Encrypted login | home

Program Information

Predicting Accelerator Dysfunction


C Able

C Able1*, C Nguyen2 , A Baydush3 , J Gersh4 , A Ndlovu5 , I Rebo6 , J Booth7 , M Perez8 , B Sintay9 , M Munley10 , (1) Florida Cancer Specialists - New Port Richey, New Port Richey, FL, (2) Wake Forest School of Medicine, Winston-salem, NC, (3) Wake Forest University School of Medicine, Winston-salem, NC, (4) Gibbs Cancer Center & Research Institute - Pelham, Greer, SC, (5) Hackensack University Medical Center, Hackensack, NJ, (6) Hackensack University Medical Center, Hackensack, NJ, (7) Royal North Shore Hospital, St Leonards, ,(8) North Sydney Cancer Center, Sydney, Australia, (9) Cone Health Cancer Center, Greensboro, NC, (10) Wake Forest University School of Medicine, Winston-salem, NC

Presentations

TU-FG-201-9 (Tuesday, August 2, 2016) 1:45 PM - 3:45 PM Room: 201


Purpose:
To develop an integrated statistical process control (SPC) framework using digital performance and component data accumulated within the accelerator system that can detect dysfunction prior to unscheduled downtime.

Methods:
Seven digital accelerators were monitored for twelve to 18 months. The accelerators were operated in a ‘run to failure mode’ with the individual institutions determining when service would be initiated. Institutions were required to submit detailed service reports. Trajectory and text log files resulting from a robust daily VMAT QA delivery were decoded and evaluated using Individual and Moving Range (I/MR) control charts. The SPC evaluation was presented in a customized dashboard interface that allows the user to review 525 monitored parameters (480 MLC parameters). Chart limits were calculated using a hybrid technique that includes the standard SPC 3σ limits and an empirical factor based on the parameter/system specification. The individual (I) grand mean values and control limit ranges of the I/MR charts of all accelerators were compared using statistical (ranked analysis of variance (ANOVA)) and graphical analyses to determine consistency of operating parameters.

Results:
When an alarm or warning was directly connected to field service, process control charts predicted dysfunction consistently on beam generation related parameters (BGP)– RF Driver Voltage, Gun Grid Voltage, and Forward Power (W); beam uniformity parameters – angle and position steering coil currents; and Gantry position accuracy parameter: cross correlation max-value. Control charts for individual MLC – cross correlation max-value/position detected 50% to 60% of MLCs serviced prior to dysfunction or failure. In general, non-random changes were detected 5 to 80 days prior to a service intervention. The ANOVA comparison of BGP determined that each accelerator parameter operated at a distinct value.

Conclusion:
The SPC framework shows promise. Long term monitoring coordinated with service will be required to definitively determine the effectiveness of the model.

Funding Support, Disclosures, and Conflict of Interest: Varian Medical System, Inc provided funding in support of the research presented.


Contact Email: