Wavelet Based MR 2D Slice Retrieval in 3D Volumes

A wavelet-based, multimodal, multiscale technique is proposed for 2D slice retrieval in 3D Magnetic Resonance (MR) volumes. The use of multiscale wavelet representation facilitates robust matching of important anatomical structures at multiple resolutions. This wavelet representation gives multiscale edges and hence it is very suitable for multimodal registration and is inherently insensitive to MR field inhomogeneity. The dyadic representation allows to use dyadic step size for optimization of registration parameter space for efficient computation. We propose a fast, 2D rigid image registration scheme using mutual information (MI) as a first step for retrieval. For efficient retrieval of relevant slices a novel MI based search scheme is developed. Experiments show promising results with respect to multi-modality, accuracy, speed and robustness.