Cellular-resolution connectomics: challenges of dense neural circuit reconstruction
TL;DR: The aim of this perspective is to summarize and quantify the challenges for data analysis in cellular-resolution connectomics and describe current solutions involving online crowd-sourcing and machine-learning approaches.
Abstract: Neuronal networks are high-dimensional graphs that are packed into three-dimensional nervous tissue at extremely high density. Comprehensively mapping these networks is therefore a major challenge. Although recent developments in volume electron microscopy imaging have made data acquisition feasible for circuits comprising a few hundreds to a few thousands of neurons, data analysis is massively lagging behind. The aim of this perspective is to summarize and quantify the challenges for data analysis in cellular-resolution connectomics and describe current solutions involving online crowd-sourcing and machine-learning approaches.
Citations
More filters
TL;DR: Hydrogel-based structures can be built from within biological tissue to allow subsequent removal of lipids without mechanical disassembly of the tissue, creating a tissue-hydrogel hybrid that is physically stable, that preserves fine structure, proteins and nucleic acids, and that is permeable to both visible-spectrum photons and exogenous macromolecules.
Abstract: With potential relevance for brain-mapping work, hydrogel-based structures can now be built from within biological tissue to allow subsequent removal of lipids without mechanical disassembly of the tissue. This process creates a tissue-hydrogel hybrid that is physically stable, that preserves fine structure, proteins and nucleic acids, and that is permeable to both visible-spectrum photons and exogenous macromolecules. Here we highlight relevant challenges and opportunities of this approach, especially with regard to integration with complementary methodologies for brain-mapping studies.
676 citations
TL;DR: Results suggest that this method for blood vessel segmentation in fundus images based on a discriminatively trained fully connected conditional random field model is suitable for the task of segmenting elongated structures, a feature that can be exploited to contribute with other medical and biological applications.
Abstract: Goal: In this work, we present an extensive description and evaluation of our method for blood vessel segmentation in fundus images based on a discriminatively trained fully connected conditional random field model. Methods: Standard segmentation priors such as a Potts model or total variation usually fail when dealing with thin and elongated structures. We overcome this difficulty by using a conditional random field model with more expressive potentials, taking advantage of recent results enabling inference of fully connected models almost in real time. Parameters of the method are learned automatically using a structured output support vector machine, a supervised technique widely used for structured prediction in a number of machine learning applications. Results: Our method, trained with state of the art features, is evaluated both quantitatively and qualitatively on four publicly available datasets: DRIVE, STARE, CHASEDB1, and HRF. Additionally, a quantitative comparison with respect to other strategies is included. Conclusion: The experimental results show that this approach outperforms other techniques when evaluated in terms of sensitivity, F1-score, G-mean, and Matthews correlation coefficient. Additionally, it was observed that the fully connected model is able to better distinguish the desired structures than the local neighborhood-based approach. Significance: Results suggest that this method is suitable for the task of segmenting elongated structures, a feature that can be exploited to contribute with other medical and biological applications.
429 citations
Cites background from "Cellular-resolution connectomics: c..."
...This property can potentially contribute to a number of different biological and medical applications where the segmentation of such structures is required, including automatic plant root phenotyping [34] or neuron analysis [35]....
[...]
TL;DR: It is argued that the cortical architecture is more heterogeneous than Brodmann's map suggests, and a triple-scale concept is proposed that includes repetitive modular-like structures and micro- and meso-maps.
342 citations
Cites background or methods from "Cellular-resolution connectomics: c..."
...The dissection of human brains into small tissue blocks is inevitable in electron microscopy (Denk and Horstmann, 2004; Helmstaedter, 2013; Lichtman and Denk, 2011) and optical coherence tomography (Magnain et al., 2015; Wang et al., 2011)....
[...]
...Besides an incredible amount of time and effort for preparation and measurement, the methods require demanding, and in some applications presently not available, computing resources to achieve the mapping of large brain regions or even a whole human brain (Helmstaedter, 2013)....
[...]
...In the near future, these storage space requirements will increase to several petabytes per dataset (Helmstaedter, 2013)....
[...]
TL;DR: It is demonstrated that X-ray ptychography—a high-resolution coherent diffractive imaging technique—can create three-dimensional images of integrated circuits of known and unknown designs with a lateral resolution in all directions down to 14.6 nanometres.
Abstract: A recently developed computational imaging technique, X-ray ptychographic tomography, is used to study integrated circuits, and a 3D image of a processor chip with a resolution of 14.6 nm is obtained. As computer chips have become increasingly crammed with nanometre-scale devices and circuitry, new microscopy techniques that can resolve the smallest features are required to enable chip design and inspection. X-ray imaging is uniquely suited for non-destructive, high-resolution imaging and Mirko Holler et al. make use of a recently developed computational imaging technique, X-ray ptychography, to generate high-resolution three-dimensional images of integrated circuits. They test X-ray ptychography on a circuit with known features, and then apply it to an Intel processor chip manufactured in the 22-nanometre technology, obtaining detailed three-dimensional maps of the devices with a resolution down to 14.6 nanometres. This technique could be used to assist quality control during chip production. Modern nanoelectronics1,2 has advanced to a point at which it is impossible to image entire devices and their interconnections non-destructively because of their small feature sizes and the complex three-dimensional structures resulting from their integration on a chip. This metrology gap implies a lack of direct feedback between design and manufacturing processes, and hampers quality control during production, shipment and use. Here we demonstrate that X-ray ptychography3,4—a high-resolution coherent diffractive imaging technique—can create three-dimensional images of integrated circuits of known and unknown designs with a lateral resolution in all directions down to 14.6 nanometres. We obtained detailed device geometries and corresponding elemental maps, and show how the devices are integrated with each other to form the chip. Our experiments represent a major advance in chip inspection and reverse engineering over the traditional destructive electron microscopy and ion milling techniques5,6,7. Foreseeable developments in X-ray sources8, optics9 and detectors10, as well as adoption of an instrument geometry11 optimized for planar rather than cylindrical samples, could lead to a thousand-fold increase in efficiency, with concomitant reductions in scan times and voxel sizes.
312 citations
TL;DR: The recent introduction of integrated light and electron microscopy systems will revolutionise correlative light and volume electron microscopeopy studies, by enabling the sequential collection of data fromLight and electron imaging modalities without intermediate specimen manipulation.
294 citations
Cites background or methods from "Cellular-resolution connectomics: c..."
...reconstruction of retina (Helmstaedter, 2013), or the 2283 years required for an individual to reconstruct 1 mm3 of mammalian cortex (Knott and Genoud, 2013) highlight the importance of progress in the fast-growing field dedicated to EM image analysis....
[...]
...In the meantime, alternative approaches have included recruitment of ‘trained’ human annotators (Helmstaedter et al., 2013), and socalled ‘crowd sourcing’ initiatives, which exploit the online power of citizen science to speed up data analysis (Giuly et al., 2013; Helmstaedter, 2013)....
[...]
...Where an attractive faç ade can be developed to present the data (and not forgetting that developing this framework, enabling widespread online access, and subsequent proofreading are additional steps in the process), these initiatives may yet recruit the hundreds of thousands of hours needed to analyse complex datasets (Helmstaedter, 2013)....
[...]
..., 2013), and socalled ‘crowd sourcing’ initiatives, which exploit the online power of citizen science to speed up data analysis (Giuly et al., 2013; Helmstaedter, 2013)....
[...]
...In ddition, the quantities of data postulated to arise from future studes are probably underestimates (Briggman and Bock, 2012; Denk t al., 2012; Helmstaedter, 2013; Knott and Genoud, 2013; Marx, 013)....
[...]
References
More filters
TL;DR: The structure and connectivity of the nervous system of the nematode Caenorhabditis elegans has been deduced from reconstructions of electron micrographs of serial sections as discussed by the authors.
Abstract: The structure and connectivity of the nervous system of the nematode Caenorhabditis elegans has been deduced from reconstructions of electron micrographs of serial sections. The hermaphrodite nervous system has a total complement of 302 neurons, which are arranged in an essentially invariant structure. Neurons with similar morphologies and connectivities have been grouped together into classes; there are 118 such classes. Neurons have simple morphologies with few, if any, branches. Processes from neurons run in defined positions within bundles of parallel processes, synaptic connections being made en passant. Process bundles are arranged longitudinally and circumferentially and are often adjacent to ridges of hypodermis. Neurons are generally highly locally connected, making synaptic connections with many of their neighbours. Muscle cells have arms that run out to process bundles containing motoneuron axons. Here they receive their synaptic input in defined regions along the surface of the bundles, where motoneuron axons reside. Most of the morphologically identifiable synaptic connections in a typical animal are described. These consist of about 5000 chemical synapses, 2000 neuromuscular junctions and 600 gap junctions.
5,491 citations
TL;DR: The findings challenge the common view that humans stand out from other primates in their brain composition and indicate that, with regard to numbers of neuronal and nonneuronal cells, the human brain is an isometrically scaled‐up primate brain.
Abstract: The human brain is often considered to be the most cognitively capable among mammalian brains and to be much larger than expected for a mammal of our body size. Although the number of neurons is generally assumed to be a determinant of computational power, and despite the widespread quotes that the human brain contains 100 billion neurons and ten times more glial cells, the absolute number of neurons and glial cells in the human brain remains unknown. Here we determine these numbers by using the isotropic fractionator and compare them with the expected values for a human-sized primate. We find that the adult male human brain contains on average 86.1 +/- 8.1 billion NeuN-positive cells ("neurons") and 84.6 +/- 9.8 billion NeuN-negative ("nonneuronal") cells. With only 19% of all neurons located in the cerebral cortex, greater cortical size (representing 82% of total brain mass) in humans compared with other primates does not reflect an increased relative number of cortical neurons. The ratios between glial cells and neurons in the human brain structures are similar to those found in other primates, and their numbers of cells match those expected for a primate of human proportions. These findings challenge the common view that humans stand out from other primates in their brain composition and indicate that, with regard to numbers of neuronal and nonneuronal cells, the human brain is an isometrically scaled-up primate brain.
1,818 citations
TL;DR: It is demonstrated that datasets meeting these requirements can be obtained by automated block-face imaging combined with serial sectioning inside the chamber of a scanning electron microscope, opening the possibility of automatically obtaining the electron-microscope-level 3D datasets needed to completely reconstruct the connectivity of neuronal circuits.
Abstract: Three-dimensional (3D) structural information on many length scales is of central importance in biological research. Excellent methods exist to obtain structures of molecules at atomic, organelles at electron microscopic, and tissue at light-microscopic resolution. A gap exists, however, when 3D tissue structure needs to be reconstructed over hundreds of micrometers with a resolution sufficient to follow the thinnest cellular processes and to identify small organelles such as synaptic vesicles. Such 3D data are, however, essential to understand cellular networks that, particularly in the nervous system, need to be completely reconstructed throughout a substantial spatial volume. Here we demonstrate that datasets meeting these requirements can be obtained by automated block-face imaging combined with serial sectioning inside the chamber of a scanning electron microscope. Backscattering contrast is used to visualize the heavy-metal staining of tissue prepared using techniques that are routine for transmission electron microscopy. Low-vacuum (20–60 Pa H2O) conditions prevent charging of the uncoated block face. The resolution is sufficient to trace even the thinnest axons and to identify synapses. Stacks of several hundred sections, 50–70 nm thick, have been obtained at a lateral position jitter of typically under 10 nm. This opens the possibility of automatically obtaining the electron-microscope-level 3D datasets needed to completely reconstruct the connectivity of neuronal circuits.
1,506 citations
TL;DR: Foldit is described, a multiplayer online game that engages non-scientists in solving hard prediction problems and shows that top-ranked Foldit players excel at solving challenging structure refinement problems in which substantial backbone rearrangements are necessary to achieve the burial of hydrophobic residues.
Abstract: A natural polypeptide chain can fold into a native protein in microseconds, but predicting such stable three-dimensional structure from any given amino-acid sequence and first physical principles remains a formidable computational challenge. Aiming to recruit human visual and strategic powers to the task, Seth Cooper, David Baker and colleagues turned their 'Rosetta' structure-prediction algorithm into an online multiplayer game called Foldit, in which thousands of non-scientists competed and collaborated to produce a rich set of new algorithms and search strategies for protein structure refinement. The work shows that even computationally complex scientific problems can be effectively crowd-sourced using interactive multiplayer games. Predicting the structure of a folded protein from first principles for any given amino-acid sequence remains a formidable computational challenge. To recruit human abilities to the task, these authors turned their Rosetta structure prediction algorithm into an online multiplayer game in which thousands of non-scientists competed and collaborated to produce new algorithms and search strategies for protein structure refinement. This shows that computationally complex problems can be effectively 'crowd-sourced' through interactive multiplayer games. People exert large amounts of problem-solving effort playing computer games. Simple image- and text-recognition tasks have been successfully ‘crowd-sourced’ through games1,2,3, but it is not clear if more complex scientific problems can be solved with human-directed computing. Protein structure prediction is one such problem: locating the biologically relevant native conformation of a protein is a formidable computational challenge given the very large size of the search space. Here we describe Foldit, a multiplayer online game that engages non-scientists in solving hard prediction problems. Foldit players interact with protein structures using direct manipulation tools and user-friendly versions of algorithms from the Rosetta structure prediction methodology4, while they compete and collaborate to optimize the computed energy. We show that top-ranked Foldit players excel at solving challenging structure refinement problems in which substantial backbone rearrangements are necessary to achieve the burial of hydrophobic residues. Players working collaboratively develop a rich assortment of new strategies and algorithms; unlike computational approaches, they explore not only the conformational space but also the space of possible search strategies. The integration of human visual problem-solving and strategy development capabilities with traditional computational algorithms through interactive multiplayer games is a powerful new approach to solving computationally-limited scientific problems.
1,265 citations
TL;DR: It is argued that the time is right to begin assimilating the wealth of data that has been accumulated over the past century and start building biologically accurate models of the brain from first principles to aid the understanding of brain function and dysfunction.
Abstract: Markram describes the impressive aims of the Blue Brain Project, in which the enormous computing power of IBM's Blue Gene supercomputer is being harnessed to build biologically accurate models of the neocortical column and, ultimately, the whole brain.
1,243 citations