Room: Karl Dean Ballroom B1
Purpose: Respiratory motion models have a number of applications in magnetic resonance image (MRI) guided radiotherapy. Model performance depends on correlation of the respiratory surrogate with internal motion. This work evaluates one external and three image-derived respiratory surrogates for motion compensation across multiple imaging planes during MRI-guided radiotherapy.
Methods: Six volunteer studies were performed using a 0.35T MRI-guided radiotherapy system while simultaneously recording an external respiratory bellows surrogate. Fifty images were obtained in an interleaved fashion at ten adjacent slice positons exhibiting respiratory motion. Three image-derived respiratory motion surrogates were evaluated: the center of k-space signal, body area within a manually selected region of interest (ROI), and a manifold learning technique known as locally linear embedding (LLE). The center of k-space signal and body area image surrogates were normalized across slices according to their 25th to 75th percentile range. Normalization of motion information across slices during manifold learning was performed by simultaneous group-wise alignment of the manifolds. Motion resolved image reconstructions were evaluated computing the normalized correlation coefficient (NCC), sum-of-squares difference (SSD) and mean registration distance (MRD) relative to a breath-hold scan for voxels containing tissue (excluding the bowel). The surrogate values corresponding to each image in the breath hold scan were used to select an image at each slice location to produce a motion resolved image reconstruction.
Results: The mean NCC for the bellows, center of k-space, body area and manifold surrogate reconstructions were 0.892, 0.882, 0.891 and 0.897 respectively. The mean SSD corresponding to these surrogates were 0.103, 0.116, 0.097 and 0.096 respectively. The MRD was 2.263 mm, 2.448 mm, 2.819 mm and 2.150 mm for these surrogates.
Conclusion: Initial results indicate optimal motion compensated reconstruction varies depending on evaluation metric. Of the image-derived respiratory surrogates, only LLE outperforms the respiratory bellows across all evaluation metrics.
Funding Support, Disclosures, and Conflict of Interest: Some authors have previously received speaking and consulting fees from ViewRay.