Voxel Stuffing: high-quality volume interpolation from multiple sequences of cross-sectional images

Abstract

This paper proposes a new algorithm, called Voxel Stuffing, to reconstruct single high-quality volume data from multiple sparsely-spaced sequences of cross-sectional images acquired by magnetic resonance imaging (MRI). Although fine and isotropic cross-sectional images can be obtained by using the most advanced MRI facilities, sparse sampling is commonly performed in the clinical examination. Intensive feasibility study was performed with three regular grid volume data sets, whose sources include an analytic function; a numerical simulation; and measurements. In either case, the Voxel Stuffing algorithm generates a higher-quality volume data from triple sequences of cross-sectional images in comparison with any volume data reconstructed linearly from a single sequence of class-sectional images. The Voxel Stuffing algorithm is extended to reconstruct a rectilinearly structured volume data set from triple non-orthogonal sequences of cross-sectional images, which are taken commonly in the general MRI clinical examination. The effectiveness of the extended Voxel Stuffing algorithm is illustrated with an MRI data set for a human brain containing a tumor.