The FlyEM project was launched around 2008 building on prior work from the labs of Dmitri Chklovskii and Rick Fetter, as well as of Ian Meinertzhagen and other Janelia Visiting Scientists. In the first few years we successfully reconstructed an array of lamina cartridges and a medulla column using a newly developed automated reconstruction pipeline for transmission electron microscopy (TEM) images. We also imaged and experimented with many other data sets, including a complete first instar larval brain sectioned and imaged at 5 tilts (a project now being continued in the Cardona lab).
The medulla reconstruction demonstrated both the feasibility and utility of a connectome at the synapse level, each of which had been doubted. It enabled the identification of the elementary motion detector circuit in the medulla, surmounting a 50-year old barrier in the analysis of motion-sensing pathways. Nonetheless, although a huge advance over the purely manual methods previously used, these reconstructions demonstrated two major limitations to further advances. First, the z-axis resolution of TEM imaging, which is determined by an irreducibly minimal section thickness, is nevertheless an order of magnitude poorer than x,y resolution available by TEM, and insufficient to trace all the neurites in the neuropile through consecutive sections. Second, the rate-limiting step of this pipeline occurs when human proofreaders exhaustively scan segmented image stacks to verify the correctness of segmentation.
FIB-SEM isotropic resolution allows tracing to be done in arbitrary orientations. The three videos show 3 orientations of the same dataset (XY, YZ, and XZ orientations).
In order to scale up our reconstruction effort we therefore needed to improve image resolution to obtain more nearly isotropic voxel dimensions, and also to accelerate our proofreading rate. After considering several alternatives, we converged on using Focused Ion Beam Scanning Electron Microscopy (FIB-SEM) to achieve isotropic imaging resolution. The quality of images produced by FIB-SEM along with improvements in automated segmentation algorithms, new work flows, and new analysis software, enabled us to scale up the pipeline speed to reconstruct circuits of larger dimensions to greater completeness than were possible with TEM.