PDF version of this entire document

Image Data

To conduct the experiments, a set of MR images of the brain was used. The set, which will be referred to as the `MGH Dataset' (see Acknowledgements), was a set of 2-D transaxial mid-brain slices, extracted at an equivalent level from each of a set of affinely aligned T1-weighted 3-D MR scans of $\mathcal{N}=36$ normal subjects. As well as the images themselves, there was access to ground-truth data, in the form of dense (pixel by pixel) anatomical label maps for the grey and white matter, the caudate nucleus, and the lateral ventricles. These labels were further divided into left and right hemispheres. The anatomical labels were obtained by manual annotation under conditions of rigorous quality control (details available at www.cma.mgh.harvard.edu). An example image and the corresponding label maps are shown in Figure .

Figure: An example affinely-aligned brain image and its accompanying anatomical labels, both overlaid and expanded, for grey matter, white matter, the lateral ventricles, and the caudate nucleus. The labels are also divided into left and right.

The set of images was non-rigidly registered using a Minimum Description Length (MDL) NRR algorithm [], and this registration was used as the starting point for a systematic evaluation of the effects of misregistration. Although the initial registration cannot be assumed to be perfect, it is reasonable to assume that each deformation degrades the registration rather than improving it.