A multi-resolution active contour framework for ultrasound image segmentation
Chapter in an edited book (author)

Altmetrics Information

Other information
AbstractA novel multi-resolution framework for ultrasound image segmentation is
presented. The framework is based on active contours and exploits both local
intensity and local phase features to deal with degradations of ultrasound images
such as low signal-to-noise, intensity inhomogeneity and speckle noise. We first
build a Gaussian pyramid for each input image and employ a local statistics
guided active contour model to delineate initial boundaries of interested objects
in the coarsest pyramid level. Both means and variances of local intensities are
utilized to handle local intensity inhomogeneity. In addition, as speckle noise is
greatly reduced in the coarsest pyramid level, the contours can avoid trapping
in local minima during the evolution. A phase-based geodesic active contour
(GAC) is implemented to refine the boundaries in finer pyramid levels.
Compared to traditional gradient-based GAC methods, the phase-based model
is more suitable for ultrasound images with low contrast and weak boundaries.
We employed the proposed framework for left ventricle, liver and kidney
segmentation in echocardiographic images; comparative experiments demonstrate
the advantages of the proposed segmentation framework.
All Author(s) ListWeiming Wang, Jing Qin, Pheng-Ann Heng, Yim-Pan Chui, Liang Li, Bing Nan Li
All Editor(s) ListC. H. Chen
Journal nameComputer Vision in Medical Imaging
Book titleComputer Vision in Medical Imaging
Series TitleSeries in Computer Vision
Volume Number2
PublisherWorld Scientific
Pages115 - 129
LanguagesEnglish-United States

Last updated on 2021-13-09 at 23:59