We propose to develop the computational infrastructure necessary for future large-scale reverse engineering of cortical circuits. Neuroscience researchers are using confocal and electron micrograph (EM) techniques to scan neural tissue at high resolution. Their goal is to capture a detailed map of all neurons and synapses within the nervous system of an organism. Through automation, it is now possible to acquire petabyte size volumes. However, there is no way to currently analyze such large datasets. We will develop an open-source system that supports remote visualization and analysis of arbitrary sized volumes. Our system will be named """"""""Open SSECRETT"""""""" and enable a collaborative effort to develop automatic segmentation of neurons and synaptic connections. The proposed system will be architected around remote data access so that geographically diverse research groups can collaborate on the enormous task of segmenting neurons from volumes in the database. Custom clients will implement various segmentation algorithms and the results will be put back in a central database. This will allow the algorithms and their results to be shared and compared. We will also develop standard clients that will allow universal access to view and explore the immense data.
Since the discovery of Golgi staining, tracing cells has revealed how individual neurons form connections in neural tissue[1]. Unfortunately, early techniques could only reveal complex neural processes by imaging a few select neurons. High-resolution volumes, generated by electron micrographs, allow all cells in a block of tissue to be traced. However, since axons make connections across large distances, it is necessary to image large tissue blocks in order to get a complete circuit. Automated sectioning and imaging are now capable of generating such volumes, but no software currently available can analyze the resulting data. Scanning a cubic centimeter of tissue at nanometer EM scale (figure 1) would produce hundreds of petabytes of data! It is a challenge to even view such large data, let alone segment circuits of neurons from it. We propose developing a scalable software database that manages exabyte sized volumes. It will support a community of researchers who are working on algorithms to automatically segment neurons and analyze resulting circuits.
