The first point to note about the results shown in Figure is, that as we might have expected from the results shown in Figure , Generalisation 60#60 is not able to discriminant between the three NRR algorithms, having insufficient sensitivity. Specificity, however, as we might have expected from its superior sensitivity, can discriminate between the pairwise and groupwise methods; both groupwise registrations give lower values of the specificity measure than the pairwise algorithm. This difference persists as we vary the number of model modes and is statistically significant. We can conclude that either of these groupwise algorithms is superior to the pairwise algorithm.
There is possibly a slight difference when the two groupwise methods are compared, since one graph tends to lie lower than the other. However, when we compare this difference to the size of the error bars on the points, it is not large enough for us state that there is a statistically significant difference between the two groupwise variants.