Encrypted login | home

Program Information

New Visualization Methods for Longitudinal MRI Registrations and Segmentations

no image available
H Veeraraghavan

H Veeraraghavan*, J Deasy , Memorial Sloan Kettering Cancer Center, New York, NY

Presentations

SU-E-J-196 Sunday 3:00PM - 6:00PM Room: Exhibit Hall

Purpose: To develop visualization techniques to facilitate easy assessment of (a) registration and (b) tracking volumetric changes in structures during radiation therapy from MRI.

Method: The frequently used method for visualizing registrations between scans is a multi-color overlay technique or deformation vector fields. However, the overlay technique is unintuitive and does not help to appreciate the quality of registration particularly when the registration mismatches are not very large. Similarly, the deformation fields give an indication of extent of deformation but do not help to assess the differences in registration. We present a mirroring and edge-augmented mirroring technique that places the fixed and moving image next to each other and allows the user to quickly assess the small differences in registration. Next, we present a volumetric intersection based 3D model to visualize the changes in diseased lymph node volumes in head and neck cancer. 3D model-based visualization provides more information about the location-specific changes in volume rather than the simplistic one dimensional information obtained from 2D plot of nodal volume changes.

Result: We show results comparing our approach with the standard color-based overlay method for comparing registrations of intra-patient registrations using T2-MRI. Upon comparing the mirroring technique with the color-overlay, one can more easily appreciate the differences in registration. Adding edge-based mirroring seems to further assist in evaluating the registration. Our approach for viewing registrations seems to be more intuitive and easy to use in order to help assess the quality of registration compared to color-based overlays. Similarly, the change volumetric model together with a 2D plot reveals more information including the locations undergoing changes and responding to treatment.

Conclusions: Better approaches are necessary for assessing the quality of registrations and changes in diseased structures over time. Our approach presents some novel visualization solutions for registration and segmentation.



Contact Email: