PDF version of this document

Method Validation

Using the perturbation method outlined succinctly in§7, we can measure the quality of different models. These models were obtained from the same the registrations with varying extents of perturbation applied,

 

Fig. 5. An older evidence that Generalisation and Specificity of brain models degrades as their registration degrades

The results of an experiment which tests the effect of increasing mis-registration are shown in Fig. 5. The curves indicate that, whichever shuffle 'scope' we choose, both Specificity and Generalisation increase, which means that they get worse as registration gets artificially degraded. These results which include a variety of shuffle distances show a consistent trend, which is an encouraging behaviour. It implies that we needn't compare each pixel against a very large region, hence efficiency if improved. Furthermore it shows that there is little dependency on yet another parameter, which is the region size for the shuffle distance.