Data-Driven, Entropy-Based Measures
for Assessing Non-Rigid Registration
We present a method for assessing the performance of non-rigid
registration algorithms, without the need for any form of ground truth.
The method exploits the fact that, given a set of non-rigidly registered
images, a generative statistical appearance model can be constructed.
The quality of the model depends on the quality of the registration,
and can be evaluated by comparing images sampled from it with the
original image set. We derive a measure which is is based on Shannon's
entropy and show that it demonstrates the loss of registration as
a set of correctly registered images is progressively perturbed. We
also show the tight correlation between our newly-proposed method
and an overlap-based measure, which is based on ground-truth anatomical
labels of the brain.