Comparison of two distance based alignment methods in medical imaging

A k-dimensional (k-d) tree based alignment and its comparison with the standard distance map based alignment is presented. We first describe a brief outline of both distance based iterative alignment algorithms. The new k-d based technique uses a modified approximate nearest neighbor (ANN) library, which is designed for both exact and approximate nearest neighbor searching in multidimensional space. We performed self-alignment tests of the k-d tree based alignment and compared two different alignment methods using a large 3D dataset of a rodent brain. The results indicate that the k-d based image alignment is highly effective, accurate, reliable, and provides compatible errors with the distance map based alignment method. On the other hand, as a big advantage, k-d tree alignment requires significantly less virtual or physical memory; a critical issue for large datasets.