1. Morphological active contour driven by local and global intensity fitting for spinal cord segmentation from MR images.
- Author
-
Fouladivanda, Mahshid, Kazemi, Kamran, Helfroush, Mohammad Sadegh, and Shakibafard, Alireza
- Subjects
- *
SPINAL cord , *MAGNETIC resonance imaging , *NEUROLOGICAL disorders , *HAUSDORFF measures , *CEREBROSPINAL fluid - Abstract
Graphical abstract Highlights • Segmentation of the spinal cord from MR images. • The segmentation method uses a morphological region based active contours model. • The local information is obtained using local morphology fitting. • Robust to intensity inhomogeneity by using local and global information. • Robust towards active contour initialization in different vertebral levels. Abstract Background Spinal cord (SC) segmentation from magnetic resonance (MR) images can be used to study neurological disorders and facilitates group analysis. Variation of intensity inhomogeneity and small cross section of SC are difficulties that restrict automizing SC segmentation. New methods In this paper we present a method for accurate SC segmentation from MR images. The proposed morphological local global intensity fitting model (MLGIF) is based on region based morphological active contour model that utilizes local and global information. The local information is obtained using local morphology fitting and has been embedded into region based active contour to deal with images intensity inhomogeneity and variable contrast levels between SC and the cerebrospinal fluid. The contour evolution has been performed using successive application of a set of morphological operators. Results The proposed method has been validated on 28 T1-weighted and 29 T2-weighted MR images and simulated MR images with different noise levels. Assessment of the results shows the accuracy of the proposed method for SC segmentation. Comparison to existing method(s) The proposed MLGIF method was comparable with existing SC segmentation methods. Between segmented images and corresponding ground truth images, the mean DICE similarity coefficient, mean conformity coefficient and mean Hausdorff distance were 0.90 (092), 0.8 (0.83) and 0.85 mm (0.70 mm), respectively, for T1(T2)-weighted images. Conclusion The MLGIF model was proposed to achieve a robust method to deal with intensity inhomogeneity and lack of contrast between SC and surrounding tissues. Moreover, accuracy and less sensitivity to initial curve were seen. [ABSTRACT FROM AUTHOR]
- Published
- 2018
- Full Text
- View/download PDF