Robustness Study

Quantitative results for robustness of FLIRT were ascertained using a consistency test. This test is designed to examine the robustness of a registration method by comparing registrations obtained using various different, but known, initial starting positions of a given image. Results showed that the method was highly consistent on a set of difficult images. Furthermore, several other available packages were tested on the same set of images and did not achieve the same level of consistency, sometimes demonstrating substantial inconsistencies. These tests (together with those presented in [12]) showed that the robustness was due to the optimisation method, not just the choice of cost function, and that in order to achieve the robust registrations, multi-resolution local optimisation alone is insufficient. Moreover, the newly proposed hybrid local-global optimisation method achieves a much greater degree of robustness, which is necessary for fully automatic use, within a prescribed time-limit (less than one hour on a PC; e.g. registering two $1 \times 1 \times 1$ mm images typically takes 15 minutes on a 500MHz Pentium III).