Region-of-interest intra-arc MV imaging to facilitate sub-mm positional accuracy with minimal imaging dose during treatment deliveries of small cranial lesions.

Abstract

To automate the generation of region-of-interest (ROI) apertures for use with megavoltage imaging for online positional corrections during cranial stereotactic radiosurgery. Digitally reconstructed radiographs (DRRs) were created for a 3D-printed skull phantom at 5degree gantry angle increments for a three-arc beam arrangement. At each angle, 3000 random rectangular apertures were generated, and 100 shifts on a grid were applied to the anatomy within the frame. For all shifts, the mutual information (MI) between the shifted and unshifted DRR was calculated to derive an average MI gradient. The top 10% of apertures that minimized registration errors were overlaid and discretely thresholded to generate imaging plans. Imaging was acquired with the skull while implementing simulated patient motion on a linac. Control point-specific couch motions were derived to align the skull to its planned positioning. Apertures with a range of repositioning errors less than 0.1mm possessed a 42% larger average MI gradient when compared with apertures with a range greater than 1mm. Dose calculations with Monte Carlo exhibited an 84% reduction in the dose received by 50% of the skull with the 50% thresholded plan when compared to a constant 22×22cm2 imaging plan. For all different imaging plans (with and without motion), the calculated median 3D-errors with respect to the tracking of a metal-BB fiducial positioned at isocenter in the skull were sub-mm except for the 80% thresholded plan. Sub-mm positional errors are achievable with couch motions derived from control point-specific ROI imaging. Smaller apertures that conform to an anatomical ROI can be utilized to minimize the imaging dose incurred at the expense of larger errors.