An online x-ray based position validation system for prostate hypofractionated radiotherapy
Purpose: Accurate positioning of the target volume during treatment is paramount for stereotactic body radiation therapy (SBRT). In this work, the authors present the development of an in-house software tool to verify target position with an Elekta-Synergy linear accelerator using kV planar images acquired during treatment delivery. Methods: In-house software, SeedTracker, was developed in MATLAB to perform the following three functions: 1. predict intended seed positions in a planar view perpendicular to any gantry angle, simulating a portal imaging device, from the 3D seed co-ordinates derived from the treatment planning system; 2. autosegment seed positions in kV planar images; and 3. report the position shift based on the seed positions in the projection images. The performance of SeedTracker was verified using a CIRS humanoid phantom (CIRS, VA, USA) implanted with three Civco gold seed markers (Civco, IA, USA) in the prostate. The true positive rate of autosegmentation (TPRseg) and the accuracy of the software in alerting the user when the isocenter position was outside the tolerance (TPRtrig) were studied. Two-dimensional and 3D static position offsets introduced to the humanoid phantom and 3D dynamic offsets introduced to a gel phantom containing gold seeds were used for evaluation of the system. Results: SeedTracker showed a TPRseg of 100% in the humanoid phantom for projection images acquired at all angles except in the ranges of 80◦-100◦ and 260◦-280◦ where seeds are obscured by anatomy. This resulted in a TPRtrig of 88% over the entire treatment range for considered 3D static offsets introduced to the phantom. For 2D static offsets where the position offsets were only introduced in the anterior-posterior and lateral directions, the TPRtrig of SeedTracker was limited by both seed detectability and positional offset. SeedTracker showed a false positive trigger in the projection angle range between 130◦-170◦ and 310◦-350◦ (a maximum of 24% of treatment time) due to limited information that can be derived from monoscopic images. The system accurately determined the dynamic trajectory of the isocenter position in the superior and inferior direction for the studied dynamic offset scenarios based on the seed position in monoscopic images. Conclusions: The developed software has been shown to accurately autosegment the seed positions in kV planar images except for two 20◦ arcs where seeds are obscured by anatomical structures. The isocenter trajectories determined by the system, based on the monoscopic images, provide useful information for monitoring the prostate position. The developed system has potential application for monitoring prostate position during treatment delivery in linear accelerator based SBRT.