A hardware-accelerated software platform for adaptive radiation therapy

Abstract

A critical requirement of successful adaptive radiotherapy (ART) is the knowledge of anatomical changes as well as actual dose delivered to the patient during the course of treatment. While cone-beam CT (CBCT) is typically used to minimize the patient setup error and monitor daily anatomical changes, its poor image quality impedes accurate segmentation of the target structures and the dose computation. We developed an integrated ART software platform that combines fast and accurate image registration, segmentation, and dose computation/ accumulation methods. The developed platform automatically links patient images, radiotherapy plan, beam and dosimetric parameters, and daily treatment information, thus providing and efficient ART workflow. Furthermore, to improve the accuracy of deformable image registration (DIR) between the planning CT and daily CBCTs, we iteratively correct CBCT intensities by matching local intensity histograms in conjunction with the DIR process. We tested our DIR method on six head and neck (HN) cancer cases, producing improved registration quality. Our method produced overall NMI of 0.663 and NCC of 0.987, outperforming conventional methods by 3.8% and 1.9%, respectively. The overall ART process has been validated on two HN cancer cases, showing differences between the planned and the actually delivered dose values. Both DIR and dose computation modules are accelerated by GPUs, and the computation time for DIR and dose computation at each fraction is ~1min.