We present a new semi-supervised method for segmenting multiple interrelated object boundaries with spherical
topology in volumetric images. The core of our method is a novel graph-theoretic algorithm that simultaneously
detects multiple surfaces under smoothness, distance, and elasticity constraints. The algorithm computes the
global optimum of an objective function that incorporates boundary, regional and surface elasticity information.
A single straight line drawn by the user in a cross-sectional slice is the sole user input, which roughly indicates
the extent of the object. We employ a multi-seeded Dijkstra-based range competition algorithm to pre-segment
the object on two orthogonal multiplanar reformatted (MPR) planes that pass through the input line. Based
on the 2D pre-segmentation results, we estimate the object and background intensity histograms, and employ
an adaptive mean-shift mode-seeking process on the object histogram to automatically determine the number of
surface layers to be segmented. The final multiple-surface segmentation is performed in an ellipsoidal coordinate
frame constructed by an automated ellipsoid fitting procedure. We apply our method to the segmentation of
liver lesions with necrosis or calcification, and various other tumors in CT images. For liver tumor segmentation,
our method can simultaneously delineate both tumor and necrosis boundaries. This capability is unprecedented
and is valuable for cancer diagnosis, treatment planning, and evaluation.
Proc. SPIE. 5370, Medical Imaging 2004: Image Processing
KEYWORDS: Image segmentation, 3D image processing, Image processing algorithms and systems, Computed tomography, Detection and tracking algorithms, Medical imaging, In vivo imaging, Detection theory, Algorithms, Algorithm development
In this paper, a novel polynomial-time algorithm is described for solving the optimal net surface detection problem on proper ordered multi-column graphs in N-D space (N ≥ 3). The method is applied to searching for optimal object boundaries with arbitrary smoothness constraints in volumetric medical images. By simple transformations, such optimal surface detection problems can be simplified to a problem of computing the minimum s-t cuts in the transformed graphs. An efficient implementation for the 3-D case that can achieve near real-time performance on moderate-sized datasets is presented. We further examine our technique in experiments by segmenting the cylindrical surfaces of human airways from pulmonary volumetric CT images, and compare the results to those produced by previous methods. By allowing full specifications of the cost-function and smoothness constraints without degrading the performance, the new algorithm is more flexible than traditional methods and guarantees global optimality. The multi-dimensional nature of the algorithm maintains continuity in higher dimensions.