Author: Youssef Ben Bouchta, Chen Cheng, Owen Thomas Dillon, Mark Gardner, Paul J. Keall, Purnima Sundaresan 👨🔬
Affiliation: Radiation Oncology Network, Western Sydney Local Health District, Image X Institute, Faculty of Medicine and Health, The University of Sydney 🌍
Purpose: There are three clinical motivations for real time IGRT in head and neck cancer radiation therapy: (1)50% of patients experience anxiety from the patient mask (2)patient motion still occurs despite the mask and (3)real time IGRT in other sites such as prostate have demonstrated 50% reduction in toxicities. For these reasons, we developed a real-time volumetric imaging method suitable for head and neck treatment on conventional linacs using a Kalman Filter (KF) approach.
Methods: The KF approach uses a novel reformulation of forward projection for vector fields to estimate the underlying 3D motion field using observed motion in 2D projection images recovered via 2D Demons x-ray image registration. The KF allows us to incorporate incomplete observation data within local smoothness assumptions on the underlying field. We simulated kV acquisition at 1Hz/6 degrees using ground truth 3D volumes, where the ground truth deformation at each time point was simulated using surface tracking traces of 12 head and neck patients. The posterior 20x20x20 masked grid of displacement vectors was interpolated and applied to the reference 3D CT to obtain volumetric images. Anatomical accuracy was evaluated via average symmetric surface distance (ASSD) and mean centroid error (MCE) metrics at each time point for relevant structures. The findings were compared with the standard of care, no volumetric imaging.
Results: For the 12 patients the tumor contour ASSD was 0.9±0.3mm (1.2±0.6mm no imaging). The tumor MCE was -0.2±0.6mm, 0.3±0.8mm, 0.1±0.4mm in LR, AP, SI directions (0.9±1.0mm, 0.8±1.1mm, -0.1±1.2mm with no imaging). The ASSD for the brainstem was 0.9±0.3mm (1.3±0.6mm no imaging). Slow baseline drift was captured more readily compared to high-frequency motion.
Conclusion: We have performed proof-of-concept real-time volumetric imaging via the simulation of realistic internal head and neck motion. Preliminary results demonstrate accurate and simultaneous tracking of both tumour and nearby organ-at-risk.