User Aligned Histogram Stacks for Visualization of Abdominal Organs via MRI

Abstract

Multi-dimensional transfer functions (MDTF) are occasionally designed as two-step approaches. At the first step, the constructed domain is modelled coarsely using global volume statistics and an initial transfer function (TF) is designed. Then, a finer classification is performed using local information to refine the TF design. In this study, both a new TF domain and a novel two-step MDTF strategy are proposed for visualization of abdominal organs. The proposed domain is generated by aligning the histograms of the slices, which are reconstructed based on user aligned majority axis/regions through an interactive Multi-Planar Reconstruction graphical user interface. It is shown that these user aligned histogram stacks (UAHS) exploit more a priori information by providing tissue specific inter-slice spatial domain knowledge. For initial TF design, UAHS are approximated using a multi-scale hierarchical Gaussian mixture model, which is designed to work in quasi real time. Then, a finer classification step is carried out for refinement of the initial result. Applications to several MRI data sets acquired with various sequences demonstrate improved visualization of abdomen.