Large Scale Synapse Detection
Synapses are interconnectivity structures that link neurons in the brain. In order to get a better understanding of the brain, it is essential to understand the properties of synapses, or the interconnects of the brain. This investigation focussed on designing visual recognition algorithms that could reliably localize these synaptic junctions from Electron Micrographs. In addition, the datasets used for this work were terabyte scale, meaning the algorithms designed should scale to large datasets. We propose a fast scanning algorithm utilizing image morphology operators to efficiently scan the entire dataset for generating a probability distribution or heat map for synapse presence. This procedure aids in filtering out the "easy" true negatives, and lets us run the computationally costlier appearance based classifiers only on regions where the scanning algorithm has a score greater than a threshold. The costlier appearance based features proposed in this work exploit discriminative visual characteristics of synapses such as the spherical shape of vesicles, ridge profiles of junctions, and the ellipticity of ribbons in a multiple kernel learning framework.
Relevant Publications
Synapse Classification and Localization in Electron Micrographs
Vignesh Jagadeesh, James Anderson, Bryan Jones, Robert Marc, Steven Fisher, B.S. Manjunath
Pattern Recognition Letters, 2013.
Vignesh Jagadeesh, James Anderson, Bryan Jones, Robert Marc, Steven Fisher, B.S. Manjunath
Pattern Recognition Letters, 2013.