Skip to main content
U.S. Department of Energy
Office of Scientific and Technical Information

A pixel-connecting algorithm for enhancement and segmentation of computed tomography scans

Thesis/Dissertation ·
OSTI ID:5734363

The objective of the study was to enhance and segment X-ray Computerized Tomography (CT) scans. To deal with the noise and spatial complexity of these images, a relaxation algorithm was developed in order to link pixels together into homogeneous regions. Each pixel is assigned a set of weighted links to its nearest neighbors. The links are initially isotropic, and are arranged into stochastic link matrices. By computing powers of the link matrix, an object-dependent weighting mask for each pixel over an expanded neighborhood is found. The masks are used to compute a similarity measure between pixels in order to adjust the interlinks. The edges, which segment the image, are identified by the below-threshold links, and the displayed mask-weighted averages result in an enhanced image. The algorithm seems robust w.r.t. the five images tested: the images based on the weighted averages have a smooth appearance with sharpened edges. The algorithm has successfully segmented primary liver tumors of varying sizes and shapes. The links which drop below threshold highlight anatomical details of the scans which are difficult to visualize with the unaided eye.

Research Organization:
Akron Univ., OH (USA)
OSTI ID:
5734363
Country of Publication:
United States
Language:
English