CAJAL3D: Towards A Fully Automatic Pipeline for Connectome Estimation from High-Resolution EM Data

In recent years, technological advances have allowed for millions of cubic microns of cortical tissue to be imaged at very high resolution (e.g., 4x4x45 nm3) using electron microscopy. A variety of efforts have successfully applied 2- and 2.5-dimensional segmentation methods to identify major image features (e.g. vesicles, mitochondria) and neurite segments over a small number of brain slices in these samples. However, the data are too large and the problem space is too big for any one group to fully analyze. We have therefore developed a common language and data repository to facilitate the sharing of connectome data and results within the scientific community. More specifically, we are designing an ecosystem of standardized interfaces and services to facilitate large-scale, collaborative connectomics, called CAJAL3D (Connectome Annotation through Joint Analysis of Large 3-dimensional Data). Our design facilitates interoperability of algorithms and the interpretability of results, by standardizing algorithm inputs and outputs through defined annotation types at all processing stages.

In addition to defining a common interface for connectomics, we developed a processing pipeline framework that implements our annotation standard and is integrated with the Open Connectome Project (OCP) web services. The framework is built on a client-server based infrastructure that facilitates scalable, distributed processing. The pipeline exchanges data and data products associated with various forms of connectomics information, ranging from raw images to processed graphs, with the OCP storage servers. The annotation standard, image processing algorithms, image and annotation storage databases and associated web services are being made available to the community as a free resource. We are in the process of soliciting feedback from the community for additional features and functionality.

DMK and WRG contributed equally to this work.