Author: Leon Dunn 👨🔬
Affiliation: IsoAnalytics Pty. Ltd. 🌍
Purpose: The purpose of this work is to present the development and results of an automated Varian Trajectory Log file processing software platform called Sentinel (IsoAnalytics Pty. Ltd.).
Methods: Sentinel was developed using MATLAB (R2022a) and features a graphical user interface which monitors two folders and automatically processes trajectory log files from up to 50 linear accelerators simultaneously, as well as simultaneously evaluating incoming plans from the TPS. The software provides automated analytics dashboards focusing on machine performance, patient QA, and treatment plan analytics. It uses the high-temporal-resolution MLC positions, as well as jaw/gantry/collimator positions recorded in log files to create fluence maps for gamma-analysis, by comparing planned to expected fluence for each field automatically. Over 100k fluence images and pass/fail outcome generated by Sentinel were then used to train a convolutional neural network to predict whether a particular plan will pass or fail the analysis when delivered and can notify the user via email when plans or log files fail analysis.
Results: To date, Sentinel has automatically and passively processed over 100,000 log files from 16 linear accelerators including TrueBeam, Edge and Halcyon models and provides a wealth of information on machine performance and statistics relative to the fleet of machines monitored. The results from each log-file and plan-check are stored in a database and have highlighted underperforming, ageing machines and plan-classes that challenge machine delivery.
Conclusion: Automated log file analysis with Sentinel provides a unique passive QA layer that enhances established plan and machine QA programs. By offering unique insights into machine performance and plan complexity, Sentinel can help medical physicists make data-driven decisions. The AI predictions built on Sentinel data have the potential to identify plans that may fail even before delivery.