Probabilistic segmentation propagation from uncertainty in registration
Ivor Simpson, Mark W. Woolrich, Julia A. Schnabel · UCL Discovery (University College London) · 2011
In this paper we propose a novel approach for incorporating measures of spatial uncertainty which are derived from non-rigid registration, into propagated segmentation labels. In current approaches to segmentation via label propagation, a point-estimate of the registration parameters is used. However, this is limited by the registration accuracy achieved. In this work, we derive local measurements of the uncertainty of a non-rigid mapping from a probabilistic registration framework. This allows us to consider the set of probable locations for a segmentation label to hold. We demonstrate the use of this method on the propagation of accurately delineated cortical labels in inter-subject brain MRI using the NIREP dataset. We find that accounting for the spatial uncertainty of the mapping increases the sensitivity of correctly classifying anatomical labels.