Lateral ventricle segmentation based on fusion of expert priors in AD

INTRODUCTION: Volume measurements of the lateral ventricles are often used as surrogate markers of atrophy and disease progression in studies of neurodegenerative brain disorders (Schnack et al., NeuroImage, 2001), (Xia et al., NeuroImage, 2004), (Nestor et al., Brain, 2008). Manual segmentation can be timeconsuming and has the drawbacks of interand intra-observer variability. Even though there is high contrast between tissue and cerebrospinal fluid (CSF) in MRI, segmentation of the lateral ventricles can be difficult. For example, partial volume effects make it difficult for region-growing algorithms to accurately segment the temporal horns and occipital poles of the ventricles. Choroid plexus has image intensity similar to grey matter (GM) on T1 weighted (T1w) images which can confound threshold-based techniques. As subjects age, the ventricles can increase in size significantly, and this can be amplified by disease and presents a challenge for warpingbased techniques. We present a precise and accurate technique to automatically segment the lateral ventricles (LV) and validate the method on a large cohort of patients with Alzheimer’s Disease (AD).