Global Neuron Shape Reasoning with Point Affinity Transformers DOI Creative Commons

Jakob Troidl,

Johannes Knittel, Wanhua Li

et al.

bioRxiv (Cold Spring Harbor Laboratory), Journal Year: 2024, Volume and Issue: unknown

Published: Nov. 24, 2024

Connectomics is a subfield of neuroscience that aims to map the brain's intricate wiring diagram. Accurate neuron segmentation from microscopy volumes essential for automating connectome reconstruction. However, current state-of-the-art algorithms use image-based convolutional neural networks are limited local shape context. Thus, we introduce new framework reasons over global with novel point affinity transformer. Our embeds (multi-)neuron cloud into fixed-length feature set which can decode any pair affinities, enabling clustering clouds automatic proofreading. We also show learned easily be mapped contrastive embedding space enables type classification using simple KNN classifier. approach excels in two demanding connectomics tasks: proofreading errors and classifying types. Evaluated on three benchmark datasets derived connectomes, our method outperforms transformers, graph networks, unsupervised baselines.

Language: Английский

Neurotransmitter classification from electron microscopy images at synaptic sites in Drosophila melanogaster DOI Creative Commons
Nils Eckstein, Alexander Shakeel Bates, Andrew Champion

et al.

Cell, Journal Year: 2024, Volume and Issue: 187(10), P. 2574 - 2594.e23

Published: May 1, 2024

High-resolution electron microscopy of nervous systems has enabled the reconstruction synaptic connectomes. However, we do not know sign for each connection (i.e., whether a is excitatory or inhibitory), which implied by released transmitter. We demonstrate that artificial neural networks can predict transmitter types presynapses from micrographs: network trained to six transmitters (acetylcholine, glutamate, GABA, serotonin, dopamine, octopamine) achieves an accuracy 87% individual synapses, 94% neurons, and 91% known cell across D. melanogaster whole brain. visualize ultrastructural features used prediction, discovering subtle but significant differences between phenotypes. also analyze distributions brain find neurons develop together largely express only one fast-acting GABA). hope our publicly available predictions act as accelerant neuroscientific hypothesis generation fly.

Language: Английский

Citations

120
NEURD offers automated proofreading and feature extraction for connectomics DOI Creative Commons
Brendan Celii, Stelios Papadopoulos, Zhuokun Ding

et al.

Nature, Journal Year: 2025, Volume and Issue: 640(8058), P. 487 - 496

Published: April 9, 2025

We are in the era of millimetre-scale electron microscopy volumes collected at nanometre resolution1,2. Dense reconstruction cellular compartments these has been enabled by recent advances machine learning3-6. Automated segmentation methods produce exceptionally accurate reconstructions cells, but post hoc proofreading is still required to generate large connectomes that free merge and split errors. The elaborate 3D meshes neurons contain detailed morphological information multiple scales, from diameter, shape branching patterns axons dendrites, down fine-scale structure dendritic spines. However, extracting features can require substantial effort piece together existing tools into custom workflows. Here, building on open source software for mesh manipulation, we present Neural Decomposition (NEURD), a package decomposes meshed compact extensively annotated graph representations. With feature-rich graphs, automate variety tasks such as state-of-the-art automated errors, cell classification, spine detection, axonal-dendritic proximities other annotations. These enable many downstream analyses neural morphology connectivity, making massive complex datasets more accessible neuroscience researchers.

Language: Английский

Citations

4
Connectomic reconstruction of a cortical column DOI Open Access
Meike Sievers, Alessandro Motta,

M. P. Schmidt

et al.

bioRxiv (Cold Spring Harbor Laboratory), Journal Year: 2024, Volume and Issue: unknown

Published: March 25, 2024

ABSTRACT The cerebral cortex of mammals has long been proposed to comprise unit-modules, so-called cortical columns. detailed synaptic-level circuitry such a neuronal network about 10 4 neurons is still unknown. Here, using 3-dimensional electron microscopy, AI-based image processing and automated proofreading, we report the connectomic reconstruction defined column in mouse barrel cortex. appears as structural feature connectome, without need for geometrical or morphological landmarks. We then used connectome definition cell types column, determine intracolumnar circuit modules, analyze logic inhibitory circuits, investigate circuits combination bottom-up top-down signals specificity input, search higher-order structure within homogeneous populations, estimate degree symmetry Hebbian learning various connection types. With this, provide first column-level description cortex, likely substrate mechanistic understanding sensory-conceptual integration learning.

Language: Английский

Citations

9
CAVE: Connectome Annotation Versioning Engine DOI Creative Commons
Sven Dorkenwald, Casey M Schneider-Mizell, Derrick Brittain

et al.

Nature Methods, Journal Year: 2025, Volume and Issue: unknown

Published: April 9, 2025

Abstract Advances in electron microscopy, image segmentation and computational infrastructure have given rise to large-scale richly annotated connectomic datasets, which are increasingly shared across communities. To enable collaboration, users need be able concurrently create annotations correct errors the automated by proofreading. In large every proofreading edit relabels cell identities of millions voxels thousands like synapses. For analysis, require immediate reproducible access this changing expanding data landscape. Here we present Connectome Annotation Versioning Engine (CAVE), a that provides scalable solutions for flexible annotation support fast analysis queries at arbitrary time points. Deployed as suite web services, CAVE empowers distributed communities perform connectome up petascale datasets (~1 mm 3 ) while annotating is ongoing.

Language: Английский

Citations

1
CAVE: Connectome Annotation Versioning Engine DOI Creative Commons
Sven Dorkenwald, Casey M Schneider-Mizell, Derrick Brittain

et al.

bioRxiv (Cold Spring Harbor Laboratory), Journal Year: 2023, Volume and Issue: unknown

Published: July 28, 2023

Abstract Advances in Electron Microscopy, image segmentation and computational infrastructure have given rise to large-scale richly annotated connectomic datasets which are increasingly shared across communities. To enable collaboration, users need be able concurrently create new annotations correct errors the automated by proofreading. In large datasets, every proofreading edit relabels cell identities of millions voxels thousands like synapses. For analysis, require immediate reproducible access this constantly changing expanding data landscape. Here, we present Connectome Annotation Versioning Engine (CAVE), a for connectome analysis up-to petascale (∼1mm 3 ) while annotating is ongoing. segmentation, CAVE provides distributed continuous versioning reconstructions. Annotations defined locations such that they can quickly assigned underlying segment enables fast queries CAVE’s arbitrary time points. supports schematized, extensible annotations, so researchers readily design novel annotation types. already used many connectomics including largest available date.

Language: Английский

Citations

19
A guide to CNN-based dense segmentation of neuronal EM images DOI Creative Commons
Hidetoshi Urakubo

Microscopy, Journal Year: 2025, Volume and Issue: unknown

Published: Jan. 13, 2025

Abstract Large-scale reconstitution of neuronal circuits from volumetric electron microscopy images is a remarkable research goal in neuroanatomy. However, the large-scale reconstruction result automatic segmentation using convolutional neural networks (CNNs), which still challenging for general researchers to perform. This review focuses on two representative CNNs dense segmentation: flood-filling (FFN) and local shape descriptors (LSD)-predicting U-Net (LSD network). It outlines their basic mechanisms, requirements, output author’s example segmentation. The FFN excels segmenting long axons, LSD network adept at myelinated axons. choice between depends target, as neither universally superior. A common limitation easy detachment thin spines parent dendrites, fundamentally unavoidable. author also introduces proposed mitigate this issue. As CNN-based automated can take months, need be aware selection an appropriate CNN, required computer resources, fundamental limitations. serves guide such

Language: Английский

Citations

0
Comparative prospects of imaging methods for whole-brain mammalian connectomics DOI Creative Commons
Logan Thrasher Collins,

Todd Huffman,

Randal A. Koene

et al.

Cell Reports Methods, Journal Year: 2025, Volume and Issue: unknown, P. 100988 - 100988

Published: Feb. 1, 2025

Language: Английский

Citations

0
AI-guided immersive exploration of brain ultrastructure for collaborative analysis and education DOI Creative Commons
Uzair Shah, Marco Agus, Daniya Boges

et al.

Computers & Graphics, Journal Year: 2025, Volume and Issue: unknown, P. 104239 - 104239

Published: May 1, 2025

Language: Английский

Citations

0
A milestone map of mouse-brain connectivity reveals challenging new terrain for scientists DOI Creative Commons

Michael Eisenstein

Nature, Journal Year: 2024, Volume and Issue: 628(8008), P. 677 - 679

Published: April 15, 2024

A pioneering 'connectomics' collaboration has successfully reconstructed one cubic millimetre of brain tissue, but researchers are still just scratching the surface complexity it contains.

Language: Английский

Citations

3
A general algorithm for consensus 3D cell segmentation from 2D segmented stacks DOI Creative Commons
Felix Zhou, Clarence Yapp, Zhiguo Shang

et al.

bioRxiv (Cold Spring Harbor Laboratory), Journal Year: 2024, Volume and Issue: unknown

Published: May 6, 2024

Cell segmentation is the fundamental task. Only by segmenting, can we define quantitative spatial unit for collecting measurements to draw biological conclusions. Deep learning has revolutionized 2D cell segmentation, enabling generalized solutions across types and imaging modalities. This been driven ease of scaling up image acquisition, annotation computation. However 3D which requires dense slices still poses significant challenges. Labelling every in slice prohibitive. Moreover it ambiguous, necessitating cross-referencing with other orthoviews. Lastly, there limited ability unambiguously record visualize 1000's annotated cells. Here develop a theory toolbox, u-Segment3D 2D-to-3D compatible any method. Given optimal segmentations, generates without data training, as demonstrated on 11 real life datasets, >70,000 cells, spanning single aggregates tissue.

Language: Английский

Citations

2