Vladimir S. Fonov1, Sridar Narayanan1,
Douglas L. Arnold1, D. Louis Collins1
1McConnell Brain Imaging Centre,
Montreal Neurological Institute, Montreal, QC, Canada
Measurement
of lateral ventricular volume is often used for measuring the progression of
neurodegenerative diseases. Existing techniques for ventricle segmentation
have been extensively validated for the normal population, but may be
suboptimal for the patients with Alzheimers Disease (AD), for example. We
propose an automated method which uses information from expert manual
segmentation combined with a population-specific atlas. Experiments were
completed with a group of 271 elderly patients from an ongoing clinical
trial, using manual segmentations as a gold standard. We found that proposed algorithm yields
accurate results with a median kappa of 0.962.