Spectral Algorithms for Segmenting Neurons in their Three-dimensional Space

Friday, October 28, 2011
Hall 1-2 (San Jose Convention Center)
Jose Farrington-Zapata , Computer Science, University of Puerto Rico-Rio Piedras, San Juan, PR
Ioannis Koutis , Computer Science, University of Puerto Rico-Rio Piedras, San Juan, PR
Richard Garcia-Lebron , Computer Science, University of Puerto Rico-Rio Piedras, San Juan, PR
Eduardo Rosa-Molinar , Biology, University of Puerto Rico-Rio Piedras, San Juan, PR
Jose Serrano-Velez , Biology, University of Puerto Rico-Rio Piedras, San Juan, PR
The automated or semi-automated segmentation of individual neurons in
electron microscopic (EM) images is a crucial step in the acquisition
and analysis of connectomes, i.e. maps of the neural connections. The
acquisition of connectomes is a computational problem of colossal size
and importance. Algorithms for it are still in their infancy.
 In this work, we adapt and combine the random walker method and
spectral rounding, two previously proposed algorithms for image
segmentation. We also introduce and investigate generalized cuts. The
random walker algorithm computes a rough shape of the neuron and then
spectral rounding or generalized cuts find a more precise
segmentation. Supplemented with the recently discovered linear system
solvers the algorithms make efficient use of 3D contextual information
to generate noise-insensitive neuron segmentation that significantly
improves upon previous algorithms. Registered serial 3D EM images of
retrograde-filled spinal motor neurons and in bloc heavy atom stained
were used in order to obtain high contrast images of neuronal soma,
individual axons, dendritic spines, and synapses in order to easily
and accurately segment neurons. Our data shows that dendrites of the
same neuron that appear disconnected or ambiguously connected in 2D
frames, are ultimately connected through the soma in the serial 3D
image.