Les épines dendritiques sont de petites protrusions post-synaptiques présentant des changements morphologiques corrélés avec la plasticité synaptique. Elles ont pour origine les filopodes dendritiques qui s’élargissent lors du contact avec l’axone. Ces changements morphologiques impliquent une grande variété de molécules dont des protéines associées à l’actine et des protéines d’adhésion. Cependant, comment ces différentes protéines sont coordonnées dans le temps et l’espace est encore largement méconnu. De plus, les techniques de microscopie conventionnelle ne permettent pas d’étudier l’organisation et la dynamique de ces protéines dans les épines dont la taille est proche de la limite de resolution. L’objectif de ma thèse a donc été d’explorer le rôle des protéines associées à l’actine ainsi que celui des protéines d’adhésion N-cadhérines dans l’organisation et la dynamique du cytosquelette d’actine des épines dendritiques. Dans une première étude, nous avons suivi la motilité des filopodes et épines dendritiques de neurones en visualisant l’actine-GFP. Nous avons couplé cette approche avec : 1) une technique de piégeage optique de microsphères recouvertes de N-cadhérines ou des substrats micro-imprimés également recouverts de N-cadhérines afin de contrôler temporellement et spatialement les adhésions cadhérine-cadhérine, 2) la stimulation pharmacologique de la myosine II afin d’induire la contraction F-actine/myosine et 3) l’expression de mutants de N-cadhérine non adhésifs. Nous avons ainsi démontré que la stabilisation des filopodes en épines était dépendante de l’engagement d’un embrayage moléculaire entre les adhésions trans-synaptiques N-cadhérine et le flux rétrograde d’actine généré par les myosines II. Dans une deuxième étude, nous avons utilisé la microscopie super-résolutive (PALM et dSTORM) et le suivi de protéines individuelles (sptPALM) pour étudier l’organisation et la dynamique à l’échelle nanométrique des protéines à l’origine des réseaux d’actine branchés dans les épines. Ainsi, nous avons caractérisé la localisation et la dynamique de l’actine, du complexe Arp2/3, du complexe WAVE, d’IRSp53, de VASP et de Rac-1. Nous avons montré que, contrairement aux structures motiles classiques comme lamellipode, le réseau d’actine branché dans les épines n’ést pas formé aux extrémités protrusives puis incorporé dans un flux rétrograde d’actine. Ce réseau est initié à la PSD puis croît vers l’extérieur afin de générer les protrusions membranaires responsablent des changements morphologiques de l’épine. Nos résultats montrent également qu’un contrôle strict de l’activité de Rac-1 est nécessaire au maintien de la morphologie des épines dendritiques et de l’architecture du réseau d’actine branché. L’ensemble de mon travail souligne l’importance du rôle de l’organisation à l’échelle nanométrique du réseau d’actine branché et des adhésions N-cadhérine dans la dynamique et la formation des épines dendritiques. Ces résultats pourraient avoir un rôle important dans la compréhension des changements morphologiques lors de la plasticité synaptique. / Dendritic spines are tiny post-synaptic protrusions exhibiting changes in morphology correlated with synaptic plasticity. They originate from motile dendritic filopodia, which enlarge after contacting axons. These morphological changes involve a wide number of molecular actors, including actin-binding proteins, and adhesion molecules. However, how these various molecular components are coordinated temporally and spatially to tune changes in spine shape remains unclear. Furthermore, conventional photonic microscopy techniques could not achieved the spatial resolution required to study the dynamic nanoscale organization of these proteins within the micron size dendritic spines. The objective of my Ph.D. was to unravel how actin-binding proteins and N-cadherin adhesion regulate the organization and dynamics of F-actin network in dendritic spines. In a first study, we measured the motility of dendritic filopodia and spines by time lapse imaging of actin-GFP in primary hippocampal neurons. We combined those measurements with: 1) manipulation of N-cadherin coated beads with optical tweezers, or micropatterns to control the timing and location of nascent N-cadherin adhesions, 2) pharmacological stimulation of myosin II to trigger contraction of the F-actin/myosin network and 3) expression of non-adhesive N-cadherin mutants to compete for the interaction between N-cadherin adhesion and F-actin. Using these different approaches we demonstrated that the stabilization of dendritic filopodia into mature spines was dependent on the engagement of a molecular clutch between trans-synaptic N-cadherin adhesions and the myosin driven F-actin flow. In a second study, we used super resolution microscopy (PALM and dSTORM) and single protein tracking (sptPALM) to study the dynamic nanoscale organizations of branched actin networks within dendritic spines. Using these technics, we characterized within dendritic spines, the localization and dynamics of actin, Arp2/3 complex, WAVE complex, IRSp53, VASP and Rac-1. We established that, opposite to classical motile structures such as the lamellipodium, branched F-actin networks in dendritic spines are not formed at the tip of membrane protrusions and incorporated in a retrograde flow. On the contrary, they are growing outwards from the PSD generating membrane protrusions responsible for spine motility. We also show that a thigh control of Rac1 activity is required to maintain dendritic spine morphology and branched actin network organization. Altogether, these studies point out the role of the nanoscale functional organization of F-actin networks and its linkage to adhesion proteins in the regulation of dendritic spine formation and dynamics. These findings may have important implications in the understanding of spine morphology changes driven by synaptic activity.

<p> The urgency for deeper knowledge about nervous system function and dysfunction has never been greater. With increasing rates of mental disorders and expanding healthcare costs, deciphering details of axonal development is essential to meeting this imperative. Models of neuronal growth are improving as roles of microtubules and motor proteins surface. However, traditional motor protein studies focus on intracellular cargo transport, leaving deficits in knowledge about how these proteins organize cytoskeletal filaments in the axon and growth cone during neuronal development. Inconsistent findings on microtubule activity in growing axons also leave gaps in quantitative assessments of microtubule translocation and assembly, limiting the ability to construct a comprehensive model of axonal elongation.</p> The goal of this study was to provide a more complete neuronal growth cone model by determining how individual microtubule translocation and assembly, mass microtubule movements, and motor protein activity contribute to axonal elongation. The underlying mechanisms of these processes were investigated by testing the roles of dynein and microtubule dynamics in axonal elongation of <i>Aplysia </i><i>californica </i>neurons using transillumination, fluorescent speckle, and super-resolution microscopy. Pharmacologically inhibiting either dynein activity or microtubule assembly reduced both bulk and individual microtubule anterograde translocation and neurite elongation rates. Suppressing both processes simultaneously had compensatory rather than additive effects. Super-resolution imaging also revealed fewer dynein motors co-localized with microtubules when microtubule assembly was inhibited. These results strongly suggest that disrupting microtubule assembly blocks neurite outgrowth partly because it inhibits dynein-mediated bulk microtubule translocation.

Computational methods in single molecule localization microscopy Abstract Fluorescence microscopy is one of the chief tools used in biomedical research as it is a non invasive, non destructive, and highly specific imaging method. Unfortunately, an optical microscope is a diffraction limited system. Maximum achievable spatial resolution is approximately 250 nm laterally and 500 nm axially. Since most of the structures in cells researchers are interested in are smaller than that, increasing resolution is of prime importance. In recent years, several methods for imaging beyond the diffraction barrier have been developed. One of them is single molecule localization microscopy, a powerful method reported to resolve details as small as 5 nm. This approach to fluorescence microscopy is very computationally intensive. Developing methods to analyze single molecule data and to obtain super-resolution images are the topics of this thesis. In localization microscopy, a super-resolution image is reconstructed from a long sequence of conventional images of sparsely distributed single photoswitchable molecules that need to be sys- tematically localized with sub-diffraction precision. We designed, implemented, and experimentally verified a set of methods for automated processing, analysis and visualization of data acquired...

No description available.

No description available.

The accurate assessment of hemodynamics and its parameters play an important role when diagnosing cardiovascular diseases. In this context, 4D Flow Magnetic Resonance Imaging (4D Flow MRI) is a non-invasive measurement technique that facilitates hemodynamic parameter assessment as well as quantitative and qualitative analysis of three-directional flow over time. However, the assessment is limited by noise, low spatio-temporal resolution and long acquisition times. Consequently, in regions characterized by transient, rapid flow dynamics, such as the aorta and heart, capturing these rapid transient flows remains particularly challenging. Recent research has shown the feasibility of machine learning models to effectively denoise and increase the spatio-temporal resolution of 4D Flow MRI. However, temporal super-resolution networks, which can generalize on unseen domains and are independent on boundary segmentations, remain unexplored.  This study aims to investigate the feasibility of a neural network for temporal super-resolution and denoising of 4D Flow MRI data. To achieve this, we propose a residual convolutional neural network (based on the 4DFlowNet from Ferdian et al.) providing an end-to-end mapping from temporal low resolution space to high resolution space. The network is trained on patient-specific cardiac models created with computational-fluid dynamic (CFD) simulations covering a full cardiac cycle. For clinical contextualization, performance is assessed on clinical patient data. The study shows the potential of the 4DFlowNet for temporal-super resolution with an average relative error of 16.6 % on an unseen cardiac domain, outperforming deterministic methods such as linear and cubic interpolation. We find that the network effectively reduces noise and recovers high-transient flow by a factor of 2 on both in-silico and in-vivo cardiac datasets. The prediction results in a temporal resolution of 20 ms, going beyond the general clinical routine of 30-40 ms. This study exemplifies the performance of a residual CNN for temporal super-resolution of 4D flow MRI data, providing an option to extend evaluations to aortic geometries and to further develop different upsampling factors and temporal resolutions. / En noggrann bedömning av hemodynamiken och dess parametrar spelar en viktig roll vid diagnos av kardiovaskulära sjukdomar. I detta sammanhang är 4D Flow Magnetic Resonance Imaging (4D Flow MRI) en icke-invasiv mätteknik som underlättar bedömning av hemodynamiska parametrar samt kvantitativ och kvalitativ analys av flöde. Bedömningen begränsas av brus, låg spatio-temporal upplösning och långa insamlingstider. I områden som karakteriseras av snabb flödesdynamik, såsom aorta och hjärta, är det därför fortfarande särskilt svårt att fånga dessa snabba transienta flöden. Ny forskning har visat att det är möjligt att använda maskininlärningsmodeller för att effektivt reducera brus och öka den spatio-temporala upplösningen i 4D Flow MRI. Nätverk för temporal superupplösning, som kan generaliseras till osedda domäner och är oberoende av segmentering, är fortfarande outforskade.  Denna studie syftar till att undersöka genomförbarheten av ett neuralt nätverk för temporal superupplösning och brusreducering av 4D Flow MRI-data. För att uppnå detta föreslår vi ett residual faltningsneuralt nätverk (baserat på 4DFlowNet från Ferdian et al.) som tillhandahåller en end-to-end-mappning från temporalt lågupplöst utrymme till högupplöst utrymme. Nätverket tränas på patientspecifika hjärtmodeller som skapats med CFD-simuleringar som spänner över en hel hjärtcykel. För klinisk kontextualisering utvärderas nätverkets prestanda på kliniska patientdata. Studien visar potentialen av 4DFlowNet för temporal superupplösning med ett genomsnittligt relativt fel på 16,6 % på en osedd hjärtdomän, vilket överträffar deterministiska metoder som linjär och kubisk interpolation. Vi konstaterar att nätverket effektivt minskar brus och återställer högtransient flöde med en faktor på 2 på både in-silico ochin-vivo hjärtdataset. Förutsägelsen resulterar i en temporal upplösning på 20 ms, vilket är mer än den allmänna kliniska rutinen på 30-40 ms. Denna studie exemplifierar prestandan hos en residual CNN för temporal superupplösning av 4D-flödes-MRI-data, vilket ger möjlighet att utvidga utvärderingarna till aortageometrier och att vidareutveckla olika uppsamplingsfaktorer och temporala upplösningar.

In synthetic aperture radar (SAR) and inverse synthetic aperture radar (ISAR), an imaging radar emits electromagnetic waves of varying frequencies towards a target and the backscattered waves are collected. By either moving the radar antenna or rotating the target and combining the collected waves, a much longer synthetic aperture can be created. These radar measurements can be used to determine the radar cross-section (RCS) of the target and to reconstruct an estimate of the target. However, the reconstructed images will suffer from spectral leakage effects and are limited in resolution. Many methods of enhancing the images exist and some are based on deep learning. Most commonly the deep learning methods rely on high-resolution ground truth data of the scene to train a neural network to enhance the radar images. In this thesis, a method that does not rely on any high-resolution ground truth data is applied to train a convolutional neural network to enhance radar images. The network takes a conventional ISAR image subject to spectral leakage effects as input and outputs an enhanced ISAR image which contains much more defined features. New RCS measurements are created from the enhanced ISAR image and the network is trained to minimise the difference between the original RCS measurements and the new RCS measurements. A sparsity constraint is added to ensure that the proposed enhanced ISAR image is sparse. The synthetic training data consists of scenes containing point scatterers that are either individual or grouped together to form shapes. The scenes are used to create synthetic radar measurements which are then used to reconstruct ISAR images of the scenes. The network is tested using both synthetic data and measurement data from a cylinder and two aeroplane models. The network manages to minimise spectral leakage and increase the resolution of the ISAR images created from both synthetic and measured RCSs, especially on measured data from target models which have similar features to the synthetic training data.  The contributions of this thesis work are firstly a convolutional neural network that enhances ISAR images affected by spectral leakage. The neural network handles complex-valued signals as a single channel and does not perform any rescaling of the input. Secondly, it is shown that it is sufficient to calculate the new RCS for much fewer frequency samples and angular positions and compare those measurements to the corresponding frequency samples and angular positions in the original RCS to train the neural network.

Super-resolution is a subfield of computer vision centered around upsampling low-resolution images to a corresponding high-resolution counterpart. This degree project investigates the suitability of a deep learning method for real-time video super-resolution. Following earlier work in the field, we use bicubic interpolation as a baseline for comparison. The deep learning method selected is specifically suited towards real-time super-resolution and consists of a motion compensation network and an upsampling network. The deep learning method and bicubic interpolation are compared by quantitatively evaluating the methods against each other in quality metrics and performance metrics. Suitable quality metrics are selected from earlier works to provide increased comparability of results, namely peak signal-to-noise ratio and structure similarity index. The performance metrics are: number of operations for a single upsampled frame, latency, throughput, and memory requirements. We apply the methods to a highly challenging publicly available dataset specifically engineered towards video super-resolution research. To further investigate the deep learning method, we propose a few modifications and study the effect on the metrics. Our findings show that the deep learning models outperform bicubic interpolation in the quality metrics, while bicubic interpolation outperformed the deep learning models in the performance metrics. We also find no significant quality metric improvement associated with having a motion compensation network for this dataset, suggesting that the dataset might be too complex for the motion compensation network. We conclude that the deep learning method exhibits real-time capabilities as the method has a throughput of around 500 frames per second for full HD super-resolution. Additionally, we show that by modifying the deep learning method, we achieve similar latency as bicubic interpolation without sacrificing throughput or quality. / Superupplösning är ett område inom datorseende centrerat kring att uppsampla lågupplösta bilder till högupplösta motsvarigheter. Detta examensarbete undersöker hur lämplig en specifik djupinlärningsmetod är för superupplösning i realtid. Enligt tidigare forskning använder vi oss av bikubisk interpolering som grund för jämförelse. Den valda djupinlärningsmetoden är speciellt anpassad till superupplösning i realtid och består av ett rörelsekompensationsnätverk och ett uppsamplingsnätverk. Djupainlärningsmetoden och interpoleringsmetoden jämförs genom att kvantitativt utvärdera metoderna mot varandra i kvalitetsmått och prestandamått. Lämpliga kvalitetsmått väljs från tidigare forskning för att ge ökad jämförbarhet mellan resultaten, nämligen maximalt signaltill- brusförhållande och strukturlikhetsindex. Prestandamätvärdena är: antal operationer för en uppsamplad bild, latens, genomströmning och minnesbehov. Vi utvärderar metoderna på ett utmanande allmänt tillgängligt dataset speciellt konstruerat för superupplösningsforskning inom video. För att ytterligare undersöka den djupa inlärningsmetoden föreslår vi några modifieringar och studerar effekten på mätvärdena. Våra resultat visar att djupinlärningsmodellerna överträffar bikubisk interpolering i kvalitetsmåtten, medan bikubisk interpolering överträffar djupinlärningsmodellerna i prestandamåtten. Vi finner inte heller någon signifikant kvalitetsmässig förbättring förknippad med att ha ett rörelsekompensationsnätverk för detta dataset, vilket kan betyda att datasetet är för komplext för rörelsekompensationnätverket. Vi drar slutsatsen att djupainlärningsmetoden uppvisar realtidsfunktioner eftersom metoden har en genomströmning på cirka 500 bilder per sekund för full HD superupplösning. Dessutom visar vi att genom att modifiera djupainlärningsmetoden uppnår vi liknande latens som bikubisk interpolering utan att offra genomströmning eller kvalitet.

Recent years have witnessed tremendous success in using deep learning approaches to handle low-level vision problems. Most of the deep learning based methods address the low-level vision problem by training a neural network to approximate the mapping from the inputs to the desired ground truths. However, directly learning this mapping is usually difficult and cannot achieve ideal performance. Besides, under the setting of unsupervised learning, the general deep learning approach cannot be used. In this thesis, we investigate and address several problems in low-level vision using the proposed approaches. To learn a better mapping using the existing data, an indirect domain shift mechanism is proposed to add explicit constraints inside the neural network for single image dehazing. This allows the neural network to be optimized across several identified neighbours, resulting in a better performance. Despite the success of the proposed approaches in learning an improved mapping from the inputs to the targets, three problems of unsupervised learning is also investigated. For unsupervised monocular depth estimation, a teacher-student network is introduced to strategically integrate both supervised and unsupervised learning benefits. The teacher network is formed by learning under the binocular depth estimation setting, and the student network is constructed as the primary network for monocular depth estimation. In observing that the performance of the teacher network is far better than that of the student network, a knowledge distillation approach is proposed to help improve the mapping learned by the student. For single image dehazing, the current network cannot handle different types of haze patterns as it is trained on a particular dataset. The problem is formulated as a multi-domain dehazing problem. To address this issue, a test-time training approach is proposed to leverage a helper network in assisting the dehazing network adapting to a particular domain using self-supervision. In lossy compression system, the target distribution can be different from that of the source and ground truths are not available for reference. Thus, the objective is to transform the source to target under a rate constraint, which generalizes the optimal transport. To address this problem, theoretical analyses on the trade-off between compression rate and minimal achievable distortion are studied under the cases with and without common randomness. A deep learning approach is also developed using our theoretical results for addressing super-resolution and denoising tasks. Extensive experiments and analyses have been conducted to prove the effectiveness of the proposed deep learning based methods in handling the problems in low-level vision. / Thesis / Doctor of Philosophy (PhD)

No description available.