International audienceModern developments in light microscopy have allowed the observation of cell deformation with remarkable spatiotemporal resolution and reproducibility. Analyzing such phenomena is of particular interest for the signal processing and computer vision communities due to the numerous computational challenges involved, from image acquisition all the way to shape analysis and pattern recognition and interpretation. This article aims at providing an up-to-date overview of the problems, solutions, and remaining challenges in deciphering the morphology of living cells via computerized approaches, with a particular focus on shape description frameworks and their exploitation using machine-learning techniques. As a concrete illustration, we use our recently acquired data on amoeboid cell deformation, motivated by its direct implication in immune responses, bacterial invasion, and cancer metastasis

Understanding the mechanisms involved in cell deformation and motility is of major interest in numerous areas of life sciences. Precise quantification of cell shape requires robust shape description tools to be amenable to subsequent analysis and classification. The main difficulty lies in the great variability of cell shapes within socalled "homogeneous" populations. While basic shape descriptors fail to provide sufficient information and lack robustness to small shape variabilities, here we investigate the use of the Spherical Harmonics transform to efficiently extract and quantify cell shape and deformation. Using real 3D+t biological imaging data sets, we show that this tool allows to precisely characterize the cell shape both in a static and a dynamic manner, allowing to extract a wide range of qualitative and quantitative parameters, such as outliers individuals, redundant shape configurations and spatiotemporal deformation patterns.

Understanding the mechanisms involved in cell deformation and motility is of major interest in numerous areas of life sciences. Precise quantification of cell shape requires robust shape description tools to be amenable to subsequent analysis and classification. The main difficulty lies in the great variability of cell shapes within a given homogeneous population. In this work, we propose a framework for cell shape extraction and classification for 3D time-lapse sequences of living cells, based on the SPherical HARMonics transform (SPHARM). Starting from an initial segmentation of the cell surface over time, this mathematical representation enables us to represent each extracted surface by a unique set of coefficients, while taking into account invariance properties such as translation or orientation. Then, unsupervised classification is conducted using a multi-class K-Means approach, so as to extract the most pertinent number of classes representing the different phases of the cell deformation. Experimental results on several sequences give encouraging results, and show that the proposed approach can be used to perform automated sequence annotation, and can be further applied to compare shape characteristics across different cell populations.

No abstract