Information-Theoretic Multi-modal Image Registration Based on the Improved Fast Gauss Transform: Application to Brain Images

摘要

Performances of multi-modality image registration methods that are based on information-theoretic registration criteria crucially depend on the specific computational implementation. We proposed a new implementation based on the improved fast Gauss transform so as to estimate, from all available intensity samples, the intensity density functions needed to compute the information-theoretic criteria. The proposed and several other state-of-the-art implementations were tested and compared in 3-D rigid-body registration of multi-modal brain volumes. Experimental results indicate that the proposed implementation achieves the most consistent spatial alignment of brain volumes at a subpixel accuracy.