Skip to Main Content
The purpose of the study was to compare the ability of a mutual information algorithm with that of a standard algorithm to align images of histological serial sections. The two align algorithms were implemented in C running on a Linux based PC. Both algorithms used the same gradient-based optimizer, but different cost functions standard (ST), and mutual information (MI) respectively. The object of the test was to align 4557 serial sections originating from a rat kidney. The alignment of kidney sections is difficult as these sections contain many nearly identical tubules, representing a high degree of translation symmetry. As a consequence there is a non-negligible chance of misalignment into a local minimum, making serial kidney sections good real life test objects for image alignment. We showed that images, which were difficult to align by the ST were easy to align with MI. We found that the most efficient strategy was first to align all 4557 images using the ST function and then to align the misaligned 54 images using the MI function.