In quantitative magnetic resonance imaging (qMRI), quantitative tissue properties can be estimated by fitting a signal model to the voxel intensities of a series of images acquired with different settings. To obtain reliable quantitative measures, it is necessary that the qMRI images are spatially aligned so that a given voxel corresponds in all images to the same anatomical location. The objective of the present study is to describe and evaluate a novel automatic groupwise registration technique using a dissimilarity metric based on an approximated form of total correlation. The proposed registration method is applied to five qMRI datasets of various anatomical locations, and the obtained registration performances are compared to these of a conventional pairwise registration based on mutual information. The results show that groupwise total correlation yields better registration performances than pairwise mutual information. This study also establishes that the formulation of approximated total correlation is quite analogous to two other groupwise metrics based on principal component analysis (PCA). Registration performances of total correlation and these two PCA-based techniques are therefore compared. The results show that total correlation yields performances that are analogous to these of the PCAbased techniques. However, compared to these PCA-based metrics, total correlation has two main advantages. Firstly, it is directly derived from a multivariate form of mutual information, while the PCA-based metrics were obtained empirically. Secondly, total correlation has the advantage of requiring no user-defined parameter.

Additional Metadata
Persistent URL,
Conference 29th IEEE Conference on Computer Vision and Pattern Recognition Workshops, CVPRW 2016
Note geen abonnement
Guyader, J.-M, Huizinga, W, Fortunati, V, Poot, D.H.J, Kranenburg, M.V. (Matthijs Van), Veenland, J.F, … Klein, S. (2016). Total Correlation-Based Groupwise Image Registration for Quantitative MRI. In IEEE Computer Society Conference on Computer Vision and Pattern Recognition Workshops (pp. 626–633). doi:10.1109/CVPRW.2016.84