A Predictive Irradiation Time Calculator for Pencil Beam Scanning Proton Therapy
Abstract
Purpose
In proton therapy treatment planning, several free parameters impact both plan quality and delivery time. These include spot and energy layer spacing, MU per spot or layer, and in some systems, utilization of adaptive apertures (AA). Decreasing delivery time while maintaining plan quality maximizes patient comfort and clinical efficiency. To better understand the impact of treatment planning approaches, an irradiation time calculator was developed based on delivery log files.
Methods
Machine logs from a single irradiation for 32 subjects over 12 months were acquired from two matched clinical synchrocyclotron treatment vaults, totaling 36 plans and 116 fields. Disease sites included breast, central nervous system, craniospinal axis, esophagus, head and neck, pelvis, spine, and thorax. 17 plans utilized dynamic apertures and 19 utilized static apertures. The timestamp, energy, spot index, spot position, charge, and AA coordinates were fit using linear-least squares, fitting coefficients for AA movement, spot position and charge, and setting energy layer switch time as constant. The plans were divided into training and testing (14) and validation (22) plans. The model was evaluated by correlation between actual and predicted delivery time.
Results
Predicted delivery times had an absolute mean regression error of 7.47±5.40 seconds in the training set and an absolute mean residual error of 5.53±4.01 seconds in the validation set. Breast plans had the lowest absolute regression error of 3.68±2.64 seconds, and pelvis had the highest absolute regression error of 15.64±9.73 seconds. Model accuracy was independent of dynamic aperture usage during planning. The model predicts an AA movement of 0.09 meters/second, scanning target speed of 1.28 meters/second, charge delivery of 144.44 picocoulombs/second, and an energy layer switch time of 242.07 milliseconds.
Conclusion
This tool can be used to predict total delivery times and allow for exploration on the impact of treatment planning parameters on delivery time.