Abstract # 2996 A New Hierarchical Symmetric Optical Flow-based Image Registration Method and its Clinical Evaluation

Presenter: Michalski, Darek

< Back to topics

> Cover Page, Authors, Institutions

> Purpose/Objective

> Materials/Methods

> Results/Discussion

> Conclusions

> View tables/images/videos

The accuracy of the method was estimated at 1.5 pixel. The worst results were obtained with finite difference method of gradient estimation. The multi‐point spatial and temporal gradient filters and B‐spline differentiation are qualitatively more robust, however they are also more time consuming methods. The registration was used to propagate contours along 3‐D CT volumes and model the thoracic tumor motion. This segmentation results differed from human delineated contours up to few pixels. The majority of the generated contours were gauged acceptable using human visual judgment criterion. The structures' dvh comparison confirmed their congruity. However, a human validation is still required after the registration. The tumor motion modeling is less sensitive to minor local discrepancies.