Interactive and intuitive segmentation of volumetric data: the segmentVIEW system and the Kooshball algorithm

We present two algorithms and a system for segmentation and analysis of three-dimensional medical image data. The Kooshball algorithm operates by recursively projecting uniform distributions of rays from seed points, with gradient filters convolved along the rays to create candidate surface points. This cloud of candidate points is pruned via several transformation and filtering steps, resulting in stacks of 2D contours which are fed into the SegmentVIEW visualization system and processed by an energy-minimizing contour fitting algorithm. The contours are displayed directly on data surfaces, and both the contours and the underlying 3D data can be interactively manipulated by the user; a central goal of the SegmentVIEW system is to integrate the three major methods of volume visualization (ray casting, surface generation and rendering, and sectioning) in a manner so that manipulations are performed directly and intuitively on data, features and objects in a 3D space. Tests with the Kooshball algorithm and the SegmentVIEW system show that complex 3D features can be rapidly segmented, quantified and visualized with a high degree of accuracy regarding feature locations and volumes