Migratory cells, for example human retinal epithelial (RPE) cells, exhibit highly variable morphology. This makes it difficult to use traditional methods, such as the landmark based Procrustes analysis or feature based analysis, to quantitatively represent their shapes. We propose a novel framework to generate a low-dimensional representation of highly variable cell shapes. The framework lends itself readily to efficient exploratory analysis of a given cell shape dataset in order to visualise morphological trends in the data and reveal the intrinsic structure of various morphology-based cell phenotypes in the data. Preliminary results show that the framework is effective in revealing consistent morphological phenotypes.