Click here to


Are you sure ?

Yes, do it No, cancel

Novel Hybrid Deformable Image Registration with Automatic Body-Cavity Segmentation to Overcome Sliding Motion for Robust Reconstruction of Time-Resolved 4DMRI

X Nie*, J Deasy , G Li , Memorial Sloan Kettering Cancer Center, New York, NY


(Tuesday, 7/16/2019) 10:30 AM - 11:00 AM

Room: Exhibit Hall | Forum 2

Purpose: This study aims to enhance the accuracy of deformable image registration (DIR) by overcoming sliding motion at the body-cavity interface through automatically segmenting the body cavity, performing separated DIRs, and finetuning whole-body DIR for time-resolved (TR) 4DMRI image reconstruction.

Methods: An automatic image segmentation tool for the body cavity was developed based on multiple 3D MR images, including both high-resolution and low-resolution images, prior to DIR in TR-4DMRI reconstruction. The automatic segmentation starts with the lung segmentation in coronal view and the most inferior lung contour was cylindrically extended into the abdomen as an initial guess for abdominal cavity. Local abdominal voxel intensities were renormalized for optimal contrast. High-resolution difference images were first applied to differentiate moving and stable tissues, followed by 3D edge extraction to modify the initial cylindrical cavity. The body cavity was applied to low-resolution image, assuming anterior body shell thickness being constant during respiration. Low-resolution image was also simulated by down-sampling high-resolution image. Segmented DIR was performed separately for body cavity and anterior body shell, leaving posterior body shell intact. Final DIR fine-tuning was applied to all body segments together. MRI images of three subjects were acquired and used under an IRB-approved protocol. As a control, conventional DIR was performed using whole-body images. The image quality was compared, referencing to the target MR image with voxel intensity correlation (VIC) for quantification.

Results: The hybrid DIR produces much closer results to the reference than the conventional DIR, both visually and quantitatively (VIC=0.95 vs. VIC=0.90). The sliding-motion artifacts, such as a misaligment at diaphragm-chest-wall interface and a deformed/moving spine, were eliminated.

Conclusion: We have demonstrated the feasibility of automatic hybrid DIR approach to enhance the accuracy and robustness of TR-4DMRI reconstruction, making the TR-4DMRI one-step closer to a clinical application for multi-breath motion simulation in radiotherapy.


Not Applicable / None Entered.


Not Applicable / None Entered.

Contact Email