TY - GEN
T1 - Multimodal surface matching
T2 - 23rd International Conference on Information Processing in Medical Imaging, IPMI 2013
AU - Robinson, Emma C.
AU - Jbabdi, Saad
AU - Andersson, Jesper
AU - Smith, Stephen
AU - Glasser, Matthew F.
AU - Van Essen, David C.
AU - Burgess, Greg
AU - Harms, Michael P.
AU - Barch, Deanna M.
AU - Jenkinson, Mark
PY - 2013
Y1 - 2013
N2 - Group neuroimaging studies of the cerebral cortex benefit from accurate, surface-based, cross-subject alignment for investigating brain architecture, function and connectivity. There is an increasing amount of high quality data available. However, establishing how different modalities correlate across groups remains an open research question. One reason for this is that the current methods for registration, based on cortical folding, provide sub-optimal alignment of some functional sub-regions of the brain. A more flexible framework is needed that will allow robust alignment of multiple modalities. We adapt the Fast Primal-Dual (Fast-PD) approach for discrete Markov Random Field (MRF) optimisation to spherical registration by reframing the deformation labels as a discrete set of rotations and propose a novel regularisation term, derived from the geodesic distance between rotation matrices. This formulation allows significant flexibility in the choice of similarity metric. To this end we propose a new multivariate cost function based on the discretisation of a graph-based mutual information measure. Results are presented for alignment driven by scalar metrics of curvature and myelination, and multivariate features derived from functional task performance. These experiments demonstrate the potential of this approach for improving the integration of complementary brain data sets in the future.
AB - Group neuroimaging studies of the cerebral cortex benefit from accurate, surface-based, cross-subject alignment for investigating brain architecture, function and connectivity. There is an increasing amount of high quality data available. However, establishing how different modalities correlate across groups remains an open research question. One reason for this is that the current methods for registration, based on cortical folding, provide sub-optimal alignment of some functional sub-regions of the brain. A more flexible framework is needed that will allow robust alignment of multiple modalities. We adapt the Fast Primal-Dual (Fast-PD) approach for discrete Markov Random Field (MRF) optimisation to spherical registration by reframing the deformation labels as a discrete set of rotations and propose a novel regularisation term, derived from the geodesic distance between rotation matrices. This formulation allows significant flexibility in the choice of similarity metric. To this end we propose a new multivariate cost function based on the discretisation of a graph-based mutual information measure. Results are presented for alignment driven by scalar metrics of curvature and myelination, and multivariate features derived from functional task performance. These experiments demonstrate the potential of this approach for improving the integration of complementary brain data sets in the future.
UR - http://www.scopus.com/inward/record.url?scp=84887011429&partnerID=8YFLogxK
U2 - 10.1007/978-3-642-38868-2_40
DO - 10.1007/978-3-642-38868-2_40
M3 - Conference contribution
C2 - 24683992
AN - SCOPUS:84887011429
SN - 9783642388675
T3 - Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics)
SP - 475
EP - 486
BT - Information Processing in Medical Imaging - 23rd International Conference, IPMI 2013, Proceedings
Y2 - 28 June 2013 through 3 July 2013
ER -