Search

International audience; Multi-parametric flow cytometry is a key technology for characterization of immune cell phenotypes. However, robust high-dimensional post-analytic strategies for automated data analysis in large numbers of donors are still lacking. Here, we report a computational pipeline, called FlowGM, which minimizes operator input, is insensitive to compensation settings, and can be adapted to different analytic panels. A Gaussian Mixture Model (GMM)-based approach was utilized for initial clustering, with the number of clusters determined using Bayesian Information Criterion. Meta-clustering in a reference donor permitted automated identification of 24 cell types across four panels. Cluster labels were integrated into FCS files, thus permitting comparisons to manual gating. Cell numbers and coefficient of variation (CV) were similar between FlowGM and conventional gating for lymphocyte populations, but notably FlowGM provided improved discrimination of "hard-to-gate" monocyte and dendritic cell (DC) subsets. FlowGM thus provides rapid high-dimensional analysis of cell phenotypes and is amenable to cohort studies.

Summary The conventional approach for microscopic 3D cellular imaging is based on axial through-stack image series which has some significant limitations such as anisotropic resolution and axial aberration. To overcome these drawbacks, we have recently introduced an alternative approach based on micro-rotation image series. Unfortunately, this new technique suffers from a huge burden of computation that makes its use quite difficult for current applications. To address these problems we propose a new imaging strategy called bi-protocol, which consists of coupling micro-rotation acquisition and conventional z-stack acquisition. We experimentally prove bi-protocol 3D reconstruction produces similar quality to that of pure micro-rotation, but offers the advantage of reduced computation burden because it uses the z-stack volume to accelerate the registration of the micro-rotation images.

Conventional flow cytometry (FC) methods report optical signals integrated from individual cells at throughput rates as high as thousands of cells per second. This is further combined with the powerful utility to subsequently sort and/or recover the cells of interest. However, these methods cannot extract spatial information. This limitation has prompted efforts by some commercial manufacturers to produce state-of-the-art commercial flow cytometry systems allowing fluorescence images to be recorded by an imaging detector. Nonetheless, there remains an immediate and growing need for technologies facilitating spatial analysis of fluorescent signals from cells maintained in flow suspension. Here, we report a novel methodological approach to this problem that combines micro-fluidic flow, and microelectrode dielectric-field control to manipulate, immobilize and image individual cells in suspension. The method also offers unique possibilities for imaging studies on cells in suspension. In particular, we report the system's immediate utility for confocal "axial tomography" using micro-rotation imaging and show that it greatly enhances 3-D optical resolution compared with conventional light reconstruction (deconvolution) image data treatment. That the method we present here is relatively rapid and lends itself to full automation suggests its eventual utility for 3-D imaging cytometry.

In order to extend the analysis of the datasets produced by lensfree video microscopy we have implemented a cell tracking algorithm to combine and correlate cell motility to the previously devised metrics to quantify e.g. cell adhesion and spreading, cell division, and cell death. In this paper we present the assessment of these new methodology on experiments involving three different cell lines, namely 3T3 fibroblast cells, primary HUVEC cells and macrophage THP1 cells. We demonstrate that the good spatial resolution and the fast frame rate obtained with of our lensfree video microscope allows standard cell tracking algorithm to be computed. The results is the possibility to analyze thousands of cells successfully tracked over tens of hours. The results is the possibility to compare different cell cultures in terms of e.g. cell motility and cell confinement ration. Ultimately we managed to measure the doubling time at single cell level over a large number of N=235 cells tracked over two days.

International audience; Quantification of basic cell functions is a preliminary step to understand complex cellular mechanisms, for e.g., to test compatibility of biomaterials, to assess the effectiveness of drugs and siRNAs, and to control cell behavior. However, commonly used quantification methods are label-dependent, and end-point assays. As an alternative, using our lensfree video microscopy platform to perform high-throughput real-time monitoring of cell culture, we introduce specifically devised metrics that are capable of non-invasive quantification of cell functions such as cell-substrate adhesion, cell spreading, cell division, cell division orientation and cell death. Unlike existing methods, our platform and associated metrics embrace entire population of thousands of cells whilst monitoring the fate of every single cell within the population. This results in a high content description of cell functions that typically contains 25,000 - 900,000 measurements per experiment depending on cell density and period of observation. As proof of concept, we monitored cell-substrate adhesion and spreading kinetics of human Mesenchymal Stem Cells (hMSCs) and primary human fibroblasts, we determined the cell division orientation of hMSCs, and we observed the effect of transfection of siCellDeath (siRNA known to induce cell death) on hMSCs and human Osteo Sarcoma (U2OS) Cells.

We introduce Auto-associative composite models, which have shown a good behavior on real data sets, and share important theoretical approximation properties. Their basic principle is to approximate iteratively data by manifolds of increasing dimension. We exhibit a special class of such models: auto-associative additive models. Their use is widespread in Projection pursuit regression. First, we show that Principal component analysis is a linear auto-associative additive model. Then, we show that principal component analysis is the only auto-associative composite model which is additive.

We have developed “Lensfree video microscopy” to bring new perspectives to cell biology, with its ability to perform real-time cell culture monitoring, with a field of view of 24mm2 encompassing several thousand cells

Owing to its simplicity, lensless imaging system is adept at continuous monitoring of adherent cells inside the incubator. The setup consists of a CMOS sensor with pixel pitch of 2.2 μm and field of view of 24 mm2, LED with a dominating wavelength of 525 nm, along with a pinhole of 150 μm as the source of illumination. The in-line hologram obtained from cells depends on the degree of cell-substrate adhesion. Drastic difference is observed between the holographic patterns of floating and adherent cells. In addition, the well-established fact of reduction of cell-substrate contact during cell division is observed with our system based on corresponding spontaneous transition in the holographic pattern. Here, we demonstrate that by recognizing this specific holographic pattern, number of cells undergoing mitosis in a cell culture with a population of approximately 5000 cells, can be estimated in real-time. The method is assessed on comparison with Edu-based proliferation assay. The approach is straightforward and it eliminates the use of markers to estimate the proliferation rate of a given cell culture. Unlike most proliferation assays, the cells are not harvested enabling continuous monitoring of cell culture.

In the context of unsupervised classification of multidimensional data, we revisit the classical mixture model in the case where the dependencies among the random variables are described by a DAG structure. This structure is considered at two levels, the original DAG and its macro-representation. This two-level representation is the main base of the proposed mixture model. To perform unsupervised classification, we propose a dedicated algorithm called EM-mDAG, which extends the classical EM algorithm. In the Gaussian case, we show that this algorithm can be efficiently implemented. This approach has two main advantages. It favors the selection of a small number of classes and it allows a semantic interpretation of the classes based on a clustering within the macro-variables.

Recently, micro-rotation confocal microscopy has enabled the acquisition of a sequence of slices of non adherent living cells obtained during a partially controlled rotation movement of the cell through the focal plane. Although we are now able to estimate the 3D position of every slice with respect to the frame, the reconstruction of the cell from the positioned slices remains a problem that this paper address. In our context, 3D spatially-varying PSF and missing data are the two main particularities of this problem. Experiments illustrate the ability of the classical EM algorithm to deconvolve efficiently cell volume and also to deal with missing data.

Smoothing by spline functions is a tried and tested method that is used in many different fields. As mentioned in the introductory chapter, the variational approach of one-dimensional splines provides an immediate introduction to the major issues inherent in modeling (regularization, Bayesian estimation, robustness, estimation of the regularization parameter, optimization) thanks to the simplicity of one-dimensional splines. Approximation splines also provide a first formalization of the intuitive concepts of smoothing and adaptive surfaces and curves. These are two essential concepts in imaging.

The two previous chapters involved modeling a random phenomenon X for which only one observed value was available.1 The information this value provided was not enough for us to deduce the random behavior of X,and this is why a prior information had to be introduced. This resulted in the use of a model for a smooth curve (image) degraded by random errors. When several observed values of the phenomenon are available, the situation is completely different, because we can then hope to deduce the variations in X without having to consider strong prior information.

This chapter deals with the processing of degradation on radiographic images. In many cases, the methods discussed can also be applied to fields other than radiography. The radiographic image acquisition process includes a physical phase in which the image is formed by radiation and is disturbed by blurring or scattering effects [17]. The second phase is the detection of the radiation by a detector characterized by a response function. In the third phase, the measuring instruments deliver a final image to which a measurement noise is added. We will present models for denoising, deblurring, scatter analysis, and sensitivity function estimation.

The purpose of reconstruction is to determine the internal structure of an object, given a set of views that are radiographic projections of the object. This technique was first used for medical applications, which are still the best-known examples of its use. It is also used for nondestructive testing in an industrial context, for example, to inspect welds or composite materials, or to monitor sensitive structures such as nuclear reactors. Here, we focus on tomographic techniques. These require the introduction of a regularization energy because the physics of the acquisition makes this an ill-conditioned problem.

We know from points (i) and (ii) of Section 1.1.2 that an inverse problem can be solved by constraining the solution either implicitly, via a prior energy U1 (x) as in Chapter 2, or explicitly by dimensionality reduction of the space containing x. The latter approach is the subject of this chapter. It is completely different from regularization with a prior energy. Dimensionality reduction has two advantages: It produces a simple exogenous model, and it is easy to handle when used as a component of a more complicated model.

One of the main tasks of nondestructive testing is to detect defects in materials. Here, we discuss the automatic detection of the projection of defects on radiographic images. We will focus in particular on gammaradiographic images in highly degraded situations that lead to barely visible defects. In this type of image, defects can easily be confused with artifacts. In fact when digitally processing a single image, it is virtually impossible to distinguish between artifacts and defects. Both entities appear as dark, elongated regions on a more or less uniform, noisy background. We will use the generic term “valleys” to describe them. This chapter is devoted to the detection of such valleys. We should emphasize that this detection problem is found in fields other than nondestructive testing [62, 82, 144]. We will conclude this chapter with an extension to the situation where we have a matched pair of images. The fusion of the two images allows us to distinguish between defects and artifacts.

These concepts of Bayesian estimation and inverse problems have already been introduced (Section 1.1). Here they are described more specifically in the Markov context, which will he one of the most important formalisms in Part III.

Quantification of cell proliferation and monitoring its kinetics are essential in fields of research such as developmental biology, oncology, etc. Although several proliferation assays exist, monitoring cell proliferation kinetics remains challenging. We present a novel cell proliferation assay based on real-time monitoring of cell culture inside a standard incubator using a lensfree video-microscope, combined with automated detection of single cell divisions over a population of several thousand cells. Since the method is based on direct visualization of dividing cells, it is label-free, continuous, and not sample destructive. Kinetics of cell proliferation can be monitored from a few hours to several days. We compare our method to a standard assay, the EdU proliferation assay, and as proof of principle, we demonstrate concentration-dependent and time-dependent effect of actinomycin D-a cell proliferation inhibitor.

We are given a set of points in a high dimensional space. For instance, this set can represent many visual appearances of an object, a face or a hand. We address the problem of approximating this set by a manifold in order to have a compact representation of the object appearance. When the scattering of this set is approximately an ellipsoid, then the problem has a well-known solution given by Principal Components Analysis (PCA). Yet, in some situations like object deplacement learning or face learning this linear technique can be ill-adapted and nonlinear approximation must be introduced. The method we propose can be seen as a Non Linear PCA (NLPCA), the main difficulty being that the data points are not ordered. We propose an index to find projection axes encouraging the choice of axes which preserve as well as possible the structure of the closest point neighborhood. These axes determine an order for visiting all the points when smoothing. Finally, a new criterion, called "generalization error" is introduced to determine the smoothing rate, that is the spline number of knots in this case. Experimental results conclude this paper: the method is tested on artificial data and on two data sets coming from databases used in visual learning.

Because of the intrinsic superimposition of structures in radiographies, their interpretation is a complex task. However, in such a situation, prior information on some of the observed objects is often available. For example, a set of reference radiographies of some structures may be available. Unfortunately, the objects' geometry is not perfectly reproducible. Thus, differences can occur between the object considered in reference radiographies and the one present in the radiographies. The authors propose a way to take into account the the deformations between the reference objects and the observed objects. A flexible model of the reference objects' radiographies is built up. This model is "learned" from radiographies of a set of reference objects. In the second section, various ways to build up a flexible model are compared. In the following section an automatic way to design a model adapted to deal with tolerances in radiographies of a class of 3D deformable objects is proposed. Finally, all applications of the method to quantitative printed circuit board solder inspection are presented. >

This presentation provides a review of our approach to custom total hip replacements including principles of design, instrumentation and technique. We focuss on the fitting of the stem from CT data and we show how strain contraints can drive a 3D deformable shape estimation of the stem, the basic shape beeing a parametric generalized cylinder

We address the problem of volume reconstruction from a sequence of cross-sections in the case where the cross-section positions are unknown. This implies performing simultaneous registration and reconstruction. We propose a statistical formulation of the problem leading to an energy minimization algorithm as well as an automatic calibration procedure for the energy parameters. This method has been developed in the context of micro-rotation confocal microscopy. Experiments in this context illustrate the ability of this method to reconstruct efficiently the object of interest.

The authors address the problem of identifying the projection of an object from incomplete data extracted from its radiographic image. They assume that the unknown object is a particular sample of a flexible object. Their approach consists first in designing a deformation model able to represent and to simulate a great variety of samples of the flexible object radiographic projection. This modellisation is achieved using a training set of complete data. Then, given the incomplete data, the identification task consists in estimating the observed object using the deformation model. The proposed modelling extracts from the training set, not only the deformation modes, but also other relevant information (such as probability distributions on the deformations, relations between deformations) to use it to regularise the identification step.

Biological studies always start from curious observations. This is exemplified by description of cells for the first time by Robert Hooke in 1665, observed using his microscope. Since then the field of microscopy and cell biology grew hand in hand, with one field pushing the growth of the other and vice-versa. From basic description of cells in 1665, with parallel advancements in microscopy, we have travelled a long way to understand sub-cellular processes and molecular mechanisms. With each day, our understanding of cells increases and several questions are being posed and answered. Several high-resolution microscopic techniques are being introduced (PALM, STED, STORM, etc.) that push the resolution limit to few tens of nm, taking us to a new era where ‘seeing is believing'. Having said this, it is to be noted that the world of cells is vast, with information spread from nanometers to millimetres, and also over extended time-period, implying that not just one microscopic technique could acquire all the available information. The knowledge in the field of cell biology comes from a combination of imaging and quantifying techniques that complement one another.Majority of modern-day microscopic techniques focuses on increasing resolution which, is achieved at the expense of cost, compactness, simplicity, and field of view. The substantial decrease in the field of observation limits the visibility to a few single cells at best. Therefore, despite our ability to peer through the cells using increasingly powerful optical instruments, fundamental biology questions remain unanswered at mesoscopic scales. A global view of cell population with significant statistics both in terms of space and time is necessary to understand the dynamics of cell biology, taking in to account the heterogeneity of the population and the cell-cell variability. Mesoscopic information is as important as microscopic information. Although the latter gains access to sub-cellular functions, it is the former that leads to high-throughput, label-free measurements. By focussing on simplicity, cost, feasibility, field of view, and time-lapse in-incubator imaging, we developed ‘Lensfree Video Microscope' based on digital in-line holography that is capable of providing a new perspective to cell culture monitoring by being able to capture the kinetics of thousands of cells simultaneously. In this thesis, we present our lensfree video microscope and its applications in in-vitro cell culture monitoring and quantification.We validated the system by performing more than 20,000 hours of real-time imaging, in diverse conditions (e.g.: 37°C, 4°C, 0% O2, etc.) observing varied cell types and culture conditions (e.g.: primary cells, human stem cells, fibroblasts, endothelial cells, epithelial cells, 2D/3D cell culture, etc.). This permitted us to develop label-free cell based assays to study the major cellular events – cell adhesion and spreading, cell division, cell division orientation, cell migration, cell differentiation, network formation, and cell death. The results that we obtained respect the heterogeneity of the population, cell to cell variability (a raising concern in the biological community) and the massiveness of the population, whilst adhering to the standard cell culture practices - a rare combination that is seldom attained by existing real-time monitoring methods.We believe that our microscope and associated metrics would complement existing techniques by bridging the gap between mesoscopic and microscopic information.

Liquid chromatography mass spectrometry is a promising technique in analytical chemistry for the discovery of protein biomarkers. This thesis deals with correcting distortions such as LC/MS image alignment and intensity standardization. We then apply a contrario detection, and study the limit of detection of the proposed algorithm., La spectrométrie de masse LC/MS est une technique de chimie analytique très prometteuse pour la recherche de biomarqueurs protéiques. Cette thèse aborde la correction de certaines distortions : l'alignement des images LC/MS et la normalisation des intensités. Nous étudions ensuite un détecteur de signaux a contrario et calculons la limite de détection.

Liquid chromatography mass spectrometry is a promising technique in analytical chemistry for the discovery of protein biomarkers. This thesis deals with correcting distortions such as LC/MS image alignment and intensity standardization. We then apply a contrario detection, and study the limit of detection of the proposed algorithm.; La spectrométrie de masse LC/MS est une technique de chimie analytique très prometteuse pour la recherche de biomarqueurs protéiques. Cette thèse aborde la correction de certaines distortions : l'alignement des images LC/MS et la normalisation des intensités. Nous étudions ensuite un détecteur de signaux a contrario et calculons la limite de détection.