REGISTER HERE: https://epfl.zoom.us/webinar/register/WN_jaLD_KmlSLaeFjp16E8YtQ
14:30 – 15:00 Introduction with A. Archetti, M. Brondi and L. Mariotti
15:00 – 15:30 Andrea Giovannucci (Joint Department of Biomedical Engineer, University of North Carolina, USA) agiovann@email.unc.edu
Neuro-image segmentation and real-time data processing
Abstract: Optical microscopy methods such as calcium and voltage imaging already enable fast activity readout (30-1000Hz) of large neuronal populations using light. However, the lack of corresponding advances in online algorithms has slowed progress in retrieving information about neural activity during or shortly after an experiment. This technological gap not only prevents the execution of novel real-time closed-loop experiments, but also hampers fast experiment-analysis-theory turnover for high-throughput imaging modalities. The fundamental challenge is to reliably extract neural activity from fluorescence imaging frames at speeds compatible with new indicator dynamics and imaging modalities. To meet these challenges and requirements, we propose a framework for Fluorescence Imaging OnLine Analysis (FIOLA). FIOLA exploits computational graphs and accelerated hardware to preprocess fluorescence imaging movies and extract fluorescence traces at speeds in excess of 300Hz on calcium imaging datasets and at speeds over 400Hz on voltage imaging datasets. Additionally, we present the first real-time spike extraction algorithm for voltage imaging data. We evaluate FIOLA on both simulated data and real data, demonstrating reliable and scalable performance. Our methods provide the computational substrate required to precisely interface large neuronal populations and machines in real-time, enabling new applications in neuroprosthetics, brain-machine interfaces, and experimental neuroscience. Moreover, this new set of tools is poised to dramatically shorten the experiment-data-theory cycle by providing immediate feedback on the activity of large neuronal populations at experimental time.
15:30 – 16:00 Daniel Sage (EPFL, CH) daniel.sage@epfl.ch
Microscopy Image Analysis – The Shift to Deep Learning?
Abstract: The quantification of microscopy images require automatic tools to extract relevant information from complex data. To tackle this task, numerous image analysis algorithms have been designed, commonly based on prior knowledge and on physical modeling. However, the recent success of deep learning (DL) in computer science has drastically changed the bioimage analysis workflows to a data-centric paradigm. While this DL technology remains relatively inaccessible to end-users, recent efforts have been proposed to facilitate the deployment of DL for some bioimage applications through new open-source software packages. Here, we present a set of open-source and user-friendly tools that allow to test DL models and to gain proficiency in DL technology: the centralized repository of bioimage model (Bioimage Model Zoo), the ready-to-use notebooks for the training, and the plugin deepImageJ that can run a DL model in ImageJ. We provide also good practice tips to avoid the risk of misuse. We address some practical issues such as the availability of a massive amount of images, the understanding of the generalizability concept, or the selection of the pre-trained models. The shift to deep learning also questions the community about the trust, reliability, and validity of such trained deep learning models.
16:00 – 16:30 Carsen Stringer (Janelia Research Campus, Ashburn (Virginia, USA)) stringerc@janelia.hhmi.org
Anatomical and functional algorithms for cellular segmentation
Abstract: Many biological applications require the segmentation of cell bodies, membranes, and nuclei from microscopy images. We therefore developed a generalist, deep learning-based segmentation algorithm called Cellpose, which can very precisely segment a wide range of image types out-of-the-box and does not require model retraining or parameter adjustments. Additionally, many scientists acquire functional imaging data of neural activity. We developed Suite2p, a functional segmentation algorithm, that performs well on a variety of functional imaging data. We will cover how each of these segmentation algorithms works, and the various circumstances where each algorithm perform well.
16:30 – 17:30 Follow-up session (open questions)
Anna Archetti (UNIPD) anna.archetti@unipd.it
Marco Brondi (VIMM) marco.brondi@in.cnr.it
Letizia Mariotti (CNR Neuroscience Institute, Padova)
Gianluca Ruffato (UNIPD) gianluca.ruffato@unipd.it
