Share Email Print
cover

Proceedings Paper

Initialisation of 3D level set for hippocampus segmentation from volumetric brain MR images
Author(s): Maryam Hajiesmaeili; Jamshid Dehmeshki; Bashir Bagheri Nakhjavanlo; Tim Ellis
Format Member Price Non-Member Price
PDF $14.40 $18.00

Paper Abstract

Shrinkage of the hippocampus is a primary biomarker for Alzheimer’s disease and can be measured through accurate segmentation of brain MR images. The paper will describe the problem of initialisation of a 3D level set algorithm for hippocampus segmentation that must cope with the some challenging characteristics, such as small size, wide range of intensities, narrow width, and shape variation. In addition, MR images require bias correction, to account for additional inhomogeneity associated with the scanner technology. Due to these inhomogeneities, using a single initialisation seed region inside the hippocampus is prone to failure. Alternative initialisation strategies are explored, such as using multiple initialisations in different sections (such as the head, body and tail) of the hippocampus. The Dice metric is used to validate our segmentation results with respect to ground truth for a dataset of 25 MR images. Experimental results indicate significant improvement in segmentation performance using the multiple initialisations techniques, yielding more accurate segmentation results for the hippocampus.

Paper Details

Date Published: 16 April 2014
PDF: 5 pages
Proc. SPIE 9159, Sixth International Conference on Digital Image Processing (ICDIP 2014), 91591D (16 April 2014); doi: 10.1117/12.2064402
Show Author Affiliations
Maryam Hajiesmaeili, Kingston Univ. (United Kingdom)
Jamshid Dehmeshki, Kingston Univ. (United Kingdom)
Bashir Bagheri Nakhjavanlo, Kingston Univ. (United Kingdom)
Tim Ellis, Kingston Univ. (United Kingdom)


Published in SPIE Proceedings Vol. 9159:
Sixth International Conference on Digital Image Processing (ICDIP 2014)
Charles M. Falco; Chin-Chen Chang; Xudong Jiang, Editor(s)

© SPIE. Terms of Use
Back to Top