Click here to

Session: External Beam Modeling [Return to Session]

Dependence of Subgroup Size On Predictive Quality Assurance for Beam Asymmetry Using Statistical Process Control and Autoregressive Integrated Moving Average Modeling

J Li, M Chan*, D Wang, Memorial Sloan Kettering Cancer Center, Basking Ridge, NJ


TU-F115-IePD-F3-5 (Tuesday, 7/12/2022) 1:15 PM - 1:45 PM [Eastern Time (GMT-4)]

Exhibit Hall | Forum 3

Purpose: AutoRegressive Integrated Moving Average (ARIMA) is a time-series forecasting algorithm that can be used to estimate future values based on its past pattern. It has been evaluated for predictive quality assurance in radiotherapy. The accuracy of modeling depends on the organization of time series from which model parameters are obtained. In this study, we investigated the relationship between the size of subgroup data and the model’s accuracy in predicting events for out-of-control-limit in beam symmetry from Linac daily QA measurements.

Methods: A Varian Trilogy in our clinic exceeded tolerances in beam symmetry due to a problem in its water-cooling system. Retrospectively, we collected daily QA data for 120 consecutive days and analyzed its historical trend using methods of Statistical Process Control (SPC) and ARIMA: data were grouped in three time-series: daily, 3-day average, and 5-day average. For each series, corresponding SPC upper and lower control limits were calculated using the first 100 days’ data. Three different ARIMA models are generated for three time-series separately; their parameters were obtained by training these models with the first 100 days’ subgroup data. Model results were assessed for their abilities to predict the last 20 days’ performance and out of SPC limits before it happened.

Results: Evaluations of models’ performance using the comparison between residues’ distributions and standard normal distributions demonstrated good fitness to the data in three established ARIMA models. Models’ forecast results for the last 20 days showed models for daily and 3-day average failed to predict the event of exceeding SPC limits, however, 5-day average model was able to predict such events from past data. It would give a 10-day advanced warning before out-of-control-limit event occurs.

Conclusion: ARIMA is a feasible algorithm in predictive Linac quality assurance. Combining with SPC, it can forecast anomalies and alert failures in beam symmetry.


Not Applicable / None Entered.


Not Applicable / None Entered.

Contact Email