Anatomy-preserving nonlinear registration of deep brain ROIs using confidence-based block-matching

Brain atlases are commonly used in a number of applications such as MRI segmentation and surgery targetting. Our goal is to register a basal ganglia atlas to a subject using MR image registration. Existing registration methods are for the most part either too constrained (linear registration) or can deform deep brain ROIs into implausible anatomical shapes. We developed a block-matching registration method suitable for atlas registration, using a new confidence-based regularization of the vector field. The method was used to register a set of 17 manually segmented MRI onto one subject. Results show that basal ganglia structures were better registered than when using an affine registration method.