Rotation-invariant cell shape representation and modeling with level sets and graphical models

It is important to be able to develop statistical models of the shape of biological cells. In this article, we consider an automated method for obtaining shape models. In this paper, we use a level-set based representation in combination with a circular-invariant density learning method (based on Gaussian mixture models) to represent shapes in a way suited to cells. We demonstrate this model on simple artificially-generated data as well as synthetic data produced using a simple simulation of cell protrusion dynamics with parameters derived from real cells. We show that the method is able to accurately capture both the similarities and variation of shapes across the data sets.