Unit: Quantitative Image Analysis - URA CNRS 2582
Director: Olivo-Marin Jean-Christophe
The research activities of the Quantitative Image Analysis (AIQ) Unit are dedicated to the development of methods for the processing and the analysis of visual information produced in biological microscopy. Our objective is to develop algorithms making it possible to quantify in an automatic way the images produced within the framework of research in biology and to facilitate the understanding of biological information contained therein.
Our main research topics are active contours and level sets models, tracking of fluorescent spots and the segmentation of color images. Our methodological research has enabled us to develop programs for the analysis of the motility and the change in shape of moving biological objects, the tracking and the trajectory calculation of fluorescent spots in 3D+t microscopy, the quantification of fluorescence and the analysis of color images.
Shape and motion analysis
(C. Zimmer, A. Dufour, S. Berlemont, A. Thébaud, J.-C. Olivo-Marin)
The methodological base of this work is the approach of the "deformable models" or " active contours ". We borrowed this approach from research in computer vision, where it is very widespread and generated a great number of applications and adaptations. Active contours are mathematical curves employed to represent contours of an object. They evolve/move repeatedly under the action of "forces" calculated starting from the image and of internal geometrical properties. More precisely, the evolution of the curves is governed by an equation conceived to minimize a mathematical function of the curve, usually called the energy. This energy is an essential ingredient of the method, and several developments described below are modifications of this energy. We continued our work to adapt the models of active contours to the needs and specificities for the biological imagery and we concentrated our efforts on coupled parametric active contours, coupled active surfaces and the target detection quasi-circulars with level-sets. Standard contours active parametric are well adapted to the follow-up of islolated objects, but do not manage to follow several objects when those come in contact, which constitutes a major limitation for the tracking of cells. In previous work, we had proposed a solution based on non-parametric coupled active contours (level-set), which helps to maintain separation between cells in contact but which is too complex from a computational point of view to be applied in routine to large sequences. We thus had the idea to introduce the coupling between active contours into the formalism of parametric active contours. That makes it possible to combine the possibility of following cells in contact with the computational effectiveness of parametric active contours. Moreover, we use from now on a method much more robust to define the cellular borders, which allows the segmentation of fuzzy cells in microscopy fluorescence on low level of signal.
Cell motility and morphology are increasingly studied in 3D natural or artificial environments. In order to track and segment cells from 3D+time image data, we have developed a method based on coupled active surfaces. This technique extends our previous work from 2D to 3D and takes advantage of cell volume homeostasis to further improve the handling of touching cells.
In images presenting large variability of intensity, color or textures, it is helpful or necessary to rely on shape in order to correctly detect and outline significant objects (e.g. cells, or nodules in histology images). Shape constraints can be incorporated into active contour attached to single objects, but extending this to an unknown number of objects is a difficult and unresolved goal. We have made a step in this direction for the special but important case of nearly circular objects. The proposed, fully automatic scheme, allows better identification of round structures in histology images where color and texture information are unreliable.
All these topics continue to find applications in collaboration projects with S. Blazquez and E. Labruyère (BCP, N. Guillén) and V. Shinin (CSD, S. Tajbaksh).
Tracking of spots in 4D microscopy images
(A. Genovesio, F. Ollivier, J.-C. Olivo-Marin)
Generally speaking, the tracking of mobile objects can be described as the combination of several stages: 1) detection, which consists in taking a significant measurement of a generally noisy signal; 2) estimation, which consists in comparing a real measurement with a prediction of its most probable state given by a model or a priori information; 3) association, to find the best correspondence between new detections and the predicted estimates. We have developed procedures which make it possible to process the stages mentioned above and are used to carry out the tracking of multiple bright spots in scenes of 3D+t video microscopy in biology. The quantitative study of the motility and of the dynamic properties of the fluorescently labelled particles allows the calculation of spatio-temporal information of statistical nature like their number, position and spatial distribution, or of dynamic nature like their speed, phases of movement and coefficients of diffusion which are used to characterize the movement of the particles. The performances of our methods were evaluated favorably on synthetic data and real data in collaboration with N. Arhel (VMV, P. Charneau) et E. Merey et G. Duménil (INSERM, CHU Necker, Unité de pathogénie des infections systématiques, X. Nassif).
Spot analysis in immunofluorescence images
(A. Genovesio, B. Zhang, J.-C. Olivo-Marin)
Automatic quantification of immunofluorescence images relies either on the detection and counting of spots superimposed 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. based on a multiscale approach that uses a shift invariant discrete wavelet transform(SI-DWT) and on the selective filtering of wavelet coefficients. This scheme allows to separate and characterize objects of different sizes by selecting only a vicinity of detail images with corresponding scales adapted to the size of the spots. The extraction step consists in retaining the significant responses of the locally supported detail signal filters to the desired features, at the different scales of the wavelet representation. This is accomplished through a denoising technique using a threshold value which is image and level dependent and which can be computed automatically from the data. The program detects spots both in 2D and in 3D images and is used by several biological scientists to quantify their images.
Color image segmentation in histology and cytology
(V. Meas-Yedid, E. Glory, J. Ettedgui, J.-C. Olivo-Marin)
In the context of our work on the segmentation of images color for applications in histopathology and cytology, we have pursued a project dedicated to the evaluation of the quality of the segmentations of color images of cells and tissues, observed in optical microscopy. We have proposed a new criterion based on the homogeneity of the color of the segmented areas and the penalization of small areas. This criterion has enabled us to notably improve the robustness of the segmentations to the variability of color labelling and histological tissue architecture. It also greatly helped to decrease significantly the error rates in applications of automatic counting of cells in culture. We also developed one plug-in allowing to represent and interact with the distribution of the colors of an image in a space of colors 3d. This module brings a great help for the evaluation and the validation of the automatic segmentations.
To characterize culture parameters of human muscular cells, an automatic tool based on image analysis has been developed. The acquisition system consists in an inverted microscope, a colour camera and a motorized stage with autofocus. Colour images are segmented by a patented algorithm which automatically and objectively quantifies cell number. Data are analysed to characterise the growth of cellular population (doubling time, size and morphology of nuclei). This tool is in use in collaboration with the start-up company Cellogos.
Keywords: image processing, motility, shape, microscopy, color image analysis, active contours, level sets, wavelets, Kalman filter, probabilistic data association filters