Reconstructing a connectome from an EM dataset often requires a large effort of proofreading automatically generated segmentations. While many tools exist to enable tracing or proofreading, recent advances in EM imaging and segmentation quality suggest new strategies and pose unique challenges for tool design to accelerate proofreading. Namely, we now have access to very large multi-TB EM datasets where (1) many segments are largely correct, (2) segments can be very large (several GigaVoxels), and where (3) several proofreaders and scientists are expected to collaborate simultaneously. In this paper, we introduce NeuTu as a solution to efficiently proofread large, high-quality segmentation in a collaborative setting. NeuTu is a client program of our high-performance, scalable image database called DVID so that it can easily be scaled up. Besides common features of typical proofreading software, NeuTu tames unprecedentedly large data with its distinguishing functions, including: (1) low-latency 3D visualization of large mutable segmentations; (2) interactive splitting of very large false merges with highly optimized semi-automatic segmentation; (3) intuitive user operations for investigating or marking interesting points in 3D visualization; (4) visualizing proofreading history of a segmentation; and (5) real-time collaborative proofreading with lock-based concurrency control. These unique features have allowed us to manage the workflow of proofreading a large dataset smoothly without dividing them into subsets as in other segmentation-based tools. Most importantly, NeuTu has enabled some of the largest connectome reconstructions as well as interesting discoveries in the fly brain.View Publication Page
Extracting a connectome from an electron microscopy (EM) data set requires identification of neurons and determination of synapses between neurons. As manual extraction of this information is very time-consuming, there has been extensive research effort to automatically segment the neurons to help guide and eventually replace manual tracing. Until recently, there has been comparatively less research on automatically detecting the actual synapses between neurons. This discrepancy can, in part, be attributed to several factors: obtaining neuronal shapes is a prerequisite first step in extracting a connectome, manual tracing is much more time-consuming than annotating synapses, and neuronal contact area can be used as a proxy for synapses in determining connections.
However, recent research has demonstrated that contact area alone is not a sufficient predictor of synaptic connection. Moreover, as segmentation has improved, we have observed that synapse annotation is consuming a more significant fraction of overall reconstruction time. This ratio will only get worse as segmentation improves, gating overall possible speed-up. Therefore, we address this problem by developing algorithms that automatically detect pre-synaptic neurons and their post-synaptic partners. In particular, pre-synaptic structures are detected using a Deep and Wide Multiscale Recursive Network, and post-synaptic partners are detected using a MLP with features conditioned on the local segmentation.
This work is novel because it requires minimal amount of training, leverages advances in image segmentation directly, and provides a complete solution for polyadic synapse detection. We further introduce novel metrics to evaluate our algorithm on connectomes of meaningful size. These metrics demonstrate that complete automatic prediction can be used to effectively characterize most connectivity correctly.
Understanding memory formation, storage and retrieval requires knowledge of the underlying neuronal circuits. In Drosophila, the mushroom body (MB) is the major site of associative learning. We reconstructed the morphologies and synaptic connections of all 983 neurons within the three functional units, or compartments, that compose the adult MB’s α lobe, using a dataset of isotropic 8-nm voxels collected by focused ion-beam milling scanning electron microscopy. We found that Kenyon cells (KCs), whose sparse activity encodes sensory information, each make multiple en passant synapses to MB output neurons (MBONs) in each compartment. Some MBONs have inputs from all KCs, while others differentially sample sensory modalities. Only six percent of KC>MBON synapses receive a direct synapse from a dopaminergic neuron (DAN). We identified two unanticipated classes of synapses, KC>DAN and DAN>MBON. DAN activation produces a slow depolarization of the MBON in these DAN>MBON synapses and can weaken memory recall.View Publication Page
We reconstructed the synaptic circuits of seven columns in the second neuropil or medulla behind the fly's compound eye. These neurons embody some of the most stereotyped circuits in one of the most miniaturized of animal brains. The reconstructions allow us, for the first time to our knowledge, to study variations between circuits in the medulla's neighboring columns. This variation in the number of synapses and the types of their synaptic partners has previously been little addressed because methods that visualize multiple circuits have not resolved detailed connections, and existing connectomic studies, which can see such connections, have not so far examined multiple reconstructions of the same circuit. Here, we address the omission by comparing the circuits common to all seven columns to assess variation in their connection strengths and the resultant rates of several different and distinct types of connection error. Error rates reveal that, overall, <1% of contacts are not part of a consensus circuit, and we classify those contacts that supplement (E+) or are missing from it (E-). Autapses, in which the same cell is both presynaptic and postsynaptic at the same synapse, are occasionally seen; two cells in particular, Dm9 and Mi1, form ≥20-fold more autapses than do other neurons. These results delimit the accuracy of developmental events that establish and normally maintain synaptic circuits with such precision, and thereby address the operation of such circuits. They also establish a precedent for error rates that will be required in the new science of connectomics.View Publication Page
Animal behaviour arises from computations in neuronal circuits, but our understanding of these computations has been frustrated by the lack of detailed synaptic connection maps, or connectomes. For example, despite intensive investigations over half a century, the neuronal implementation of local motion detection in the insect visual system remains elusive. Here we develop a semi-automated pipeline using electron microscopy to reconstruct a connectome, containing 379 neurons and 8,637 chemical synaptic contacts, within the Drosophila optic medulla. By matching reconstructed neurons to examples from light microscopy, we assigned neurons to cell types and assembled a connectome of the repeating module of the medulla. Within this module, we identified cell types constituting a motion detection circuit, and showed that the connections onto individual motion-sensitive neurons in this circuit were consistent with their direction selectivity. Our results identify cellular targets for future functional investigations, and demonstrate that connectomes can provide key insights into neuronal computations.View Publication Page
Recent results have shown the possibility of both reconstructing connectomes of small but biologically interesting circuits and extracting from these connectomes insights into their function. However, these reconstructions were heroic proof-of-concept experiments, requiring person-months of effort per neuron reconstructed, and will not scale to larger circuits, much less the brains of entire animals. In this paper we examine what will be required to generate and use substantially larger connectomes, finding five areas that need increased attention: firstly, imaging better suited to automatic reconstruction, with excellent z-resolution; secondly, automatic detection, validation, and measurement of synapses; thirdly, reconstruction methods that keep and use uncertainty metrics for every object, from initial images, through segmentation, reconstruction, and connectome queries; fourthly, processes that are fully incremental, so that the connectome may be used before it is fully complete; and finally, better tools for analysis of connectomes, once they are obtained.View Publication Page
Pixel and superpixel classifiers have become essential tools for EM segmentation algorithms. Training these classifiers remains a major bottleneck primarily due to the requirement of completely annotating the dataset which is tedious, error-prone and costly. In this paper, we propose an interactive learning scheme for the superpixel classifier for EM segmentation. Our algorithm is 'active semi-supervised' because it requests the labels of a small number of examples from user and applies label propagation technique to generate these queries. Using only a small set (< 20%) of all datapoints, the proposed algorithm consistently generates a classifier almost as accurate as that estimated from a complete groundtruth. We provide segmentation results on multiple datasets to show the strength of these classifiers.View Publication Page
We aim to improve segmentation through the use of machine learning tools during region agglomeration. We propose an active learning approach for performing hierarchical agglomerative segmentation from superpixels. Our method combines multiple features at all scales of the agglomerative process, works for data with an arbitrary number of dimensions, and scales to very large datasets. We advocate the use of variation of information to measure segmentation accuracy, particularly in 3D electron microscopy (EM) images of neural tissue, and using this metric demonstrate an improvement over competing algorithms in EM and natural images.View Publication Page
The ability to automatically segment an image into distinct regions is a critical aspect in many visual processing applications. Because inaccuracies often exist in automatic segmentation, manual segmentation is necessary in some application domains to correct mistakes, such as required in the reconstruction of neuronal processes from microscopic images. The goal of the automated segmentation tool is traditionally to produce the highest-quality segmentation, where quality is measured by the similarity to actual ground truth, so as to minimize the volume of manual correction necessary. Manual correction is generally orders-of-magnitude more time consuming than automated segmentation, often making handling large images intractable. Therefore, we propose a more relevant goal: minimizing the turn-around time of automated/manual segmentation while attaining a level of similarity with ground truth. It is not always necessary to inspect every aspect of an image to generate a useful segmentation. As such, we propose a strategy to guide manual segmentation to the most uncertain parts of segmentation. Our contributions include 1) a probabilistic measure that evaluates segmentation without ground truth and 2) a methodology that leverages these probabilistic measures to significantly reduce manual correction while maintaining segmentation quality.View Publication Page
The most established method of reconstructing neural circuits from animals involves slicing tissue very thin, then taking mosaics of electron microscope (EM) images. To trace neurons across different images and through different sections, these images must be accurately aligned, both with the others in the same section and to the sections above and below. Unfortunately, sectioning and imaging are not ideal processes - some of the problems that make alignment difficult include lens distortion, tissue shrinkage during imaging, tears and folds in the sectioned tissue, and dust and other artifacts. In addition the data sets are large (hundreds of thousands of images) and each image must be aligned with many neighbors, so the process must be automated and reliable. This paper discusses methods of dealing with these problems, with numeric results describing the accuracy of the resulting alignments.View Publication Page
A central problem in neuroscience is reconstructing neuronal circuits on the synapse level. Due to a wide range of scales in brain architecture such reconstruction requires imaging that is both high-resolution and high-throughput. Existing electron microscopy (EM) techniques possess required resolution in the lateral plane and either high-throughput or high depth resolution but not both. Here, we exploit recent advances in unsupervised learning and signal processing to obtain high depth-resolution EM images computationally without sacrificing throughput. First, we show that the brain tissue can be represented as a sparse linear combination of localized basis functions that are learned using high-resolution datasets. We then develop compressive sensing-inspired techniques that can reconstruct the brain tissue from very few (typically 5) tomographic views of each section. This enables tracing of neuronal processes and, hence, high throughput reconstruction of neural circuits on the level of individual synapses.View Publication Page
Reconstructing neuronal circuits at the level of synapses is a central problem in neuroscience, and the focus of the nascent field of connectomics. Previously used to reconstruct the C. elegans wiring diagram, serial-section transmission electron microscopy (ssTEM) is a proven technique for the task. However, to reconstruct more complex circuits, ssTEM will require the automation of image processing. We review progress in the processing of electron microscopy images and, in particular, a semi-automated reconstruction pipeline deployed at Janelia. Drosophila circuits underlying identified behaviors are being reconstructed in the pipeline with the goal of generating a complete Drosophila connectome.View Publication Page
This paper addresses the problem of jointly clustering two segmentations of closely correlated images. We focus in particular on the application of reconstructing neuronal structures in over-segmented electron microscopy images. We formulate the problem of co-clustering as a quadratic semi-assignment problem and investigate convex relaxations using semidefinite and linear programming. We further introduce a linear programming method with manageable number of constraints and present an approach for learning the cost function. Our method increases computational efficiency by orders of magnitude while maintaining accuracy, automatically finds the optimal number of clusters, and empirically tends to produce binary assignment solutions. We illustrate our approach in simulations and in experiments with real EM data.View Publication Page