QA program?
• Automation road to improvements?
• Machine learning in RT. (Smart Plan Import)
• Linac QA using EPID.
• QA trend for hyprofractionation (daily Linac QA).
Sunday, March 23, 2014
QA in RT
Physicians
Physicists
Therapists
Nurses
Dosimetrists
Administration
Medical residents
Physics residentsQA
IT
Sunday, March 23, 2014
QA is a multi-layered series of
cross-checks done by ALL
professionals involved.
QA in RT
Physicians
Physicists
Therapists
Nurses
Dosimetrists
Administration
Medical residents
Physics residents
IT
Sunday, March 23, 2014
Quality of QA program?
• QA in RT good to prevent severe errors.• According to ASTRO estimates, preventable mistakes occur in less than 0.01%
of the time (NY 2011).
• In contrast, the rate of occurrence of preventable events is 3X higher for surgery (http://www.hopkinsmedicine.org/surgery/faculty/Makary).
• Most of the treatment plan checks are performed by individuals using checklists.
• Evaluation of tests results too reliant and staff expertise.
• IMRT and VMAT delivery QA labor intensive and limited.
Good at preventing the preventables?
Room for improvement:
Sunday, March 23, 2014
Quality of QA program?
• QA in RT good to prevent severe errors.• According to ASTRO estimates, preventable mistakes occur in less than 0.01%
of the time (NY 2011).
• In contrast, the rate of occurrence of preventable events is 3X higher for surgery (http://www.hopkinsmedicine.org/surgery/faculty/Makary).
• Most of the treatment plan checks are performed by individuals using checklists.
• Evaluation of tests is too reliant on staff expertise.
• IMRT and VMAT delivery QA labor intensive and limited.
Good at preventing the preventables?
Room for improvement:
Sunday, March 23, 2014
Building a New Approach
From Discovery to Design: The Evolution of Human Factors in Healthcarehttp://www.longwoods.com/content/22900
• Human factor engineering for safe implementation of new technology.
• People focused approaches tend to be less effective.
• System focused and automation would be more effective.
Sunday, March 23, 2014
Building a New Approach
From Discovery to Design: The Evolution of Human Factors in Healthcarehttp://www.longwoods.com/content/22900
• Human factor engineering for safe implementation of new technology.
• People focused approaches tend to be less effective.
• System focused and automation would be more effective.
Sunday, March 23, 2014
Building a New Approach
From Discovery to Design: The Evolution of Human Factors in Healthcarehttp://www.longwoods.com/content/22900
• Human factor engineering for safe implementation of new technology.
• People focused approaches tend to be less effective.
• System focused and automation would be more effective.
Sunday, March 23, 2014
Machine Learning in RT
• ML algorithm to determine if tests or series of parameters are considered normal.
• Algorithm must be trained with a set of normal tests or parameters.
• Multi-parameter algorithm (prescription, MU, gantry angles, modulation,etc.).
• Algorithm playing a bigger role in RT.
Sunday, March 23, 2014
Machine Learning in RT
• ML algorithm to determine if tests or series of parameters are considered normal.
• Algorithm must be trained with a set of normal tests or parameters.
• Multi-parameter algorithm (prescription, MU, gantry angles, modulation,etc.).
• Algorithm playing a bigger role in RT.
Sunday, March 23, 2014
Machine Learning in RT
• ML algorithm to determine if tests or series of parameters are considered normal.
• Algorithm must be trained with a set of normal tests or parameters.
• Multi-parameter algorithm (prescription, MU, gantry angles, modulation,etc.).
• Algorithm playing a bigger role in RT.
Sunday, March 23, 2014
Machine Learning in RT
• ML algorithm to determine if tests or series of parameters are considered normal.
• Algorithm must be trained with a set of normal tests or parameters.
• Multi-parameter algorithm (prescription, MU, gantry angles, modulation,etc.).
• Algorithm could play an important role in RT.
Sunday, March 23, 2014
Smart Plan Import
• Train (using ML) patient management system (MOSAIQ or Aria) to detect when parameters are outside the norm.
• Smart Plan Import use patient information and prescription and treatment technique to verify key parameters.
• Smart plan could detect when imaging reference information is not adequate.
• Concept of virtual machine could be used.
Sunday, March 23, 2014
Smart Plan Import
• Train (using ML) patient management system (MOSAIQ or Aria) to detect when parameters are outside the norm.
• Smart Plan Import uses past experiences along with patient information, prescription and treatment technique to verify key parameters.
• Smart plan could detect when imaging reference information is not adequate.
• Concept of virtual machine could be used.
Sunday, March 23, 2014
Smart Plan Import
• Train (using ML) patient management system (MOSAIQ or Aria) to detect when parameters are outside the norm.
• Smart Plan Import uses past experiences along with patient information, prescription and treatment technique to verify key parameters.
• Smart plan could detect when imaging reference information is not adequate.
• Concept of virtual machine could be used.
Sunday, March 23, 2014
Smart Plan Import
• Train (using ML) patient management system (MOSAIQ or Aria) to detect when parameters are outside the norm.
• Smart Plan Import uses past experiences along with patient information, prescription and treatment technique to verify key parameters.
• Smart plan could detect when imaging reference information is not adequate.
• Concept of virtual machine could be used.
Sunday, March 23, 2014
Linac QA• AAPM TG-142 provide
recommendations to assure quality of treatments.
• UCSF QA procedures take 30 mins daily + 6 hours monthly.
• IMRT and VMAT is commonly performed on all new plans prior to Tx.
• With automation and 2D detectors more and better verifications could be done.
Sunday, March 23, 2014
Linac QA• AAPM TG-142 provide
recommendations to assure quality of treatments.
• UCSF QA procedures take 30 mins daily + 6 hours monthly.
• IMRT and VMAT is commonly performed on all new plans prior to Tx.
• With automation and 2D detectors more and better verifications could be done.
Sunday, March 23, 2014
Linac QA• AAPM TG-142 provide
recommendations to assure quality of treatments.
• UCSF QA procedures take 30 mins daily + 6 hours monthly.
• IMRT and VMAT is commonly performed on all new plans prior to Tx.
• With automation and 2D detectors more and better verifications could be done.
Sunday, March 23, 2014
Linac QA• AAPM TG-142 provide
recommendations to assure quality of treatments.
• UCSF QA procedures take 30 mins daily + 6 hours monthly.
• IMRT and VMAT is commonly performed on all new plans prior to Tx.
• With automation and 2D detectors more and better verifications could be done.
Sunday, March 23, 2014
TG-142 / Daily QA
Sunday, March 23, 2014
Make Daily QA Part of the Clinic
Sunday, March 23, 2014
TG-142 / Monthly QA
Sunday, March 23, 2014
Example of Monthly QA
Sunday, March 23, 2014
Systematic & Complete Docs
Sunday, March 23, 2014
Is the Current Daily QA Adequate for Hypofractionation?
• High dose per fraction.
• Heavily modulated treatments.
• Absolutely needs patient specific QA at the moment.
• 2D detectors and automation could open the door to a more complete daily QA.
Sunday, March 23, 2014
Proposed Automated Daily QA
Therapistarrives 30
mins before Tx
Deliver daily QA sequence
(25 mins)
EPID Images analyzed by software
Sunday, March 23, 2014
Daily Machine Verification
Sunday, March 23, 2014
Virtual WedgesVirtual Wedges
Sunday, March 23, 2014
Flatness and Symmetry
Sunday, March 23, 2014
MLC SequencePatient specific QA vs. same daily MLC sequence
vs.
Sunday, March 23, 2014
In Summary
• Machine learning algorithms could play a significant role in the day-to-day verifications.
• Patient-specific QA is labor intensive and effort is under way to obtain more by doing less.
• A trend towards hypofractionation may also call for revised daily QA procedure.
• Automation can save time and detect machine problems (i.e. MLC) before they lead to errors or down time.
Sunday, March 23, 2014
In Summary
• Machine learning algorithms could play a significant role in the day-to-day verifications.
• Patient-specific QA is labor intensive and efforts are under way to obtain more by doing less.
• A trend towards hypofractionation may also call for revised daily QA procedure.
• Automation can save time and detect machine problems (i.e. MLC) before they lead to errors or down time.
Sunday, March 23, 2014
In Summary
• Machine learning algorithms could play a significant role in the day-to-day verifications.
• Patient-specific QA is labor intensive and efforts are under way to obtain more by doing less.
• A trend towards hypofractionation may also call for a more comprehensive daily QA procedure.
• Automation can save time and detect machine problems (i.e. MLC) before they lead to errors or down time.
Sunday, March 23, 2014
In Summary
• Machine learning algorithms could play a significant role in the day-to-day verifications.
• Patient-specific QA is labor intensive and efforts are under way to obtain more by doing less.
• A trend towards hypofractionation may also call for a more comprehensive daily QA procedure.
• Automation can save time and detect machine problems (i.e. MLC) before they lead to errors or down time.
Sunday, March 23, 2014
Automated QA:Is this the Future?
Olivier Morin, PhDUniversity of California San Francisco
UCSF Annual CourseMarch 23rd, 2014
Sunday, March 23, 2014