An automated images-to-graphs framework for high resolution connectomics

Abstract

Reconstructing a map of neuronal connectivity is a critical challenge in contemporary neuroscience. Recent advances in high-throughput serial section electron microscopy (EM) have produced massive 3D image volumes of nanoscale brain tissue for the first time. The resolution of EM allows for individual neurons and their synaptic connections to be directly observed. Recovering neuronal networks by manually tracing each neuronal process at this scale is unmanageable, and therefore researchers are developing automated image processing modules. Thus far, state-of-the-art algorithms focus only on the solution to a particular task (e.g., neuron segmentation or synapse identification). In this manuscript we present the first fully-automated images-to-graphs pipeline (i.e., a pipeline that begins with an imaged volume of neural tissue and produces a brain graph without any human interaction). To evaluate overall performance and select the best parameters and methods, we also develop a metric to assess the quality of the output graphs. We evaluate a set of algorithms and parameters, searching possible operating points to identify the best available brain graph for our assessment metric. Finally, we deploy a reference end-to-end version of the pipeline on a large, publicly available data set. This provides a baseline result and framework for community analysis and future algorithm development and testing. All code and data derivatives have been made publicly available in support of eventually unlocking new biofidelic computational primitives and understanding of neuropathologies.

Quantifying Mesoscale Neuroanatomy Using X-Ray Microtomography

“…Thus information about the location and size of cells and vessels, obtained through X-ray image analysis, can be used to improve the alignment of EM datasets and also facilitate merging 2D outputs from segmented EM images (Gray Roncal et al, 2015). Because our open-access pipeline for X-ray image analysis has been integrated using community standard tools and approaches, we can readily combine existing EM analysis pipelines with our methods to analyze a dataset imaged using µCT and EM.…”

“…With more training data from human annotators, we can leverage more sophisticated nonlinear classification strategies such as convolutional neural networks for segmentation and axon tracing. These approaches have been shown to achieve state-of-the-art performance in the identification of synapses and segmentation of cell bodies in EM data (Gray Roncal et al, 2015; Turaga et al, 2010). Finally, improvements in the spatial resolution of tissue samples will aid in the challenge of resolving adjacent neural structures as separate objects, thereby leading to more efficient and robust approaches for cell detection.…”

“…However, with developments in large-scale EM connectomics (Helmstaedter et al, 2011; Gray Roncal et al, 2015) and the techniques we present here for µCT, far larger and more comprehensive datasets are possible at a scale previously unachieved. Indeed, it is possible to scale up µCT imaging to large volumes of human brain tissue, such as the cerebellum and frontal cortex (Mizutani et al, 2010b; Hieber et al, 2016).…”

A Multicore Path to Connectomics-on-Demand

“…One should contrast this with the recent connectomics system of Roncal et al [58] that uses a cluster of 100 AMD Opteron cores, 1 terabyte of memory, and 27 GeForce GTX Titan cards to process a terabyte in 4.5 weeks, and the stateof-the-art distributed MapReduce based system of Plaza and Berg [56] that uses 512 cores and 2.9 terabytes of memory to process a terabyte of data in 140 hours (not including skeletonization). Importantly, the speed of our pipeline does not come at the expense of accuracy, which is on par or better than existing systems in the literature [29,56,58] (using the accepted variation of information (VI) measure [44]). Our high-level pipeline design builds on prior work [29,42,51,52,55,56].…”

The Block Object Storage Service (bossDB): A Cloud-Native Approach for Petascale Neuroscience Discovery

“…In designing our Block and Object Storage Service Database (bossDB), we researched several related efforts, including DVID [11] which excels in versioned terascale storage and CATMAID [12] which provides a mature manual annotation platform. We previously worked with Neu-roData to develop ndstore [13], which originated and implemented many of the design principles necessary to store and access high-dimensional imaging datasets, including an efficient internal data representation and associated spatial indexing scheme; the Reusable Annotation Markup for Open Neuroscience (RAMON), an annotation schema for connectomics [14]; an API to remotely access services; and MATLAB and Python toolkits to facilitate usability. Based on this prior research and an understanding of the evolving requirements driven by new and maturing imaging modalities, we created a robust, cloudnative petascale datastore with a number of services and support tools ( Figure 1).…”