The PCA-Demons Algorithm for Medical Image Registration

Deformable medical image registration is vital to study anatomical morphology. As for clinical practical use, registration needs to be accurate, efficient and robust to avoid influences caused by various biases added to medical images. In this paper, principal component analysis (PCA) is implemented under the state-of-the-art diffeomorphic log demons registration framework (The corresponding algorithm is denoted as PCA-demons). Features detected by PCA are combined with correlation like sum of squared deviations (SSD), kendall, pearson and spearman as a new similarity standard. The PCA-demons was tested on public three dimensional (3D) and two dimensional (2D) datasets in the intra-subject and inter-subject manner accordingly. Experiment results show that PCA-demons is more robust, generic and converges faster compared with demons algorithm, even when noises exist in both fixed and moving images.