Imaging ONE WORLD - "A deep learning strategy to segment nuclei with only 3 annotated images"

14 December 2020

Online

RMS Hosted Event

+ Add to calendar

Overview

This week will feature Thierry Pecot from Medical University of South Carolina.

Scientific Organisers: Stefanie Reichelt, Alex Sossick, Nick Barry, Alessandro Esposito and Kirti Prakash

The meeting will begin at 1pm UK Time.

As part of the 'Imaging ONE WORLD' series, the focus of these lectures is on microscopy and image analysis methods and how to apply these to your research. Almost all aspects of imaging such as sample preparation, labelling strategies, experimental workflows, ‘how-to’ image and analyse, as well as facilitating collaborations and inspiring new scientific ideas will be covered. Speakers will be available for questions and answers. The organisers, CRUK CI core facility staff, Gurdon Institute, MRC-LMB, MRC Cancer Unit and NPL will be able to continue the discussion and provide advice on your imaging projects.


Speaker

  • Thierry Pecot

    Medical University of South Carolina
    Thierry Pécot, a CZI Imaging Scientist, is an engineer and applied mathematician with expertise in bioimage informatics. He has developed algorithms and analysis pipelines for a variety of biological applications. Since he joined the Hollings Cancer Center at the Medical University of South Carolina in 2017, he has been developing tools to process immunofluorescence multiplexed images to profile tumor immune status. 

     


Speaker Abstract

Deep learning has revolutionized the automatic processing of images. While deep convolutional neural networks have demonstrated astonishing segmentation results for many biological objects acquired with microscopy, the good performance of this technology relies on large training datasets. In this talk, I will present how to reliably segment nuclei in 2D immunofluorescence multiplexed images with only 3 annotated images by taking advantage of an efficient annotation tool, by using data augmentation,  by deploying generative adversarial networks and by combining segmentation methods.



Share this