Computational analysis of LDDMM for brain mapping

Abstract

One goal of computational anatomy (CA) is to develop tools to accurately segment brain structures in healthy and diseased subjects. In this paper, we examine the performance and complexity of such segmentation in the framework of the large deformation diffeomorphic metric mapping (LDDMM) registration method with reference to atlases and parameters. First we report the application of a multi-atlas segmentation approach to define basal ganglia structures in healthy and diseased kids' brains. The segmentation accuracy of the multi-atlas approach is compared with the single atlas LDDMM implementation and two state-of-the-art segmentation algorithms—Freesurfer and FSL—by computing the overlap errors between automatic and manual segmentations of the six basal ganglia nuclei in healthy subjects as well as subjects with diseases including ADHD and Autism. The high accuracy of multi-atlas segmentation is obtained at the cost of increasing the computational complexity because of the calculations necessary between the atlases and a subject. Second, we examine the effect of parameters on total LDDMM computation time and segmentation accuracy for basal ganglia structures. Single atlas LDDMM method is used to automatically segment the structures in a population of 16 subjects using different sets of parameters. The results show that a cascade approach and using fewer time steps can reduce computational complexity as much as five times while maintaining reliable segmentations.

Parametric Surface Diffeomorphometry for Low Dimensional Embeddings of Dense Segmentations and Imagery

“…Our goal is to simultaneously identify parametric geodesic coordinates of each subcortical motor structure (caudate, putamen, globus pallidus, as well as the thalamus), from the T1 image data directly. To this end we include a dataset obtained as part of a study of children with attention deficit disorder and autism spectrum disorder [52]. Patients have a mean age of 10.2 yrs, and T1-weighted 3D-volume MPRAGE coronal images were acquired from a Philips 3T Achieva MRI scanner (Best, the Netherlands) using an 8-channel head coil (TR = 7.99 ms, TE = 3.76 ms, Flip angle = 8°, voxel size = 1mm isotropic).…”

Shape and diffusion tensor imaging based integrative analysis of the hippocampus and the amygdala in Alzheimer's disease

“…These template surfaces for our structures of interest were created manually, ensuring sufficient smoothness and correct anatomical topology. Each optimized diffeomorphism was obtained through using LDDMM with appropriately selected parameters [47] to map the template segmentation to the scan-specific segmentation of the same structure. This surface-generation methodology was used to create the target shapes whose localized surface-based morphometrics, in terms of the surface areas associated to vertices of the triangulated mesh, were then extracted.…”

An image registration pipeline for analysis of transsynaptic tracing in mice

“…This cascading-α approach, though computationally more expensive, has been shown to be more robust in practice. 17 For the STP image the cascading-α sequence of 0.05, 0.02 and then 0.01 was found to be sufficient. A flowchart which summarizes the entire registration procedure is shown in Fig.…”