11.07.2020, Saturday 09:00 - 11:30 Hall L Technical Workshop
Chair(s)
  • Gerald M. Rubin, Ashburn, United States of America
  • Kei Ito, Cologne, Germany
Date
11.07.2020, Saturday
Session Time
09:00 - 11:30
Session Description
People who plan electron microscopy connectomics tend to underestimate the sheer amount of work required after serial section images are obtained. Here technical challenges and solutions for the fly connectome projects with transmission and focused ion beam scanning EM will be reported and shared with the people working on diverse organisms. Timing is ideal, because a series of FIB-SEM connectome papers will be published just before the Forum. To promote communication we provide ten minute discussion for each talk. S. Plaza established a computational platform for analyzing EM section images and visualizing/annotating the traced data. M. Januszewski developed the first automated neuron segmentation algorithm that works with reliable precision, which was a breakthrough for FIB-SEM analysis. A. Cardona established a platform for cooperative manual segmentation/annotation and organized a large collaborative project. R. Parekh trained and coordinated a team of tens of technicians who segment, proofread, and annotate data of both TEM and FIB-SEM with high throughput. H. Otsuna developed systems for finding homologous neurons between EM and light microscopy images, making it possible to switch EM and LM analyses freely. (The proposal might conflict with the FENS policy in that 3 speakers come from a single institute. It is inevitable given that all the large-scale fly EM connectomics are done at Janelia. We appreciate your flexible evaluation in this regard for this rare occasion.)
Technical Workshop

Chairperson introduction (ID 5637)

Lecture Time
09:00 - 09:03
Speaker(s)
  • Kei Ito, Cologne, Germany
  • Gerald M. Rubin, Ashburn, United States of America
Authors
  • Kei Ito, Cologne, Germany
  • Gerald M. Rubin, Ashburn, United States of America
Technical Workshop

Computational platform for processing EM data (ID 1136)

Lecture Time
09:03 - 09:28
Speaker(s)
  • Stephen Plaza, Ashburn, United States of America
Authors
  • Stephen Plaza, Ashburn, United States of America

Abstract

Abstract Body

Janelia FlyEM and Google produced the largest dense connectome ever consisting of around 25,000 neurons and 20 million synaptic connections in the central fly brain. In this talk, I will first highlight the various methods that made this reconstruction possible and then discuss the feasibility and economics of future connectomic efforts. This large connectome poses many challenges for data interpretation. The second part of this talk will discuss various considerations for using this data for different types of biological questions. To simplify data analysis, our team introduces compact data representations and many tools for navigating the dataset.

Hide
Technical Workshop

Automated segmentation of neurons (ID 1137)

Lecture Time
09:28 - 09:53
Speaker(s)
  • Michal Januszewski, Zürich, Switzerland
Authors
  • Michal Januszewski, Zürich, Switzerland

Abstract

Abstract Body

Dense mapping of neural circuits at synaptic level of detail currently requires tracing neurons within volume EM datasets. When done manually, this process is laborious, error-prone, and hard to scale up to keep pace with the increasing size and number of volumes available for study. Automated methods to perform the tracing are therefore necessary. Within the Connectomics at Google team, we developed a segmentation technique called Flood-Filling Networks (FFNs) based on a recurrent convolutional neural network, which has established a new state of the art for segmentation of blockface volume EM data.

I will discuss how FFNs work, and how we used them together with machine learning-based image normalization methods to segment a 20 TB FIB-SEM volume of roughly half of a drosophila brain acquired by the FlyEM team at Janelia. This fully automated reconstruction formed the basis of an extensive proofreading effort organized by FlyEM, which culminated in the recent release of the hemibrain connectome -- currently the largest synapse-resolution map of brain connectivity ever produced in any species.

Hide
Technical Workshop

Organizing collaborative segmentation and analysis (ID 1138)

Lecture Time
09:53 - 10:18
Speaker(s)
  • Albert Cardona, Cambridge, United Kingdom
Technical Workshop

From EM segmentation to connectome: The human factor (ID 1139)

Lecture Time
10:18 - 10:43
Speaker(s)
  • Ruchi Parekh, Ashburn, United States of America
Authors
  • Ruchi Parekh, Ashburn, United States of America

Abstract

Abstract Body

Generating large volumes of synapse-level connectomes has been a long-standing challenge. To produce such connectomes requires capabilities in tissue preparation, imaging, segmentation, proofreading and analysis. Advances in imaging technology, and automation in segmentation are speeding up the process of generating connectomes. However, generating a connectome still requires significant manual proofreading to fix segmentation errors. Once a connectome exists, the ability to analyze and interpret the data relies greatly on expertise across disciplines including detailed understanding of the underlying data and the methods of reconstruction. The proofreading and analysis demands are a major human resource bottleneck that will only get worse as volumes get bigger. The Connectome Annotation Team (CAT) at Janelia was built to address this bottleneck by efficiently adding resources and scaling up the proofreading operation. We created high-throughput pipelines for interviewing, hiring and training personnel to become expert annotators. We developed expertise across multiple EM volumes including the hemibrain and FAFB datasets of the fruit fly, and by working closely with labs in addressing their connectomic goals we are able to help solve specific problems. In this talk I will present how CAT was built, its ongoing operations and the resulting effect on the proofreading and analysis bottleneck.
Hide
Technical Workshop

**LIVE LECTURE **Bridging electron microscopy and light microscopy connectome data (ID 1140)

Lecture Time
10:43 - 11:08
Speaker(s)
  • Hideo Otsuna, Ashburn, United States of America
Authors
  • Hideo Otsuna, Ashburn, United States of America
  • John Bogovic, Ashburn, United States of America
  • Shinya Takemura, Ashburn, United States of America
  • Kazunori Shinomiya, Ashburn, United States of America
  • Stephan Saalfeld, Ashburn, United States of America
  • Takashi Kawase, Ashburn, United States of America

Abstract

Abstract Body

The single cell type specific split-GAL4 driver lines are a powerful tool for studying the Drosophila central nervous system in the neuronal morphology and animal behavior. However, choosing lines to intersect that have overlapping patterns restricted to one to a few neurons has been cumbersome. Our newly created method lets users search the light microscopy (LM) GAL4 data from the electron microscopy (EM) single neuron. Also, the method allows searching from the LM neurons to EM neurons. The "VVDviewer" 3D rendering software and "Color depth MIP search" Fiji plug-in to dramatically improve the speed of querying large EM-LM datasets of potential lines to intersect and aid in the split-GAL4 creation.
Hide
Technical Workshop

Live panel (ID 5638)

Lecture Time
11:08 - 11:30
Speaker(s)
  • Kei Ito, Cologne, Germany
  • Gerald M. Rubin, Ashburn, United States of America
  • Stephen Plaza, Ashburn, United States of America
  • Michal Januszewski, Zürich, Switzerland
  • Albert Cardona, Cambridge, United Kingdom
  • Ruchi Parekh, Ashburn, United States of America
  • Hideo Otsuna, Ashburn, United States of America
Authors
  • Kei Ito, Cologne, Germany
  • Gerald M. Rubin, Ashburn, United States of America
  • Stephen Plaza, Ashburn, United States of America
  • Michal Januszewski, Zürich, Switzerland
  • Albert Cardona, Cambridge, United Kingdom
  • Ruchi Parekh, Ashburn, United States of America
  • Hideo Otsuna, Ashburn, United States of America