Here are some basic statistical concepts needed to perform statistical analysis on segmented binary images. Neuroanatomical regions are often segmented manually or automatically using various segmentation tools such as mixture models, Mumford-Shah model or level set. The segmented regions of interest are assigned value 1 while the background is assigned value 0. Then it is natural to model the segmentation error using the Bernoulli distribution. In this chapter, we will introduce Bernoulli models for analyzing such binary images and apply them in detecting the regions of stroke lesions in diffusionweighted images (DWI).