Project

Back to overview

A Higher Level for Neuron Reconstruction

Applicant Cook Matthew
Number 160133
Funding scheme Project funding
Research institution Institut für Neuroinformatik Universität Zürich Irchel und ETH Zürich
Institution of higher education University of Zurich - ZH
Main discipline Information Technology
Start/End 01.07.2016 - 30.06.2019
Approved amount 357'926.00
Show all

Keywords (5)

Connectomics; Biomedical Image Processing; Neural Reconstruction; Electron Microscopy Image Segmentation; Neural Image Segmentation

Lay Summary (German)

Lead
Für ein besseres Verständnis des Gehirns ist es notwendig, die Struktur, insbesondere den zellulären Schaltplan des Gehirns, zu kennen. Dieser Schaltplan ist ­ bis auf wenige, sehr kleine Gehirnausschnitte ­ bisher nicht verfügbar. Dank Fortschritte in der 3D­-Elektronenmikroskopie ist es heute möglich, Datensätze von größeren Gehirnausschnitten aufzunehmen. Die genaue Verschaltung aller Nervenzellen können dadurch aufgeklärt werden. Dafür ist jedoch ein manuelles “Nachzeichnen” (d.h., Segmentieren) jeder einzelnen Zelle vonnöten. Dies ist ­ bei Datensätzen von mehreren Terabytes Größe kein praktikables Vorgehen. Ziel ist es daher, die Segmentierung der Nervenzellen automatisiert mit Hilfe von Algorithmen durchzuführen.
Lay summary

Inhalt und Ziel des Forschungsprojekt

Wir mochten den Stand der Technik im Bereich der automatisierten Nervenzell­-Segmentierung auf verschiedene Arten vorantreiben:

  1.   Durch Methoden, die auf hoheren Abstraktionsebenen agieren, und auch Merkmale wie beispielsweise die Form der Zellen erfassen. Diese Merkmale konnen benutzt werden, um Segmentierungsvorschlage zu evaluieren. 

  2. Durch Methoden, die autonom anhand manuell generierter Segmentierungen lernen. Die dafür entwickelten Algorithmen für Maschinelles Lernen, insbesondere Strukturellem Lernen, werden das Einstellen von Modell-Parametern erheblich erleichtern.

  3. Durch die Entwicklung intuitiver Fehler­maße, die es erlauben, Segmentierungsvorschlage auf ihre Richtigkeit zu untersuchen und so verschiedene Algorithmen objektiv zu vergleichen.

Wissenschaftlicher und gesellschaftlicher Kontext des Forschungsprojekts

Mit diesem Forschungsprojekt werden neue Methoden im Bereich des Maschinellen Lernens und der Bildverarbeitung entwickelt. Wir hoffen dadurch, den Gehirnforschern Zugang zu genaueren Schaltplanen des Gehirns zu verschaffen. Den Wissenschaftlern wird es dadurch ermoglicht, Modelle zu testen und das Gehirn besser zu verstehen. 

Direct link to Lay Summary Last update: 20.06.2016

Responsible applicant and co-applicants

Employees

Publications

Publication
Microtubule Tracking in Electron Microscopy Volumes
Eckstein Nils, Buhmann Julia, Cook Matthew, Funke Jan (2020), Microtubule Tracking in Electron Microscopy Volumes, in International Conference on Medical Image Computing and Computer-Assisted Intervention (MICCAI), Springer, Lima, Peru.
Automatic Detection of Synaptic Partners in a Whole-Brain Drosophila EM Dataset
Buhmann Julia, Sheridan Arlo, Gerhard Stephan, Krause Renate, Nguyen Tri, Heinrich Larissa, Schlegel Philipp, Lee Wei-Chung Allen, Wilson Rachel, Saalfeld Stephan, others (2019), Automatic Detection of Synaptic Partners in a Whole-Brain Drosophila EM Dataset, Cold Spring Harbor Laboratory, Cold Spring Harbor.
Synaptic Partner Prediction from Point Annotations in Insect Brains
Buhmann Julia, Krause Renate, Lentini Rodrigo Ceballos, Eckstein Nils, Cook Matthew, Turaga Srinivas, Funke Jan (2018), Synaptic Partner Prediction from Point Annotations in Insect Brains, Springer International Publishing, Cham.
TED: A Tolerant Edit Distance for segmentation evaluation
Funke Jan, Klein Jonas, Moreno-Noguer Francesc, Cardona Albert, Cook Matthew (2017), TED: A Tolerant Edit Distance for segmentation evaluation, in Methods, 115, 119-127.
Structured learning of assignment models for neuron reconstruction to minimize topological errors
Funke Jan, Klein Jonas, Moreno-Noguer Francesc, Cardona Albert, Cook Matthew (2016), Structured learning of assignment models for neuron reconstruction to minimize topological errors, in 2016 IEEE 13th International Symposium on Biomedical Imaging (ISBI 2016), Prague, Czech RepublicIEEE, Prague.
Tracking of microtubules in anisotropic volumes of neural tissue
Buhmann Julia M., Gerhard Stephan, Cook Matthew, Funke Jan (2016), Tracking of microtubules in anisotropic volumes of neural tissue, in 2016 IEEE 13th International Symposium on Biomedical Imaging (ISBI 2016), Prague, Czech RepublicIEEE, Prague.

Datasets

synful_fafb

Author Buhmann, Julia; Funke, Jan; Cook, Matthew
Publication date 13.12.2019
Persistent Identifier (PID) https://doi.org/10.1101/2019.12.12.874172
Repository https://cremi.org
Abstract
The synful_fafb dataset consists of 244 M synaptic partners predicted in the full adult fly brain Electron Microscopy dataset [Zheng et al. 2018].For more information on the dataset and how to interact with it, see this github repository: https://github.com/funkelab/synful_fafb.

Collaboration

Group / person Country
Types of collaboration
Frances Moreno-Noguer, IRI, UPC Barcelona Spain (Europe)
- in-depth/constructive exchanges on approaches, methods or results
- Publication
- Research Infrastructure
Albert Cardona, Janelia Farm (Howard Hughes Medical Institute) United States of America (North America)
- in-depth/constructive exchanges on approaches, methods or results
- Publication
- Exchange of personnel
Stephan Saalfeld, HHMI Janelia United States of America (North America)
- in-depth/constructive exchanges on approaches, methods or results
- Publication
- Research Infrastructure
Srini Turaga, HHMI Janelia United States of America (North America)
- in-depth/constructive exchanges on approaches, methods or results
- Publication
- Research Infrastructure

Scientific events

Active participation

Title Type of contribution Title of article or contribution Date Place Persons involved
Connectomics Poster Synaptic Partner Prediction from Point Annotations in Insect Brains 15.04.2020 Berlin, Germany Buhmann Julia;
Connectomics Talk given at a conference Tracking of Microtubules 17.04.2019 Berlin, Germany Eckstein Nils;


Knowledge transfer events

Active participation

Title Type of contribution Date Place Persons involved
Deep Learning: Anwendung in den Neurowissenschaften Talk 18.12.2019 Zurich, Switzerland Buhmann Julia;
Synapse Predictions in Drosophila (Schulvortrag) Talk 19.03.2019 Ashburn, Virginia, United States of America Buhmann Julia;


Associated projects

Number Title Start Funding scheme
149861 Neuron Reconstruction from Electron Microscopy Images 01.12.2013 Project funding

Abstract

To better understand the brain, the substrate of our cognition, we require detailed knowledge of its structure, in particular its ``wiring diagram''. Unfortunately, an exact diagram of even the tiniest part of a brain is still not available today!Thanks to great advances in volume electron microscopy, human experts can now reliably trace individual neurons. Unfortunately, the effort for a manual tracing of {\it all} processes in the gigavoxel images that are now becoming standard is prohibitive: it is measured in multiple decades. And yet, a fully manual tracing still is the current state of the art.The reason for this sorry state of affairs is that automated methods still break down in the multiple parts of a typical volume image where neural processes become very thin, the staining is imperfect or the signal-to-noise ratio is poor. And yet, the high agreement obtained between independent human tracings testifies to the fact that these obstacles are, in principle, surmountable.We propose to work on next-generation algorithms that surpass the state of the art in a number of ways: Firstly, we will develop higher-order factors that can capture properties such as shape. These factors will be used both post hoc, to evaluate predictions based on local evidence alone, to obtain uncertainty measures that can be used for interactive training or semi-manual tracing. The same factors will also be used during inference. The resulting optimization is very hard and calls for an informed generation of competing hypotheses which we will also develop and refine. An important part of this proposal is a structured learning framework, where we wish to learn from cheap annotations: skeletons rather than pixel-level annotations. To that end, we will also need a meaningful error metric, which we are confident we can find based on a concept of edit distance.In summary, we will leverage some of the latest techniques in computer vision and machine learning, and push the boundary in a few important places. We thus hope to enable neuroscientists to ask new questions, and answer them, based on the massive amounts of data they currently generate.
-