Enhancing Linear Accelerator Performance through Predictive Modeling of Downtime and Interlock Incidents
Abstract
Purpose
Radiation therapy is a critical modality in cancer treatment, relying on medical Linear Accelerators (LINACs) to deliver precise radiation doses while minimizing exposure to surrounding healthy tissues. LINAC systems generate extensive operational and administrative data that are managed by integrated control software. Safety interlocks are designed to automatically interrupt treatment when operational parameters exceed predefined thresholds. Component failures and interlock events can lead to unplanned machine downtime, disrupting clinical workflows, delaying patient treatments, and increasing operational costs. This study aims to develop predictive machine learning models for LINAC interlock occurrences and downtime events to enhance system reliability and reduce service interruptions.
Methods
Operational data were collected from the machine malfunction logbook of a Varian Clinical System at Square Hospitals Ltd. Two datasets were constructed: one for predicting interlock events and another for forecasting downtime duration. An LSTM (Long Short-Term Memory) neural network was implemented for interlock prediction and trained over 300 epochs. For downtime prediction, six machine learning models were developed and evaluated. Model performance was assessed using accuracy metrics and mean squared error (MSE), with k-fold cross-validation applied for model validation.
Results
The LSTM model achieved an accuracy of 72% in predicting interlock events, with a top-2 prediction accuracy of 84%. Among the evaluated downtime prediction models, Bayesian regression demonstrated the best performance, achieving an accuracy of 98.21% and an MSE of 0.0179. Cross-validation results confirmed the robustness and consistency of the predictive models.
Conclusion
Machine learning can significantly enhance the prediction and analysis of LINAC interlock and downtime events, enabling proactive maintenance strategies. These findings support improved patient care, workflow efficiency, and hospital management, ensuring minimal service disruptions in radiation therapy.