SU-E-J-198: Evaluation of a Free-Form Intensity-Based Deformable Registration Method Using the POPI Model




Our goal is to evaluate the accuracy of a free-form intensity-based deformable registration method using the POPI model.


Five subjects with 4DCT datasets from the POPI model (Vandemeulebrouke Med Phys 2011) were used to assess deformable registration accuracy. Each subject contained 100 or more identified landmark points that corresponded between the end-inspiratory and end-expiratory phases. The 0% phase was registered to the 50% phase first using a rigid alignment followed by a freeform intensity-based deformable registration. Landmark displacement was measured after the rigid registration (initial displacement) and the deformable registration (residual error). A single subject (POPI2) with significant initial displacement was also registered using an interactive tool to influence the deformation by locking local alignments to help guide the deformation. Error was measured with and without using the tool for this subject.


The average initial displacement prior to deformable registration ranged from 5.7 mm to 14.0 mm. The mean (SD) residual error after deformation ranged from 1.0 mm (0.9) to 1.7 mm (3.3) for four of the five subjects. For POPI2 the initial residual deformable error was 4.6 mm (6.8). Using seven local alignments to guide the deformation the error for POPI2 decreased to 1.2 mm (1.0). The average error across all 5 subjects was 1.3 mm (2.0).


A free-form intensity-based deformable registration method was found to provide good accuracy with an average error of 1.3 mm. A method for locally guided deformation allowed for the accurate registration of a challenging case with significant respiratory motion. References1. Vandemeulebrouke et al. Med Phys 2011.2.

AS Nelson is an employee with ownership at MIM Software, Inc. JW Piper is an employee with ownership at MIM Software, Inc.