Room: Track 2
Purpose: To investigate a multi-temporal resolution 3D motion prediction scheme that accounts for both breathing and slow drifting motion in the abdomen using radial MRI.
Methods: Ten-minute MRI scans were acquired for 8 patients using a volumetric golden-angle stack-of-stars sequence. The first five minutes were used to construct a model for breathing motion prediction, which consists of temporal deformation vector fields (tDVFs) with 170 ms intervals that deformably map the motion state of each k-space radial sample (spoke) to a reference scan. Gaussian kernel regression was used to learn a nonlinear mapping between observation (the most recently acquired 5 spokes) and prediction (first principal component coefficient of DVFs associated with a future spoke 340 ms ahead). A slow motion model was built by reconstructing images from breathing motion-corrected spokes using a temporal-view sharing filter with a resolution of 17 sec. Motion states were determined by deformably aligning the reference scan to reconstructions. Extrapolating the first principal component coefficient from a sliding window of 4 prior deformation fields predicts future slow motion states 8.5 sec ahead. The second five-minute scan was used to compare actual motion states obtained from retrospective reconstruction/deformation with predictions. Motion traces of abdominal organ centroids were also evaluated for both breathing and slow drifting motion.
Results: The averaged mean absolute errors between predicted and actual deformation vectors for 14000 breathing/1400 slow motion states were 0.51/0.31mm, 0.66/0.29mm and 0.89/0.43mm in lateral, anterior-posterior and superior-inferior directions respectively. The root-mean-square errors of motion traces in the superior-inferior direction induced by breathing were 0.39/mm,1.39mm, and 1.91mm for liver, small bowel and stomach respectively and were less than 0.2 mm for slow drifting motion of all organs evaluated.
Conclusion: Multi-temporal resolution motion models support volumetric predictions of breathing and slow drifting motion with sufficient accuracy and temporal resolution for MR-based tracking.
Funding Support, Disclosures, and Conflict of Interest: Supported by NIH R01 EB016079