PDF version of this document

Assessing the Accuracy of Non-Rigid Registration With and Without Ground Truth

Abstract:

We compare two methods for assessing the performance of groupwise non-rigid registration algorithms. One approach, which has been described previously, utilizes a measure of overlap between data labels. Our new approach exploits the fact that, given a set of non-rigidly registered images, a generative statistical appearance model can be constructed. We observe that the quality of the model depends on the quality of the registration, and can be evaluated by comparing synthetic images sampled from the model with the original image set. We derive indices of model specificity and generalisation that can be used to assess model/registration quality. We show that both approaches detect the loss of registration as a set of correctly registered MR images of the brain is progressively perturbed. We compare the sensitivities of the different methods and show that, as well as requiring no ground truth, our new specificity measure provides the most sensitive approach to detecting misregistration.