Automatic lateral ventricle segmentation in infant population with high risk of autism

Autism is a developmental disorder with high degree of variability (Geschwind 2008). There are indications that abnormal subjects brain growth (Piven 1995) may result in different brain development trajectories in the early stages of life. Previous studies showed that significant differences in the shape of lateral ventricles (LV) could be found in primary school-age populations (Vidal 2008). In the current study, we present a new method of automatic segmentation of LV on MRI scans of young infants (6-24 month old), and apply it to a large database of scans of healthy infants and infants with high risk of autism spectrum disorder (ASD). Our method relies on the combination of state-of-the-art: non-local patch-based segmentation technique (Coupe 2011) and non-linear registration (Collins 1995). Our results show high degree of accuracy when compared to the expert segmentations.