Share Email Print

Proceedings Paper

Automatic 3D segmentation of spinal cord MRI using propagated deformable models
Author(s): B. De Leener; J. Cohen-Adad; S. Kadoury
Format Member Price Non-Member Price
PDF $14.40 $18.00

Paper Abstract

Spinal cord diseases or injuries can cause dysfunction of the sensory and locomotor systems. Segmentation of the spinal cord provides measures of atrophy and allows group analysis of multi-parametric MRI via inter-subject registration to a template. All these measures were shown to improve diagnostic and surgical intervention. We developed a framework to automatically segment the spinal cord on T2-weighted MR images, based on the propagation of a deformable model. The algorithm is divided into three parts: first, an initialization step detects the spinal cord position and orientation by using the elliptical Hough transform on multiple adjacent axial slices to produce an initial tubular mesh. Second, a low-resolution deformable model is iteratively propagated along the spinal cord. To deal with highly variable contrast levels between the spinal cord and the cerebrospinal fluid, the deformation is coupled with a contrast adaptation at each iteration. Third, a refinement process and a global deformation are applied on the low-resolution mesh to provide an accurate segmentation of the spinal cord. Our method was evaluated against a semi-automatic edge-based snake method implemented in ITK-SNAP (with heavy manual adjustment) by computing the 3D Dice coefficient, mean and maximum distance errors. Accuracy and robustness were assessed from 8 healthy subjects. Each subject had two volumes: one at the cervical and one at the thoracolumbar region. Results show a precision of 0.30 ± 0.05 mm (mean absolute distance error) in the cervical region and 0.27 ± 0.06 mm in the thoracolumbar region. The 3D Dice coefficient was of 0.93 for both regions.

Paper Details

Date Published: 21 March 2014
PDF: 7 pages
Proc. SPIE 9034, Medical Imaging 2014: Image Processing, 90343R (21 March 2014); doi: 10.1117/12.2043183
Show Author Affiliations
B. De Leener, École Polytechnique de Montréal (Canada)
J. Cohen-Adad, École Polytechnique de Montréal (Canada)
Univ. de Montréal (Canada)
S. Kadoury, École Polytechnique de Montréal (Canada)

Published in SPIE Proceedings Vol. 9034:
Medical Imaging 2014: Image Processing
Sebastien Ourselin; Martin A. Styner, Editor(s)

© SPIE. Terms of Use
Back to Top