Purpose: During radiation therapy, respiratory motion may lead to inter- and intra-fraction variations in patient anatomy that result in discrepancies between the planned and the delivered dose. Continuous knowledge of a patient's three-dimensional anatomy is required to fully assess the dosimetric impact of this motion, and would enable treatment adaptation to account for any deviations from the planned therapy. However, current clinical imaging techniques on standard linear accelerators do not provide sufficient information to fully reconstruct 3D anatomy in real time. We present a technique for generating continuous volumetric images during radiotherapy treatments using periodic planar kV images and an external respiratory surrogate signal.
Methods: Using the on-board imager, kV radiographs are acquired every 3 seconds during therapy and used to fit a principal component motion model. A multi-dimensional correlation relationship is then established between the model and the external surrogate signal, enabling continuous volumetric image reconstruction between kV imaging time points. Performance of the algorithm was evaluated using 10 digital eXtended CArdiac-Torso (XCAT) phantom datasets, enabling a comparison against a full 3D ground truth. The phantoms were programmed with measured 3D tumor motion traces and the measured external surrogate trace was used, enabling the method to be assessed with realistic irregularities breathing pattern and changes in internal-external correlation.
Results: The three-dimensional tumor positions are reconstructed with an average root mean square error of 1.47 mm, and an average 95th percentile of 2.80 mm.
Conclusion: Through combining kV imaging and external surrogate information, this technique enables continuous 3D image generation on current clinical accelerators using image-derived information without the additional dose of continuous kV imaging. It provides a new capability for treatment verification and calculating delivered dose in the presence of irregular respiratory motion. Application of this technique to clinically acquired patient datasets is ongoing.
Funding Support, Disclosures, and Conflict of Interest: This project was supported through a Master Research Agreement with Varian Medical Systems.