Multimodel Coregistration Evaluation

The poster evaluates registration evaluation metrics and proposes a combination of new and adapted metrics to address common problems with conventional measures.These problems include misleading baseline similarity, sensitivity to feature density and image overlap, and uncertainty about whether the correct cells are matched.The proposed framework combines chance-corrected NMI (NNMI), cell match counts, and local evaluation around matched cells.The poster also applies these metrics to a few registration tests.In these tests, fully automatic registration did not achieve satisfactory alignment, but existing automatic methods helped improve an initial manual registration.