Spot analysis for immunofluorescence images (G. Cuartero, J.-C. Olivo-Marin)
Automatic quantification of immunofluorescence images relies either on the detection and counting of spots superposed on biological structures, usually immersed in a non-uniform background, or on the outlining of larger cellular compartments. We have developed methods for spot detection and characterisation that allow a fast and reproducible quantitative analysis of these images. From the algorithmic point of view, the problem of spot detection is treated as a generation-recombination process of multi-resolution response elements obtained from a wavelet representation of the image. In the generation step, our algorithm retains at each resolution level only the significant responses of a compact support detail filter, followed by adaptive thresholding. In the recombination step, a local correlation coefficient is computed from the filtered wavelet coefficients at each location in the image. We are currently working on detecting spots in 3D space in order to extend this method to dynamic microscopy.
Quantification of multi-modal images by active contours (V. Meas-Yedid, C. Zimmer, V. Der Aprahamian, F. Cloppet, G. Stamon, J.-C. Olivo-Marin)
Our purpose is to develop automatic or semi-automatic tools able to characterise the morphological changes of cells and to quantify the cytoskeleton molecules involved in cell interactions. The algorithm draws on several complementary information sources provided by different modes of microscopy, specifically fluorescence and phase contrast. In the first step, cell outlines are extracted from a phase contrast image with the help of active contours, thus providing a description of their morphology. In the second step, the fluorescence image is used to quantify cytoskeleton molecules in specific compartments defined by the contours extracted in the first step, via a sequence of thresholdings and Boolean operators. This project is part of the PTR (Transverse Research Project) Study of celllular polarity by image analysis, performed in collaboration with the "Unité de Biologie des Interactions Cellulaires" and the "Laboratoire des Systèmes Intelligents de Perception" (SIP) of University Paris V.
Analysis of color images (V. Meas-Yedid, F. Marache, J.-C. Olivo-Marin)
A first study on color images resulted in a method that produces a grey-level image consisting of a linear combination of the principal components of the selected color space. We implemented a trainable neural network (multi-level perceptron) that yields satisfactory image segmentation. In collaboration with the "Unité Génétique Mycobactérienne", we applied this technique to segment, and subsequently quantify the cells present in the lungs after an infection.
Shape and motion analysis (C. Zimmer, V. Der Aprahamian, S. Carme, V. Meas-Yedid, J.-C. Olivo-Marin)
We are adapting the approach of active contours for the segmentation and tracking of deformable objects in motion from large image sequences. In this model, the object outlines are obtained from a parametric curve, by minimising an energy functional that depends on the curve's geometry and on the image data in its neighbourhood. We use the non-potential force field provided by the gradient vector flow model, which reduces the sensitivity of segmentation to initial conditions. Combined with a dilation of the initial contours, this modification increases convergence robustness. This project is part of the PTR Study of celllular polarity by image analysis. The targeted applications are: (i) the study of motility and morphology changes in amoeba, in collaboration with the "Unité Pathogénie Microbienne Moléculaire" and (ii) the quantitative analysis of morphological changes undergone by T lymphocytes in the early stages of the immune response, in collaboration with the "Unité Biologie des Interactions Cellulaires".