Paper Info
Title
MIND Demons for MR-to-CT deformable image registration in image-guided spine surgery.
Abstract
Purpose: Intraoperative localization of target and normal anatomy can be achieved through multi-modality deformable image registration to resolve planning data (e.g., tumor boundaries and adjacent vital structures) defined in preoperative MR in up-to-date intraoperative CT. We propose a new symmetric diffeomorphic deformable registration method to align preoperative MR to intraoperative CT using a modality-independent neighborhood descriptor (MIND) for image-guided spine intervention. Method: The method estimates symmetric time-dependent diffeomorphisms with smoothness priors on both the velocity fields and the diffeomorphisms using a Demons framework. Analysis of energy formulation (asymmetry vs symmetry), similarity metrics (L-2 vs Huber norms), and optimization techniques (a gradient descent method (GD), a Gauss-Newton method (GN) with an explicit matrix inversion (EMI), and GN with the Sherman-Morrison formula (SMF)) was performed in simulation, and performance was measured in terms of target registration error (TRE), properties of diffeomorphisms, and computational efficiency. Result: MIND Demons with the symmetric energy functional and the Huber metric yielded topology-preserving and invertible deformations with nonzero-positive Jacobian determinants, sub-voxel invertibility errors (<0.01 mm), and ability to resolve large deformation with TRE = 1.8 mm compared to 6.2 mm for an asymmetric method with L2, 4.9 mm for the asymmetric method with Huber, and 2.9 mm for the symmetric method with L2. MIND Demons using GN+SMF yielded TRE comparable to its variant using GD and GN+EMI but improved convergence and computation time. Conclusions: A promising method for MR-to-CT image registration has been developed, incorporating the symmetric energy functional and the robust Huber metric for a stable and reliable estimate of symmetric diffeomorphisms, and GN with SMF for computational efficiency. The approach yielded viscoelastic deformations with registration accuracy potentially suitable for high-precision image-guided interventions.
Year
DOI
Venue
2016
10.1117/12.2208621
Proceedings of SPIE
Keywords
Field
DocType
deformable image registration,Demons algorithm,symmetric diffeomorphism,multimodality image registration,MIND,CT,MRI,image-guided surgery
Computer vision,Normalization (statistics),Imaging phantom,Image-guided surgery,Artificial intelligence,Mutual information,Energy functional,Surgery,Exponential map (Riemannian geometry),Image registration,Geodesic,Physics
Conference
Volume
ISSN
Citations 
9786
0277-786X
0
PageRank 
References 
Authors
0.34
0
9
Name
Order
Citations
PageRank
S. Reaungamornrat1214.46
De Silva Tharindu2198.48
Ali Uneri311323.38
j p wolinsky442.83
a j khanna552.98
G Kleinszig63512.24
sebastian vogt734.20
Jerry L. Prince84990488.42
J H Siewerdsen910028.22