Program Information

Adaptive Direction-Dependent Regularization for CT Abdomen Deformable Image Registration

D Yang

Y Fu , S Liu , H Li , D Yang*, Washington University School of Medicine, St. Louis, MO

Presentations

SU-K-201-11 (Sunday, July 30, 2017) 4:00 PM - 6:00 PM Room: 201

Purpose: Conventional isotropic smoothing in deformable image registration (DIR) methods cannot effectively estimate complex organ motions, which often result in over-smoothed motion field across organ boundaries. To accurately estimate human abdominal motion, an adaptive direction-dependent regularization technique is proposed.

Methods: An adaptive direction-dependent regularization filter is designed to replace the isotropic smoothing filter of the Horn-Schunck optical flow DIR method. The adaptive filter applies a Gaussian isotropic filter in the normal direction and a bilateral filter in the tangential direction of the organ boundaries. The adaptive filter ensures a smooth motion field in the normal direction, and meanwhile allows sliding motion along the organ boundaries. Intensity gradients of the CT images were used to estimate the primary normal direction at organ boundaries. End-inhalation and end-exhalation of the 4D abdomen CT images of the same patient were registered using the proposed method. For comparison, DIRs were performed using three different regularization methods, Gaussian isotropic filtering (G_iso), bilateral filtering (G_bil) and the proposed adaptive direction-dependent filter (G_dd) on three patient datasets. For each case, registration accuracy was evaluated against 120 manually defined landmarks across the abdomen and lower lung area.

Results: The results show that the registration with G_dd performed the best among the three regularization methods. Average target registration errors (TRE) for DIR with G_iso, G_bil, G_dd are 6.62 mm, 5.86 mm and 4.89 mm respectively.

Conclusion: An adaptive direction-dependent regularization technique is proposed for DIR to support sliding motion along the tangential direction of organ boundaries without the need of prior organ segmentation, and meanwhile to ensure motion smoothness in the normal direction. It is useful to improve the overall DIR accuracy when sliding motion of organs is involved.

Funding Support, Disclosures, and Conflict of Interest: Funding: AHRQ R01-HS022888 No conflict of interest Disclosures: Authors have technology licensing fee from Viewray


Contact Email: