<p dir="ltr">This dissertation encompasses three studies in distinct yet impactful domains: B2B marketing, real-time video super-resolution (VSR), and smart office document routing systems. In the B2B marketing sphere, the study addresses the extended buying cycle by developing an algorithm for customer data aggregation and employing a CatBoost model to predict potential purchases with 91% accuracy. This approach enables the identification of high-potential<br>customers for targeted marketing campaigns, crucial for optimizing marketing efforts.<br>Transitioning to multimedia enhancement, the dissertation presents a lightweight recurrent network for real-time VSR. Developed for applications requiring high-quality video with low latency, such as video conferencing and media playback, this model integrates an optical flow estimation network for motion compensation and leverages a hidden space for the propagation of long-term information. The model demonstrates high efficiency in VSR. A<br>comparative analysis of motion estimation techniques underscores the importance of minimizing information loss.<br>The evolution towards smart office environments underscores the importance of an efficient document routing system, conceptualized as an online class-incremental image classification challenge. This research introduces a one-versus-rest parametric classifier, complemented by two updating algorithms based on passive-aggressiveness, and adaptive thresholding methods to manage low-confidence predictions. Tested on 710 labeled real document<br>images, the method reports a cumulative accuracy rate of approximately 97%, showcasing the effectiveness of the chosen aggressiveness parameter through various experiments.</p>

No description available.

Correlative light and electron microscopy (CLEM) is a powerful approach to investigate the molecular ultrastructure of labeled cell compartments. However, quantitative CLEM studies are rare, mainly due to small sample sizes and the sensitivity of fluorescent proteins to strong fixatives and contrasting reagents for EM. Here, we show that fusion of a self-labeling protein to insulin allows for the quantification of age-distinct insulin granule pools in pancreatic beta cells by a combination of super resolution and transmission electron microscopy on Tokuyasu cryosections. In contrast to fluorescent proteins like GFP organic dyes covalently bound to self-labeling proteins retain their fluorescence also in epoxy resin following high pressure freezing and freeze substitution, or remarkably even after strong chemical fixation. This enables for the assessment of age-defined granule morphology and degradation. Finally, we demonstrate that this CLEM protocol is highly versatile, being suitable for single and dual fluorescent labeling and detection of different proteins with optimal ultrastructure preservation and contrast.

Le cerveau est composé de neurones et de cellules gliales qui jouent un rôle de soutien et de protection du réseau cellulaire. L’espace extra-cellulaire (ECS) correspond à l’espace qui existe entre ces cellules. Les modifications de sa structure peuvent dépendre de plusieurs paramètres comme l’âge, l’apprentissage ou les maladies neuro-dégénératives. Le volume de l’ECS correspond à environ 20$%$ du volume total du cerveau et les neurotransmetteurs et autres molécules circulent dans cet espace pour assurer une communication neuronale optimale. Cependant, les dimensions et la viscosité locale de cet espace restent encore mal-connues. L’ECS est composé entre autres de protéoglycans, de glycoaminoglycans (acide hyaluronique…) et de fluide cérébrospinal. Nous avons proposé dans cette thèse une stratégie pour mesurer les dimensions et les propriétés rhéologiques de l’espace extra-cellulaire de tranches de cerveaux de rats maintenue en vie à l’aide du suivi de nanotubes de carbone individuels luminescents. Pour ces applications, nous avons étudier la biocompatibilité et le rapport signal sur bruit de nos échantillons de nanotubes afin de les détecter en profondeur dans les tranches de cerveaux et de pouvoir mesurer leurs propriétés de diffusion. / The brain is mainly composed of neurons which ensure neuronal communication and glialcells which play a role in supporting and protecting the neural network. The extracellular space corresponds to the space that exists between all these cells and represents around 20 %of the whole brain volume. In this space, neurotransmitters and other molecules circulate into ensure optimal neuronal functioning and communication. Its complex organization whichis important to ensure proper functioning of the brain changes during aging, learning or neurodegenerative diseases. However, its local dimensions and viscosity are still poorly known.To understand these key parameters, in this thesis, we developed a strategy based on the tracking of single luminescent carbon nanotubes. We applied this strategy to measure the structural and viscous properties of the extracellular space of living rodent brains slices at the nanoscale. The organization of the manuscript is as follows. After an introduction of the photoluminescence properties of carbon nanotubes, we present the study that allowed us to select the optimal nanotube encapsulation protocol to achieve our biological applications. We also present a quantitative study describing the temperature increase of the sample when laser irradiations at different wavelengths are used to detect single nanotubes in a brain slice.Thanks to a fine analysis of the singular diffusion properties of carbon nanotubes in complex environments, we then present the strategy set up to reconstruct super-resolved maps (i.e. with resolution below the diffraction limit) of the brain extracellular space morphology.We also show that two local properties of this space can be extracted : a structural complexity parameter (tortuosity) and the fluid’s in situ viscosity seen by the nanotubes. This led us to propose a methodology allowing to model the viscosity in situ that would be seen, not by the nanotubes,but by any molecule of arbitrary sizes to simulate those intrinsically present or administered in the brain for pharmacological treatments. Finally, we present a strategy to make luminescent ultra-short carbon nanotubes that are not intrinsically luminescent and whose use could be a complementary approach to measure the local viscosity of the extracellular space of the brain.

There are many clinical situations where magnetic resonance imaging (MRI) is preferable over other imaging modalities, while the major disadvantage is the relatively long scan time. Due to limited resources, this means that not all patients can be offered an MRI scan, even though it could provide crucial information. It can even be deemed unsafe for a critically ill patient to undergo the examination. In MRI, there is a trade-off between resolution, signal-to-noise ratio (SNR) and the time spent gathering data. When time is of utmost importance, we seek other methods to increase the resolution while preserving SNR and imaging time. In this work, I have studied one of the most promising methods for this task. Namely, constructing super-resolution algorithms to learn the mapping from a low resolution image to a high resolution image using convolutional neural networks. More specifically, I constructed networks capable of transferring high frequency (HF) content, responsible for details in an image, from one kind of image to another. In this context, contrast or weight is used to describe what kind of image we look at. This work only explores the possibility of transferring HF content from T1-weighted images, which can be obtained quite quickly, to T2-weighted images, which would take much longer for similar quality. By doing so, the hope is to contribute to increased efficacy of MRI, and reduce the problems associated with the long scan times. At first, a relatively simple network was implemented to show that transferring HF content between contrasts is possible, as a proof of concept. Next, a much more complex network was proposed, to successfully increase the resolution of MR images better than the commonly used bicubic interpolation method. This is a conclusion drawn from a test where 12 participants were asked to rate the two methods (p=0.0016) Both visual comparisons and quality measures, such as PSNR and SSIM, indicate that the proposed network outperforms a similar network that only utilizes images of one contrast. This suggests that HF content was successfully transferred between images of different contrasts, which improves the reconstruction process. Thus, it could be argued that the proposed multi-contrast model could decrease scan time even further than what its single-contrast counterpart would. Hence, this way of performing multi-contrast super-resolution has the potential to increase the efficacy of MRI.

L’imagerie pondérée en diffusion (DWI) représente un outil prometteur pour augmenter la spécificité de l’IRM mammaire en vue du dépistage du cancer du sein. L’épaisseur de coupe pour une acquisition ayant un rapport signal sur bruit suffisant et couvrant les seins dans un temps compatible avec un examen clinique, reste égale ou supérieur à 3 mm, limitant la possibilité de dépistage. Dans ce travail, une méthode DWI isotrope a été développée pour obtenir des images haute résolution isotropes (1x1x1 mm3) couvrant entièrement les seins. Ces images sont obtenues en combinant : (i) une séquence à train de lecture segmenté (rs-EPI) qui correspond à plusieurs segments de lecture EPI avec écho navigation, permettant d’obtenir de hautes résolutions dans le plan, (ii) une stratégie de super-résolution (SR) consistant à acquérir trois jeux de données avec des coupes épaisses (3 mm) et des décalages de 1 mm dans le sens de coupe entre chaque acquisition et (iii) une méthode de reconstruction dédiée pour obtenir des données isotropes 1x1x1 mm3. Plusieurs schémas de reconstruction basés sur différentes régularisations ont été étudiés. La SR proposée a été comparée aux acquisitions natives de 1x1x1 mm3 sans algorithme SR sur huit sujets sains et des fantômes synthétiques. Pour valider la méthode SR, nous avons utilisé plusieurs méthodes : des simulations Monte-Carlo, des mesures de SNR et des métriques de netteté et enfin le coefficient de diffusion apparent (ADC). Ces validations ont aussi été confirmées par des mesures expérimentales sur fantômes contenant des objets de dimensions et diffusion calibrées. Un nouveau protocole de recherche clinique est proposé pour évaluer l’efficacité de la séquence de diffusion à haute résolution sur le dépistage d’un cancer mammaire, dans le but de remplacer la séquence de perfusion avec injection de produit de contraste utilisée en IRM mammaire. / Diffusion-weighted imaging (DWI) is a promising tool to increase the specificity of MRI for breast cancer screening. However, the field of view covering the breasts makes the DWI at high resolution difficult and the images obtained have low signal-to-noise ratios (SNR). The current DWI techniques are limited by the spatial resolution, mainly a slice thickness greater than or equal to 3 mm. In this work, an isotropic DWI method was developed to obtain high resolution isotropic images (1x1x1 mm3) covering the entire breast. These images are obtained by combining: (i) a readout-segmented DW-EPI sequence (rs-EPI), with several segments of k-space and echo navigator providing high in-plane resolution, (ii) a super-resolution (SR) strategy, which consists of acquiring three datasets with thick slices (3 mm) and 1mm-shifts in the slice direction, (iii) and combining them into a 1x1x1 mm3 dataset using a dedicated reconstruction. Several SR reconstruction schemes were investigated, based on different regularizations. The proposed SR strategy was compared to native 1x1x1 mm3 acquisitions (i.e. with 1 mm slice thickness) on eight healthy subjects, and synthetics phantoms. To validate the SR method, we used several methods: Monte Carlo simulations, SNR measurements and sharpness metrics, the apparent diffusion coefficient (ADC) values in normal breast tissue and breast diffusion/resolution phantom were also compared. A new clinical research protocol is proposed to evaluate the effectiveness of the high resolution diffusion sequence on breast cancer screening. The aim of this protocol is to replace the contrast-enhanced perfusion by the diffusion sequence for screening.

La super-résolution est une technique de traitement d'images qui consiste en la reconstruction d'une image hautement résolue à partir d'une ou plusieurs images bassement résolues.Cette technique est apparue dans les années 1980 pour tenter d'augmenter artificiellement la résolution des images et donc de pallier, de façon algorithmique, les limites physiques des capteurs d'images.Comme beaucoup des techniques de reconstruction en traitement d'images, la super-résolution est connue pour être un problème mal posé dont la résolution numérique est mal conditionnée. Ce mauvais conditionnement rend la qualité des images hautement résolues reconstruites très sensible au choix du modèle d'acquisition des images, et particulièrement à la modélisation de la réponse impulsionnelle de l'imageur.Dans le panorama des méthodes de super-résolution que nous dressons, nous montrons qu'aucune des méthodes proposées par la littérature ne permet de modéliser proprement le fait que la réponse impulsionnelle d'un imageur est, au mieux, connue de façon imprécise. Au mieux l'écart existant entre modèle et réalité est modélisé par une variable aléatoire, alors que ce biais est systématique.Nous proposons de modéliser l'imprécision de la connaissance de la réponse impulsionnelle par un ensemble convexe de réponses impulsionnelles. L'utilisation d'un tel modèle remet en question les techniques de résolution. Nous proposons d'adapter une des techniques classiques les plus populaires, connue sous le nom de rétro-projection itérative, à cette représentation imprécise.L'image super-résolue reconstruite est de nature intervalliste, c'est à dire que la valeur associée à chaque pixel est un intervalle réel. Cette reconstruction s'avère robuste à la modélisation de la réponse impulsionnelle ainsi qu'à d'autres défauts. Il s'avère aussi que la largeur des intervalles obtenus permet de quantifier l'erreur de reconstruction. / Super-resolution is an image processing technique that involves reconstructing a high resolution image based on one or several low resolution images. This technique appeared in the 1980's in an attempt to artificially increase image resolution and therefore to overcome, algorithmically, the physical limits of an imager.Like many reconstruction problems in image processing, super-resolution is known as an ill-posed problem whose numerical resolution is ill-conditioned. This ill-conditioning makes high resolution image reconstruction qualityvery sensitive to the choice of image acquisition model, particularly to the model of the imager Point Spread Function (PSF).In the panorama of super-resolution methods that we draw, we show that none of the methods proposed in the relevant literature allows properly modeling the fact that the imager PSF is, at best, imprecisely known. At best the deviation between model and reality is considered as being a random variable, while it is not: the bias is systematic.We propose to model scant knowledge on the imager's PSF by a convex set of PSFs. The use of such a model challenges the classical inversion methods. We propose to adapt one of the most popular super-resolution methods, known under the name of "iterative back-projection", to this imprecise representation. The super-resolved image reconstructed by the proposed method is interval-valued, i.e. the value associated to each pixel is a real interval. This reconstruction turns out to be robust to the PSF model and to some other errors. It also turns out that the width of the obtained intervals quantifies the reconstruction error.

The mammalian hippo signalling pathway controls cell proliferation and apoptosis via transcriptional co-activators YAP and TAZ, and as such is a key regulator of organ and tissue growth. Multiple cellular components converge in this pathway, including the actin cytoskeleton, which is required for YAP/TAZ activity. The precise mechanism by which the mechanical actomyosin network regulates Hippo signalling, however, is unknown. Optical methods provide a non-invasive way to image and study the biomechanics of cells. In the past two decades, super-resolution fluorescence microscopy techniques that break the diffraction limit of light have come to the fore, enabling visualisation of intracellular detail at the nanoscale level. Optical trapping, on the other hand, allows precise control of micron-sized objects such as cells. Here, super resolution structured illumination microscopy (SIM) and elastic resonator interference stress microscopy (ERISM) were used to investigate a potential role for the FERM-domain protein FRMD6, or Willin, in the mechanical control of the Hippo pathway in a neuronal cell model. A double optical trap was also integrated with the Nikon-SIM with the aim of cell stretching. Willin expression was shown to modify the morphology, neuronal differentiation, actin cytoskeleton and forces of SH-SY5Y cells. Optical trapping from above the SIM objective, however, was demonstrated to be ineffective for manipulation of adherent cells. The results presented here indicate a function for Willin in the assembly of actin stress fibres that may be the result of an interaction with the Hippo pathway regulator AMOT. Further investigation, for example by direct cell stretching, is required to elucidate the exact role of Willin in the mechanical control of YAP/TAZ.

Dans cette thèse, nous avons construit et optimisé des méthodes de microscopie de fluorescence super-résolue stochastique, polarisée et quantitative qui nous permettent d'imager l'orientation moléculaire dans des environnements dynamiques et statiques a l’échelle de la molécule unique et avec une résolution nanoscopique. En utilisant un montage de microscopie super-résolue à lecture stochastique en combinaison avec une détection polarisée, nous avons pu reconstruire des images d'anisotropie de fluorescence avec une résolution spatiale de 40 nm. En particulier, nous avons pu imager l'ordre orientationnel d'assemblages biomoléculaires et cellulaires. Pour l'imagerie cellulaire, nous avons pu étudier la capacité d'étiquettes de marquer fluorophoresde reporter quantifier l'orientation moléculaire dans l'actine et les microtubules dans des cellules fixées. Nous avons également mis à profit la meilleure résolution et la détection polarisée pour étudier l'ordre moléculaire d’agrégats d’amyloïdes a l’échelle nanoscopique. Enfin, nous avons étudié l'interaction de la protéine de réparation RAD51 avec l'ADN par microscopie de fluorescence polarisée super-résolue pour quantifier l'ordre orientationnel de l'ADN et de la protéine RAD51 afin de comprendre la recombinaison homologue du mécanisme de réparation de l'ADN. / .In this thesis we built and optimized quantitative polarized stochastic super-resolution fluorescence microscopy techniques that enabled us to image molecular orientation behaviors in static and dynamic environments at single molecule level and with nano-scale resolution. Using a scheme of stochastic read-out super resolution microscopy in combination with polarized detection, we can reconstruct fluorescence anisotropy images at a spatial resolution of 40 nm. In particular, we have been able to use the techniques to quantify the molecular orientationalorder in cellular and bio-molecular assemblies. For cellular imaging, we could quantify the ability of fluorophore labels to report molecular orientation of actin and microtubules in fixed cells. Furthermore, we used the improvements of resolution and polarization detection to study molecular order of amyloid aggregates at a nanoscopic scale. Also, we studied repair protein RAD51` s interaction with DNA by using dual color polarized fluorescence microscopy, to quantify the orientational order of DNA and RAD51 to understand the homologous recombination of DNA repair mechanism.

La microscopie à éclairements structurés(structured illumination microscopy, SIM) permet de dépasser la limite de résolution en microscopie optique due à la diffraction, en éclairant l’objet avec un ensemble de motifs périodiques parfaitement connus. Cependant, il s’avère difficile de contrôler exactement la forme des motifs éclairants. Qui plus est, de fortes distorsions de la grille de lumière peuvent être générées par l’échantillon lui-même dans le volume d’étude, ce qui peut provoquer de forts artefacts dans les images reconstruites. Récemment, des approches dites blind-SIM ont été proposées, où les images sont acquises à partir de motifs d’éclairement inconnus, non-périodiques, de type speckle,bien plus faciles à générer en pratique. Le pouvoir de super résolution de ces méthodes a été observé, sans forcément être bien compris théoriquement. Cette thèse présente deux nouvelles méthodes de reconstruction en microscopie à éclairements structurés inconnus (blind speckle-SIM) : une approche conjointe et une approche marginale. Dans l’approche conjointe, nous estimons conjointement l’objet et les motifs d’éclairement au moyen d’un modèle de type Basis Pursuit DeNoising (BPDN) avec une régularisation en norme lp,q où p=>1 et 0<q<=1. La norme lp,q est introduite afin de prendre en compte une hypothèse de parcimonie sur l’objet. Dans l’approche marginale, nous reconstruisons uniquement l’objet et les motifs d’éclairement sont traités comme des paramètres de nuisance. Notre contribution est double. Premièrement, une analyse théorique démontre que l’exploitation des statistiques d’ordre deux des données permet d’accéder à un facteur de super résolution de deux, lorsque le support de la densité spectrale du speckle correspond au support fréquentiel de la fonction de transfert du microscope. Ensuite, nous abordons le problème du calcul numérique de la solution. Afin de réduire à la fois le coût de calcul et les ressources en mémoire, nous proposons un estimateur marginal à base de patches. L’élément clé de cette méthode à patches est de négliger l’information de corrélation entre les pixels appartenant à différents patches. Des résultats de simulations et en application à des données réelles démontrent la capacité de super résolution de nos méthodes. De plus, celles-ci peuvent être appliquées aussi bien sur des problèmes de reconstruction 2D d’échantillons fins, mais également sur des problèmes d’imagerie 3D d’objets plus épais. / Conventional structured illumination microscopy (SIM) can surpass the resolution limit inoptical microscopy caused by the diffraction effect, through illuminating the object with a set of perfectly known harmonic patterns. However, controlling the illumination patterns is a difficult task. Even worse, strongdistortions of the light grid can be induced by the sample within the investigated volume, which may give rise to strong artifacts in SIM reconstructed images. Recently, blind-SIM strategies were proposed, whereimages are acquired through unknown, non-harmonic,speckle illumination patterns, which are much easier to generate in practice. The super-resolution capacity of such approaches was observed, although it was not well understood theoretically. This thesis presents two new reconstruction methods in SIM using unknown speckle patterns (blind-speckle-SIM): one joint reconstruction approach and one marginal reconstruction approach. In the joint reconstruction approach, we estimate the object and the speckle patterns together by considering a basis pursuit denoising (BPDN) model with lp,q-norm regularization, with p=>1 and 0<q<=1. The lp,q-norm is introduced based on the sparsity assumption of the object. In the marginal approach, we only reconstruct the object, while the unknown speckle patterns are considered as nuisance parameters. Our contribution is two fold. First, a theoretical analysis demonstrates that using the second order statistics of the data, blind-speckle-SIM yields a super-resolution factor of two, provided that the support of the speckle spectral density equals the frequency support of the microscope point spread function. Then, numerical implementation is addressed. In order to reduce the computational burden and the memory requirement of the marginal approach, a patch-based marginal estimator is proposed. The key idea behind the patch-based estimator consists of neglecting the correlation information between pixels from different patches. Simulation results and experiments with real data demonstrate the super-resolution capacity of our methods. Moreover, our proposed methods can not only be applied in 2D super-resolution problems with thin samples, but are also compatible with 3D imaging problems of thick samples.