Real-Time Automated Target Volume and Motion Quantification for Adaptive Four-Dimensional MRI/CT Guided Radiation Therapy

Abstract

To achieve real-time automated delineation of gross target volume (GTV) and to quantify changes in volume/position of the GTV during treatment, using a novel Morphological Processing and Successive Localization (MPSL) approach, for enabling adaptive four-dimensional (4D) MRI/CT guided radiation therapy. 4DCT scans for 10 non-small cell lung cancer patients and LUNGMAN anthropomorphic chest phantom were performed. In the phantom experiments, a moving lung tumor was simulated using a 6 cm diameter sphere (Virtual Water). Washington University 4D Motion Phantom was used to generate realistic motion profiles for the sphere. Real-time position management (RPM) was used for monitoring the breathing pattern of the patients and a motion file was obtained for retrospective phase binning. Cine CT images were correlated with respective breathing phase during retrospective phase binning. The average breathing time for one breathing cycle in the patients was 3 to 6 sec. Dynamic lung-MRI data of 30 sec duration was also acquired in five subjects (two normal, two with non-severe and one with severe lung disease) using a multi-echo vastly under-sampled isotropic projection (ME-VIPR) acquisition. The subjects performed a 4 sec inhalation, breath-hold of 10 sec, forced exhalation and normal breathing for the remaining duration, in sequence, during the 30 sec scan. Iterative Highly Constrained Back Projection (I-HYPR) algorithm was used to reconstruct different phase volumes separated by duration of 0.5 sec. Multi-detector CT (MDCT) data in the same subjects was also acquired. Automated target delineation and the subsequent volume and tumor motion quantification were performed using a novel MPSL approach (described elsewhere). A root mean square (RMS) error of less than 5% was achieved between input motion profiles and the profiles quantified by MPSL, in phantom experiments. The target volumes in the MR/CT in-vivo data were auto-contoured using MPSL and the results were compared with deformable registration based segmentation (conducted using MIMVista software). The amplitude of tumor motion quantified by MPSL in the ten cancer patients was of the range 4 to 15 mm. The correlation between the breathing trace and the tumor motion trajectory in each of three dimensions was analyzed. Importantly, MPSL segmentation of one volume (128×128×128) was achieved in 2 sec. The real-time automated volume/motion quantification achieved in this work is very attractive for reducing the time required for contouring the target volumes, for improving the reproducibility of the segmentation results and for introducing real-time adaptive radiotherapy in the treatment of lung cancer.