Image-driven population analysis through mixture modeling

We present a novel, image-driven population analysis framework, called iCluster. iCluster processes a large set of images to determine a partitioning of the population based on image similarities, while establishing a dense spatial correspondence across individuals and computing the templates that represent each subpopulation. In experiments, we show that an image-driven partitioning of a large population reveals age effects on neuroanatomy and correlates with mental health.