Program Information
Temporal Latency Characterization of An In-House Gating System and a Compensation Strategy for Gated SBRT
B Barraclough*, J Park , F Li , B Lu , J Li , C Liu , G Yan , University of Florida, Gainesville, FL
Presentations
MO-DE-FS1-10 (Monday, July 31, 2017) 1:45 PM - 3:45 PM Room: Four Seasons 1
Purpose: To characterize the temporal latency of an in-house gating system and to validate an artificial neural network-based solution to compensate for the latency for gated SBRT.
Methods: The gating system consists of an infrared-based optical tracking system (OTS) and the Elekta Response module. The overall system latency includes the OTS response time and the linac beam on/off delay in gating mode. The latter was measured with a 2D diode array (50 ms temporal resolution). The overall latency was measured with a 4D motion phantom in two ways (1) imaging a moving wire attached to the 4D phantom (with known velocity) with EPID and tracking the distance it over-traveled (2) irradiating a moving film attached to the 4D phantom and examining the profile shape change due to delay. To compensate for the system latency, two adaptive neural networks (ANNs) were trained to predict patient breathing pattern: (1) an adaptive ANN with tapped delay line that used only the OTS signal as input (2) a nonlinear autoregressive exogenous model (NARX) that used both the OTS signal and the diaphragm positions delineated from CBCT projections. In NARX, only the OTS signal was used for prediction since diaphragm positions were not available post-CBCT. We evaluated the prediction accuracy of both ANNs at the boundary of the gating window (gating threshold).
Results: The overall system latency was ~780 ms; for motion prediction, the NARX achieved sufficient accuracy in training; but the prediction performance was worse than the adaptive ANN that accurately predicted the motion by ~540 ms.
Conclusion: The OTS response time dominated the linac beam on/off delay. An adaptive neural network with tapped delay line outperforms a NARX. Such a neural network needs to be in place to ensure the success of free-breathing gated radiotherapy.
Contact Email: