Room: Karl Dean Ballroom B1
Purpose: Monitoring breathing motion is a central consideration in precision radiotherapy. In this work, we explore the use of ultra-wide band (UWB) radar for monitoring and detecting changes in simulated breathing motion during dynamic IMRT delivery.
Methods: Breathing motion patterns were simulated using a respiratory motion phantom positioned on the treatment couch of a c-arm linear accelerator. The radar signals from the phantom were measured during VMAT treatment delivery using a single UWB radar system. Radar signals attributed to the phantom and linac were binned according to their respective distances from the radar. The bins corresponding to phantom were also identified by capturing the signal from phantom motion in isolation. Phantom motion signals measured in the presence of high-noise background from linac motion were recovered by 1) estimating the noise power using an artificial neural network (ANN), 2) enhancing the background-subtracted signal using range-bin weighted averaging over the target range-bins, and 3) applying a filter to remove residual noise.
Results: Phantom motion radar signals recovered from high-noise background are in excellent agreement with reference data recorded by the respiratory motion phantom. The algorithm successfully recovered simple sinusoidal patterns as well as complex patterns mimicking deep inspiration breath hold treatments during VMAT treatment delivery. Changes in breathing patterns were detected during VMAT delivery with a total processing delay time of three seconds or less on a 2.16 GHz processor. The ANN used to estimate noise from linac motion did not require re-training in order to detect noise that differed from the training reference.
Conclusion: Ultra-wide band radar can be used to monitor and detect simulated breathing motion in the presence of a high-noise background associated with VMAT treatment delivery. The signal recovery algorithm is robust against a range of breathing patterns and linac motion noise variations.
Not Applicable / None Entered.