Predictive Maintenance of Medical Accelerators A Customized Dashboard

Predictive Maintenance of Medical Accelerators: A Customized Dashboard Interface CM Able, MS, AH Baydush, Ph. D. , C Nguyen, BS, J Gersh, Ph. D. , A Ndlovu, Ph. D. , I Rebo, Ph. D. , J Booth. Ph. D. , M. Perez, Ph. D. , B Sintay, Ph. D. , and MT Munley, Ph. D. December 3, 2014 RSNA Meeting, Chicago, IL

Disclosures: This work was supported by a grant from Varian Medical Systems, Inc.

Accelerator Predictive Maintenance: Statistical Process Control (SPC) Machine operational and performance data can be represented by a continuous probability density function – typically a Gaussian distribution. § Statistical Process Control (SPC) techniques detect nonrandom changes in a process or system § Process Control Charts are the primary tool of SPC § Individual – individual parameter versus time § Moving Range – difference between adjacent individual data points versus time § Standard SPC methods utilize the mean and standard deviation of the process or system being monitored to detect non-random changes, i. e. potential failures

VMAT QA Daily Delivery § VMAT delivery of a robust, technically challenging QA treatment (Snooker Cue*) § Maximizes gantry and MLC speed and speed changes, both with and against gravity § Automated and performed daily § 5 sites consisting of 7 Truebeam linacs currently under investigation § Trajectory and text log files are decoded analyzed * Van. Esch et al, Med Phys 38(9), 5146 -5166 (2011).

Parameters Analyzed § 525 total parameters analyzed § 35 from text log file § 490 from trajectory log file § 482 of which are MLC/carriage related § 40 are raw measured values § 485 are software processed values § Speed - distance change over time § Cross correlation § With small, challenging subsection of baseline run § Analyze location and peak cross correlation value

Pd. M Software: Initial Launch

Pd. M Software: Machine Selection

Pd. M Software: 9 Major Sub-systems

Pd. M Software: RF Generation 5 Parameters

Pd. M Software: Gun Control 5 Parameters

Pd. M Software: Energy Control 4 Parameters

Pd. M Software: Beam Uniformity 11 Parameters

Pd. M Software: DC Voltage Generation 7 Parameters

Pd. M Software: System Regulation 3 Parameters

Pd. M Software: Jaws and Gantry 10 Parameters: 6 measured, 4 calculated

Pd. M Software: MLC Bank (A or B) 4 Parameters per leaf: all calculated

Pd. M Software: Analysis Pane

Individual and Moving Range Charts • Upper – Individual • Lower – MR • White line – Baseline Mean • Red Lines – Upper and lower control limits • Orange data – warning condition • Red data – Alarm condition

Horizontal Histogram and Parameter Summary Statistics • Upper – Horizontal frequency histogram • White line – Baseline Grand Mean • Red Lines – Upper and lower control limits • Middle – Parameter summary statistics • Lower – Saved user comments

Hybrid Upper and Lower Control Limits Blue Line – Standard SPC limits (+/-3 SD) using measured data Green Line – Limits using system specifications (MLC +/-2 mm/s) Red Line – Hybrid method combining standard SPC method an empirical factor based on clinical requirements or global variation determined from a large sample of accelerators

Pd. M Software: Example Detection – Beam Steering • Radial Angle Steering Coil Current values deviated to an alarm state • Deviation was identified by morning flatness/symmetry checks • Change in symmetry was confirmed by scanning water phantom • Adjustments were made and values were back at baseline

Pd. M Software: MLC Motor Failure Prediction? Cross correlation (CC) of baseline MLC position (location versus time during delivery) and its position during each daily delivery • This chart shows the CC maximum value is decreasing • Decreasing CC max value indicates the MLC is lagging compared to its baseline performance • MLC performance is evaluated by speed, CC maximum value and CC maximum value location

Predictive Maintenance: Next Steps We have deployed an alpha version of the True. Beam Pd. M software to our data collection partners. We are now correlating our findings with field service reports in an effort to validate the utility and effectiveness of our technique for predictive ability. Thank You!