Scoring functions of volume matching

Volume matching algorithms yield a set of putative alignments between a pair of volumes. These are typically scored according to some criterion internal to the software used. However, there are a range of scoring functions which can be applied to get external validation, and these can be sensitive to different characteristics of the volumes. If there is a consensus between several different scores, then one can have some confidence in the proposed alignment.

Several scoring functions have been designed for quantifying the quality of alignments. A comparison is available of these functions in the Vasishtan and Topf (2011) review paper:

1. Local Cross-Correlation Coefficient (Roseman AM. 2000) – a correlation coefficient between the density values of the two maps, calculated for the overlap region only
2. Laplacian-filtered CCF (Chacon and Wriggers, 2002) – the Laplacian filter helps to pick out the surfaces of the volumes
3. Overlap score (Chimera, Pettersen et al. 2004) – fraction of overlapping voxels with respect to the smaller of the two volumes
4. Core-weighted envelope score (Wu et al. 2003) – a cumulative measure based on the number of overlapping voxels, with a penalty for non-overlapping voxels
5. Mutual information score (Shatsky et al. 2009) – a measure of how similar the joint distribution of density values in the superimposed maps is to the individual density distributions
6. Chamfer distance (Knossow et al. 2008) – an average distance between nearest neighbour points on the surfaces of the two superimposed maps.
7. Normal Vector score (Ceulemans and Russell. 2004) – an average angle between vectors normal to the two surfaces

Many of these scoring functions depend on the surface of the volume being identified as a particular contour level of the map. The choice of contour level is not necessarily straightforward.