Reconstructing the brain: from image stacks to neuron synthesis

Abstract

Large-scale brain initiatives such as the US BRAIN initiative and the European Human Brain Project aim to marshall a vast amount of data and tools for the purpose of furthering our understanding of brains. Fundamental to this goal is that neuronal morphologies must be seamlessly reconstructed and aggregated on scales up to the whole rodent brain. The experimental labor needed to manually produce this number of digital morphologies is prohibitively large. The BigNeuron initiative is assembling community-generated, open-source, automated reconstruction algorithms into an open platform, and is beginning to generate an increasing flow of high-quality reconstructed neurons. We propose a novel extension of this workflow to use this data stream to generate an unlimited number of statistically equivalent, yet distinct, digital morphologies. This will bring automated processing of reconstructed cells into digital neurons to the wider neuroscience community, and enable a range of morphologically accurate computational models.

Computational synthesis of cortical dendritic morphologies

“…These efforts took the study of neuronal morphologies to the next era of digital reconstructions (Winnubst et al, 2019, Gouwens et al, 2019. Digital reconstructions of physiologically realistic neuronal networks (Markram et al, 2015;Egger et al, 2014) rely on distinct neuronal shapes of different cell types to approximate biological diversity (Shillcock et al, 2016;Landau et al, 2016;Ramaswamy et al, 2012). We are still far, however, from having enough reconstructions of unique morphologies to populate biologically realistic networks of a whole brain region (1 M neurons for the mouse somatosensory cortex, 10 M neurons for the mouse isocortex) (Ero et al, 2018;Herculano-Houzel et al, 2006), due to the expensive and tedious process of neuronal reconstructions (Farhoodi et al, 2019).…”

Brain Mapping and Synapse Quantification In vivo: It's Time to Imaging

“…How are these new neuroanatomical data fitting into- or complementing- other alternative scientific strategies to achieve brain mapping? I will consider several ongoing extraordinarily ambitious projects dedicated to reconstructing the brains of mammals by tackling one of their essential building blocks: The neurons (Shillcock et al, 2016 ). Among them, the BigNeuron initiative project ( http://bigneuron.org ) provides neuronal reconstruction algorithms in one open-source platform, enabling researchers to compare and test their own algorithms with a large set of image slices.…”

BigNeuron: A resource to benchmark and predict best-performing algorithms for automated reconstruction of neuronal morphology

“…A faithful annotation of the fine details of neuron morphology is particularly relevant for estimating potential connectivity, given that artifacts introduce biases in the estimation of connections between brain regions. This is even more relevant when whole brain modeling techniques rely on synthetic generation of neuronal populations based on available annotation data (Shillcock et al 2016). Similarly, fine-scale neuromorphological properties such as the radius of the neurite segments have a crucial impact on the electrical modeling of neurons, and seemingly small artifacts in the reconstruction process can lead to fully altered tree topology leading to unreliable simulation of signal integration and transmission.…”