|
|
AbstractsRobert HaaseScaDS.AI, Leipzig University Title: Large Language Models for Bio-image Analysis Abstract: Generative artificial intelligence, and large language models (LLMs) in particular, are changing the way we use computers. Recent development in this field has impact on literally all research fields. In this talk we learn how LLMs can be used to facilitate microscopy bio-image analysis. We get an insight into how the technology works and what its current limitations are.
Guillaume Jacquemet Åbo, Akademi University, Turku, Finland, Title: Studying cancer cell metastasis in the era of deep learning for microscopy Abstract: To disseminate, cancer cells use the vascular system in which they must survive and escape by attaching to and crossing the vascular wall. However, how cancer cells arrest and cross the vasculature remains poorly understood. We set up a microfluidic system to perfuse human pancreatic cancer cells (PDAC) on top of human endothelial cells. This system enabled us to capture the dynamics of cancer cell extravasation with unprecedented detail through live imaging. To analyze these complex datasets, we developed and utilized a suite of open-source image analysis software. Fast4DReg, a Fiji plugin, corrects time-lapse imaging drift. ZeroCostDL4Mic and DL4MicEverywhere harness deep learning for artificial labeling and segmentation. TrackMate v7 offers enhanced cell tracking capabilities by incorporating advanced segmentation algorithms. CellTracksColab facilitates the compilation and analysis of tracking data, providing insights into cell behavior during extravasation. We report that PDAC cells arrest nearby endothelial cell-cell junctions. PDAC cells then use long filopodia protrusions to probe for gaps. When gaps are found, the extravasation process starts, and PDAC cells progressively traverse the endothelial layer between endothelial cells. We are currently targeting filopodia regulators to assess their contribution to the extravasation process. By combining advanced imaging strategies and image analysis software development, we are advancing our understanding of cancer cell extravasation. Anca-Ioana GrapaSarcoma Molecular Pathology Group, Institute of Cancer Research, London, UK Title:Prognostic Quantitative Biomarkers in Histopathology Images: Exploring Tumor and Microenvironmental Spatial Heterogeneity Abstract: The quantitative assessment of tumor and microenvironmental (TME) spatial heterogeneity within histological diagnostic samples can provide valuable insights into patient survival and the risk of disease relapse. We illustrate this concept in two studies by leveraging various artificial intelligence (AI)/machine learning and spatial analysis methods. First, lung adenocarcinoma (LUAD), the most common histologic variant of lung cancer, exhibits morphological diversity characterized by the presence of multiple histological patterns. A deep learning-based segmentation method was first developed to delineate and map existing growth patterns in diagnostic H&E slides. Subsequently, novel graph-based ecological metrics assessing pattern intermixing diversity were shown to be prognostic of disease-free survival in a large retrospective cohort of 834 patients. In a different study focusing on rhabdomyosarcoma (RMS), an aggressive soft tissue paediatric sarcoma, we investigated the relevance of tumor vasculature for patient stratification. We developed several AI-based methods to map and classify endothelial cells on H&E slides. Our analysis revealed that vessel density was prognostic in 181 RMS patient samples. These findings were further extended through a spatial analysis of the geographical distribution of vessels in relation to tumor hypoxia regions, contributing to a better understanding of the impact of angiogenesis for patient outcome. Overall, the proposed frameworks demonstrate the importance of developing quantitative biomarkers to assess tumor and TME spatial diversity in clinical and biological investigations related to patient stratification and disease relapse risk assessment.
Thomas BoudierMorpheme, Sophia Antipolis, France Title: Data Organization and Automation for Image Analysis Abstract: The last decade saw the development of both new microscopy imaging techniques and new Deep-Learning methods, leading to an increasing production of imaging and labelled data. The data generated are numerous and large, they hence need to be stored and organized to facilitate the automation of analysis pipelines. In this talk, we briefly review existing tools for image analysis automation, and present our approach, named TAPAS, for data organization and automation focusing on multidimensional data. We will give examples of applications in different biological domains such as immunology, neurobiology or cell biology.
Jenifer Croce, Sébastien Schaub, [Eric Debreuve]LBDV/Morpheme, Villefranche-sur-Mer, France Title: Hyperspectral imaging: separating the “good” from the “bad” in a fluorescence image Abstract: Fluorescence imaging is a powerful, highly sensitive, and non-invasive technique widely employed in various life science fields. Despite many technological advancements, fluorescence imaging of marine samples (animals or plants) remains challenging, due to the inevitable and prevalent endogenous fluorescence, known as autofluorescence, in these samples. Indeed, many marine organisms exhibit autofluorescence due to the presence of fluorescent proteins, such as GFP or DsRed, originally discovered in marine organisms. This thereby complicates the choice of exogenous fluorescent probes or fluorophores to be used in experimental approaches such as immunohistochemistry or expression of fluorescently tagged proteins. Even if genetic approaches to eliminate autofluorescence are now conceivable, they remain species-specific, highly difficult to develop, time-consuming, costly, and personnel-intensive. To overcome the issue represented by autofluorescence, in any given organism (animal or plant, marine or terrestrial), we thus aim at developing a new methodology based on hyperspectral imaging. In this context our current objectives are to 1) characterize the autofluorescence of some marine organisms we commonly use during our experimental work, 2) separate by hyperspectral imaging endogenous and exogenous fluorophore signals, and 3) build our own fluorescence-related database to support researchers choosing fluorescent probes and fluorophores. For now, we focus, as a starting point, on the sea urchin Paracentrotus lividus and the three microalgae sea urchin larvae are fed with. Each alga exhibits an autofluorescence that varies both in terms of spectrum and localization. We will show how our method already provides some clues to optimize the experimental work conducted on sea urchin larvae and how we plan to improve our method currently following the Matching pursuit approach.
Kevin LEBRIGANDHead IPMC Bioinformatics Hub Title: Imaging-based Spatial Transcriptomics Short summary: What we can expect from commercial systems (Mercope, Xenium, Cosmx) that are about to revolutionise the field of imaging-based spatial transcriptomics, allowing 1000's of genes to be profiled simultaneously at sub-cellular resolution (100nm) in large tissue section (1-3 cm2). Frédéric Brau and Sylvain FeliciangeliIPMC, Valbonne, France Title: Study of Intracellular pH with Fluorescence Lifetime IMaging (FLIM) Abstract: At the cellular level, the proper functioning of intracellular organelles critically depends on the control of homeostasis and the regulation of their pH. Despite their importance, the actors and mechanisms involved in these processes remain largely poorly understood, in part because of the difficulty to measure pH in organelles.To address this question, different techniques were developed based on the property of fluorescent dyes or proteins to exhibit changes in fluorescence intensity upon changes in pH, such as ratiometric probes. However, these techniques have a major technical limitation: as the fluorescence signal decreases, the signal-to-noise ratio increases and hence the measurement uncertainty.To overcome this limitation, we took advantage of another property of certain fluorescent proteins, in which changes in pH induce a modulation of the lifetime at the electronic excited state (fluorescence lifetime). This lifetime can be measured in cellulo and imaged by Fluorescence Lifetime Imaging Microscopy (FLIM) techniques. We characterized the pH sensitivity of fluorescent proteins and developed different probes to target specific compartments. Using these tools, we were able to evaluate the pH of various cellular compartments. We could characterize the effect of ion channels expression and establish the role of a so far uncharacterized channel as a modulator of lysosomal pH.In our talk we will present our approach with different targeted proteins and the way data could be analyzed, including new representations of FLIM data and analysis.
Maximilian Fürthauer
iBV,Nice, France Title: A live imaging-based screen of the Zebrafish Extracellular Vesicle/Particle secretome Abstract: Extracellular Vesicles (EVs) have emerged as vectors of biological information that control numerous aspects of cancer biology. In addition to EVs, non-membranous Extracellular Particles (EPs) can transport cancer-related cargo and promote tumour growth. The precise roles of EVPs remain however poorly understood, notably due to technical limitations that hamper the analysis of their functions in vivo. Our interdisciplinary consortium is taking advantage of the unique transparency of the embryonic zebrafish to establish tools for the in vivo imaging and computational analysis of EVP secretion and transport. To this aim, the Fürthauer lab has started to use high speed live imaging to characterize EVP behaviour in various organs (heart & blood vessels, brain ventricles & cerebrospinal canal…) during the first 32 hours of zebrafish development. Of particular interest, this approach provides evidence that particular organs secrete only certain types of EVPs. An interdisciplinary collaboration with the group of Xavier Descombes whose expertise lies in the computational analysis of image-based biological datasets, has moreover allowed to generate quantitative analysis pipelines to study EVP transport dynamics. Taking advantage of these approaches, we aim to perform a first imaging-based small-scale screen of the EVP secretome in an intact living vertebrate. Bianca Silva
IPMC, Valbonne, France Title : Whole-brain activity mapping by multiplexed immediate early genes analysis
Abstract: Understanding brain-wide neural activity patterns is essential for unraveling the network processes underlying behavior. While various techniques offer different spatial and temporal resolutions, none provide a perfect solution for large-scale interrogations. Immediate early gene (IEG) expression emerges as a valuable tool for mapping neural activity across the brain. Despite limitations in temporal resolution, IEGs offer versatility and accessibility, making them ideal for investigating brain-wide networks. In this study, we developed ABBA, a novel automated pipeline for brain section alignment and quantification. Leveraging ABBA, we compared the induction patterns of three widely used IEGs, cFos, Arc, and NPAS4, in mice across different behavioral conditions. Our results revealed substantial variations in IEG expression levels across brain regions. Interestingly, different IEGs reflected different behavioural conditions in different areas. In light of this, we propose a new activity mapping pipeline based on the multiplexed imaging of all three IEGs. ABBA provides a scalable solution for brain-wide quantification of 2D sections, offering flexibility, ease of use, and reliable results. Integrated seamlessly with existing bio-imaging software, ABBA streamlines the alignment process, facilitating efficient and accurate outcomes for neuroscience research.
Frédéric Luton, Xavier DescombesIPMC/Morpheme, Valbonne/Sophia Antipolis Title: Characterization of the basement membrane architecture through mathematical analyses of 3D images Abstract: In epithelial cancers, the transition from benign to invasive stages corresponds to the infiltration of the basement membrane, a multiprotein complex that forms a porous sheet surrounding the epithelial tissue. Using confocal microscopy, we have established a protocol to image the architecture of the basement membrane assembled around normal or invasive human mammary cell spheroids cultured in 3D matrices. We are developing a computer program to automatically and quantitatively analyze the three-dimensional reconstruction images of the basement membrane to determine its geometric parameters and extract discriminative features between non-invasive and invasive cells. Our final goal is to identify the molecular and structural alterations that facilitate the infiltration of the basement membrane by invasive cells. Further investigations into these alterations may open the avenue for predicting metastatic progression and combat the invasion of tumors identified at an early stage Damien AmbrosettiCHU, Nice Title: To be announced Abstract: To be announced
Emmanuel BouilholiBV, Nice, France Title: Deep Learning for stain separation in histopathology images Abstract: To be announced |
Online user: 2 | Privacy |