3D Segmentation of Neurons from Electron Micrographs - PowerPoint PPT Presentation

1 / 13
About This Presentation
Title:

3D Segmentation of Neurons from Electron Micrographs

Description:

Neurons interpret electrochemical signals from multiple ... 'Canny' Edge Detection Algorithm Outline: Remove white noise. Apply an edge-detecting operator. ... – PowerPoint PPT presentation

Number of Views:32
Avg rating:3.0/5.0
Slides: 14
Provided by: sebasti67
Category:

less

Transcript and Presenter's Notes

Title: 3D Segmentation of Neurons from Electron Micrographs


1
3D Segmentation of Neurons from Electron
Micrographs
  • Nathaniel Thomas
  • under the direction of Prof. Sebastian Seung and
  • Mr. Srinivas Turaga
  • Research Science Institute 2006

2
Neurons in the Brain
  • Neuron function is the most fundamental level of
    brain function.
  • Neurons interpret electrochemical signals from
    multiple other neurons and output a single on/off
    signal.
  • The brain is essentially a large neural network.
  • Fully understanding the brain requires schematics
    of the brains neural circuitry.

3
Neural Circuitry Schematics
  • We have 3D electron micrographs (EMs) of neurons
    in the brain.
  • We want a schematic of the neural circuitry in
    the images.
  • This requires finding the edges of neurons in the
    electron micrographs.

4
Electron Micrographs of Neurons
Electron micrograph of rat retina tissue
5
Edge Detection
  • Canny Edge Detection Algorithm Outline
  • Remove white noise.
  • Apply an edge-detecting operator.
  • Thin the detected edges to increase precision.
  • Connect small discontinuities in edges using
    hysteresis.

6
Step 1 Remove Noise
  • 1D Gaussian filter

Before filter
Intensity
Space
After filter
7
Step 2 Apply Operator
1D Image
Operator
Output
First Derivative of a Gaussian
Second Derivative of a Gaussian
8
Step 3 Non-Maximum Suppression
With NMS
Without NMS
9
Step 4 Hysteresis
Strong Edges
Weak Edges
10
Neurons Identified in EMs
11
3D Segmentation of Neurons
12
Future Work
  • Increase the precision of 3D segmentation by
    making more intelligent algorithms
  • Algorithms might have prior knowledge about types
    of junctions between cells.
  • Algorithms could identify sites of
    electrochemical interaction in the 3D
    segmentation to create neural schematics.

13
Acknowledgements
  • Many thanks to
  • Professor Seung for the opportunity to work in
    his lab
  • Mr. Srini Turaga for his guidance throughout this
    project
  • The Center for Excellence in Education and the
    Massachusetts Institute of Technology for their
    efforts and resources supporting the Research
    Science Institute.
Write a Comment
User Comments (0)
About PowerShow.com