DeepTEGINN: Deep Learning Based Tools to Extract Graphs from Images of Neural Networks

Evi Zouganeli; Gustavo Borges Moreno e Mello; Ioanna Sandvig; Sidney Pontes-Filho; Stefano Nichele; Vibeke Devold Valderhaug

arxiv: 1907.01062 · v1 · pith:TNT5DM2Cnew · submitted 2019-07-01 · 💻 cs.CV · q-bio.NC

DeepTEGINN: Deep Learning Based Tools to Extract Graphs from Images of Neural Networks

Gustavo Borges Moreno e Mello , Vibeke Devold Valderhaug , Sidney Pontes-Filho , Evi Zouganeli , Ioanna Sandvig , Stefano Nichele This is my paper

Pith reviewed 2026-05-25 11:35 UTC · model grok-4.3

classification 💻 cs.CV q-bio.NC

keywords deep learninggraph extractionbrain tissueneural networksimage processinggraph theorytoolboxneuroscience

0 comments

The pith

A deep learning toolbox extracts graphs from brain tissue images to replace manual neuron tracing.

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

The paper presents DeepTEGINN as a toolbox that applies deep learning to convert images of brain cells into graph representations of neuron networks. This approach combines image processing with graph theory so that researchers can study network structure and function without tracing and classifying cells by hand. A sympathetic reader would care because nervous tissue varies widely, and manual methods limit how much data can be processed. If the toolbox works as described, it would let more scientists apply graph-based analysis to large image sets and test ideas about how network layout supports brain computations.

Core claim

The toolbox provides an easy-to-use framework that lets system neuroscientists generate graphs from images of brain tissue by combining image processing, deep learning, and graph theory, serving as a direct alternative to the laborious manual process of tracing, sorting, and classifying neurons.

What carries the argument

The DeepTEGINN toolbox, which integrates deep learning models for computer vision tasks into pipelines that output graphs from heterogeneous brain images.

If this is right

System neuroscientists gain an alternative to manual tracing, sorting, and classifying of neurons in tissue images.
Training and use of deep learning for these computer vision tasks becomes simpler and more accessible.
Graph extraction from large brain image datasets gains time efficiency.
Methods from graph theory and cellular automata become easier to apply to real neuron networks.
Machine learning approaches for this task reach a wider community beyond computer vision specialists.

Where Pith is reading between the lines

These are editorial extensions of the paper, not claims the author makes directly.

If the extracted graphs prove reliable, they could support automated reconstruction of large-scale brain connectivity maps from microscopy stacks.
The same pipeline might be tested on images from other neural preparations to check whether the underlying models generalize beyond the training tissue types.
End-to-end workflows could combine the toolbox output directly with existing graph-analysis libraries to model network dynamics without intermediate manual steps.

Load-bearing premise

Deep learning models trained on brain images will produce accurate graphs across varied nervous tissue without extensive manual correction on new datasets.

What would settle it

Run the toolbox on a new set of brain images, extract the graphs, and compare their connectivity statistics and node properties against graphs produced by independent expert manual tracing on the same images.

Figures

Figures reproduced from arXiv: 1907.01062 by Evi Zouganeli, Gustavo Borges Moreno e Mello, Ioanna Sandvig, Sidney Pontes-Filho, Stefano Nichele, Vibeke Devold Valderhaug.

**Figure 2.** Figure 2: Graph extraction example. A- 256 x 256 Crop of a raw image as acquired from the microscope. B- Segmentation of the electrode area in grey. [PITH_FULL_IMAGE:figures/full_fig_p004_2.png] view at source ↗

read the original abstract

In the brain, the structure of a network of neurons defines how these neurons implement the computations that underlie the mind and the behavior of animals and humans. Provided that we can describe the network of neurons as a graph, we can employ methods from graph theory to investigate its structure or use cellular automata to mathematically assess its function. Although, software for the analysis of graphs and cellular automata are widely available. Graph extraction from the image of networks of brain cells remains difficult. Nervous tissue is heterogeneous, and differences in anatomy may reflect relevant differences in function. Here we introduce a deep learning based toolbox to extracts graphs from images of brain tissue. This toolbox provides an easy-to-use framework allowing system neuroscientists to generate graphs based on images of brain tissue by combining methods from image processing, deep learning, and graph theory. The goals are to simplify the training and usage of deep learning methods for computer vision and facilitate its integration into graph extraction pipelines. In this way, the toolbox provides an alternative to the required laborious manual process of tracing, sorting and classifying. We expect to democratize the machine learning methods to a wider community of users beyond the computer vision experts and improve the time-efficiency of graph extraction from large brain image datasets, which may lead to further understanding of the human mind.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, simulated authors' rebuttal, and a circularity audit. Tearing a paper down is the easy half of reading it; the pith above is the substance, this is the friction.

Desk Editor's Note private letter to a colleague

This is a toolbox description paper that packages existing DL segmentation and graph tools for brain images but supplies zero validation metrics or generalization tests.

read the letter

The paper's core offering is DeepTEGINN, a software framework that wires together standard image-processing steps, off-the-shelf deep learning segmentation, and graph-construction routines so neuroscientists can turn micrographs of neural tissue into usable graphs without writing their own code. That workflow is clearly laid out and the stated intent—to lower the barrier for non-experts and cut down on manual tracing—is straightforward. The authors correctly note that nervous tissue varies and that manual graph extraction is slow, so the motivation is reasonable on its face. Beyond the packaging, nothing new is introduced in algorithms or theory. The text stays at the level of describing intended use and high-level components. The obvious gap is the complete absence of any numbers. There are no training-set sizes, no pixel- or node-level accuracy figures, no comparison against human tracers on held-out images, and no check on how well the models hold up when staining, resolution, or anatomy changes. The central promise that the toolbox serves as a practical alternative therefore rests on an untested transfer assumption. A reader who wants a ready-made pipeline might still find the description useful as a starting point for their own implementation, but anyone evaluating the claim that it reduces laborious manual work will come away empty-handed. This kind of paper is mainly for labs that need a concrete example of how to chain these particular libraries together. It does not contain a result that would change how the field measures or models networks, so I would not bring it to a reading group or cite it. It also does not look ready for peer review; the missing empirical checks are load-bearing for the stated purpose.

Referee Report

1 major / 0 minor

Summary. The manuscript introduces DeepTEGINN, a toolbox that combines image processing, deep learning, and graph theory to extract graphs from images of brain tissue. It positions the tool as an easy-to-use framework for system neuroscientists that simplifies deep-learning training for computer vision tasks and serves as an alternative to manual tracing, sorting, and classification of neurons in heterogeneous nervous tissue.

Significance. If the described pipeline were shown to generalize reliably, the work could improve time-efficiency for processing large brain-image datasets and broaden access to machine-learning methods beyond computer-vision specialists, supporting downstream graph-theoretic or cellular-automaton analyses of neural circuits.

major comments (1)

[Abstract] Abstract: the central claim that the toolbox 'provides an alternative to the required laborious manual process' is unsupported; the text supplies no training-set statistics, cross-dataset performance numbers, ablation studies on tissue heterogeneity, or quantitative comparison against manual ground truth on held-out images.

Simulated Author's Rebuttal

1 responses · 0 unresolved

We thank the referee for highlighting the need for the abstract to be supported by evidence in the manuscript. We address the single major comment below.

read point-by-point responses

Referee: [Abstract] Abstract: the central claim that the toolbox 'provides an alternative to the required laborious manual process' is unsupported; the text supplies no training-set statistics, cross-dataset performance numbers, ablation studies on tissue heterogeneity, or quantitative comparison against manual ground truth on held-out images.

Authors: We agree that the abstract asserts an alternative to manual tracing without accompanying quantitative validation in the manuscript. The paper presents the DeepTEGINN toolbox and its integration of image processing, deep learning, and graph theory as a framework, but does not report training-set sizes, cross-dataset metrics, ablations, or direct comparisons to manual ground truth. We will revise the abstract to remove or qualify the claim of providing a practical alternative to laborious manual processes. revision: yes

Circularity Check

0 steps flagged

No derivation chain or fitted predictions present; software description only.

full rationale

The paper is a description of a toolbox combining image processing, deep learning, and graph theory methods to extract graphs from brain images. No equations, first-principles derivations, parameter fitting, or predictions are claimed or performed. The central claim is that the toolbox simplifies graph extraction as an alternative to manual tracing, but this rests on an untested generalization assumption rather than any self-referential reduction of results to inputs. No self-citations are load-bearing for any derivation, and the work is self-contained as a practical software contribution without circular structure.

Axiom & Free-Parameter Ledger

0 free parameters · 0 axioms · 0 invented entities

The paper introduces a software toolbox without new mathematical derivations, fitted parameters, or postulated physical entities. It relies on standard assumptions from computer vision and neuroscience that are not audited here because only the abstract is available.

pith-pipeline@v0.9.0 · 5789 in / 1099 out tokens · 22188 ms · 2026-05-25T11:35:44.892103+00:00 · methodology

discussion (0)

Reference graph

Works this paper leans on

21 extracted references · 21 canonical work pages · 5 internal anchors

[1]

A review of graph theory application to the facilities layout problem,

M. M. Hassan and G. L. Hogg, “A review of graph theory application to the facilities layout problem,” Omega, vol. 15, no. 4, pp. 291–300, 1987

work page 1987
[2]

Multiscale functional connectivity estimation on low- density neuronal cultures recorded by high-density cmos micro electrode arrays,

A. Maccione, M. Garofalo, T. Nieus, M. Tedesco, L. Berdondini, and S. Martinoia, “Multiscale functional connectivity estimation on low- density neuronal cultures recorded by high-density cmos micro electrode arrays,” Journal of neuroscience methods , vol. 207, no. 2, pp. 161–171, 2012

work page 2012
[3]

Computational matter: Evolving computational functions in nanoscale materials,

H. Broersma, J. F. Miller, and S. Nichele, “Computational matter: Evolving computational functions in nanoscale materials,” in Advances in Unconventional Computing , pp. 397–428, Springer, 2017

work page 2017
[4]

Towards making a cyborg: A closed-loop reservoir-neuro system,

P. Aaser, M. Knudsen, O. H. Ramstad, R. van de Wijdeven, S. Nichele, I. Sandvig, G. Tufte, U. Stefan Bauer, Ø. Halaas, S. Hendseth, et al. , “Towards making a cyborg: A closed-loop reservoir-neuro system,” in Artiﬁcial Life Conference Proceedings 14 , pp. 430–437, MIT Press, 2017

work page 2017
[5]

Deep Reservoir Computing Using Cellular Automata

S. Nichele and A. Molund, “Deep reservoir computing using cellular automata,” arXiv preprint arXiv:1703.02806 , 2017

work page internal anchor Pith review Pith/arXiv arXiv 2017
[6]

Neuroplastic- ity in stroke recovery. the role of microglia in engaging and modifying synapses and networks,

I. Sandvig, I. L. Augestad, A. K. H ˚aberg, and A. Sandvig, “Neuroplastic- ity in stroke recovery. the role of microglia in engaging and modifying synapses and networks,” European Journal of Neuroscience , vol. 47, no. 12, pp. 1414–1428, 2018

work page 2018
[7]

Neﬁ: Network extraction from images,

M. Dirnberger, T. Kehl, and A. Neumann, “Neﬁ: Network extraction from images,” Scientiﬁc reports, vol. 5, p. 15669, 2015

work page 2015
[8]

E. R. Kandel, J. H. Schwartz, T. M. Jessell, D. of Biochemistry, M. B. T. Jessell, S. Siegelbaum, and A. Hudspeth, Principles of neural science , vol. 4. McGraw-hill New York, 2000

work page 2000
[9]

Opencv is an open-source, computer-vision library for extract- ing and processing meaningful data from images.,

B. Gary, “Opencv is an open-source, computer-vision library for extract- ing and processing meaningful data from images.,” Dr. Dobbs, no. 25, pp. 120–126, 2000

work page 2000
[10]

scikit-image: image processing in python,

S. Van der Walt, J. L. Sch ¨onberger, J. Nunez-Iglesias, F. Boulogne, J. D. Warner, N. Yager, E. Gouillart, and T. Yu, “scikit-image: image processing in python,” PeerJ, vol. 2, p. e453, 2014

work page 2014
[11]

Automatic differentiation in py- torch,

A. Paszke, S. Gross, S. Chintala, G. Chanan, E. Yang, Z. DeVito, Z. Lin, A. Desmaison, L. Antiga, and A. Lerer, “Automatic differentiation in py- torch,” Open review preprint https://openreview.net/pdf?id=BJJsrmfCZ, 2017

work page 2017
[12]

A Neural Algorithm of Artistic Style

L. A. Gatys, A. S. Ecker, and M. Bethge, “A neural algorithm of artistic style,” arXiv preprint arXiv:1508.06576 , 2015

work page internal anchor Pith review Pith/arXiv arXiv 2015
[13]

Image inpainting for irregular holes using partial convolutions,

G. Liu, F. A. Reda, K. J. Shih, T.-C. Wang, A. Tao, and B. Catanzaro, “Image inpainting for irregular holes using partial convolutions,” in Proceedings of the European Conference on Computer Vision (ECCV) , pp. 85–100, 2018

work page 2018
[14]

Very Deep Convolutional Networks for Large-Scale Image Recognition

K. Simonyan and A. Zisserman, “Very deep convolutional networks for large-scale image recognition,” arXiv preprint arXiv:1409.1556 , 2014

work page internal anchor Pith review Pith/arXiv arXiv 2014
[15]

YOLOv3: An Incremental Improvement

J. Redmon and A. Farhadi, “Yolov3: An incremental improvement,” arXiv preprint arXiv:1804.02767 , 2018

work page internal anchor Pith review Pith/arXiv arXiv 2018
[16]

An improved watershed algorithm based on efﬁcient computation of shortest paths,

V . Osma-Ruiz, J. I. Godino-Llorente, N. S ´aenz-Lech´on, and P. G ´omez- Vilda, “An improved watershed algorithm based on efﬁcient computation of shortest paths,” Pattern Recognition, vol. 40, no. 3, pp. 1078–1090, 2007

work page 2007
[17]

W-Net: A Deep Model for Fully Unsupervised Image Segmentation

X. Xia and B. Kulis, “W-net: A deep model for fully unsupervised image segmentation,” arXiv preprint arXiv:1711.08506 , 2017

work page internal anchor Pith review Pith/arXiv arXiv 2017
[18]

Improved zhang-suen thinning algorithm in binary line drawing applications,

W. Chen, L. Sui, Z. Xu, and Y . Lang, “Improved zhang-suen thinning algorithm in binary line drawing applications,” in 2012 International Conference on Systems and Informatics (ICSAI2012) , pp. 1947–1950, IEEE, 2012

work page 2012
[19]

Exploring network structure, dy- namics, and function using networkx,

A. Hagberg, P. Swart, and D. S Chult, “Exploring network structure, dy- namics, and function using networkx,” tech. rep., Los Alamos National Lab.(LANL), Los Alamos, NM (United States), 2008

work page 2008
[20]

Functional connectivity estimation over large networks at cellular resolution based on electrophysiological recordings and structural prior,

S. Ullo, T. R. Nieus, D. Sona, A. Maccione, L. Berdondini, and V . Murino, “Functional connectivity estimation over large networks at cellular resolution based on electrophysiological recordings and structural prior,” Frontiers in Neuroanatomy, vol. 8, p. 137, 2014

work page 2014
[21]

Microsoft coco: Common objects in context,

T.-Y . Lin, M. Maire, S. Belongie, J. Hays, P. Perona, D. Ramanan, P. Doll ´ar, and C. L. Zitnick, “Microsoft coco: Common objects in context,” in European conference on computer vision , pp. 740–755, Springer, 2014

work page 2014

[1] [1]

A review of graph theory application to the facilities layout problem,

M. M. Hassan and G. L. Hogg, “A review of graph theory application to the facilities layout problem,” Omega, vol. 15, no. 4, pp. 291–300, 1987

work page 1987

[2] [2]

Multiscale functional connectivity estimation on low- density neuronal cultures recorded by high-density cmos micro electrode arrays,

A. Maccione, M. Garofalo, T. Nieus, M. Tedesco, L. Berdondini, and S. Martinoia, “Multiscale functional connectivity estimation on low- density neuronal cultures recorded by high-density cmos micro electrode arrays,” Journal of neuroscience methods , vol. 207, no. 2, pp. 161–171, 2012

work page 2012

[3] [3]

Computational matter: Evolving computational functions in nanoscale materials,

H. Broersma, J. F. Miller, and S. Nichele, “Computational matter: Evolving computational functions in nanoscale materials,” in Advances in Unconventional Computing , pp. 397–428, Springer, 2017

work page 2017

[4] [4]

Towards making a cyborg: A closed-loop reservoir-neuro system,

P. Aaser, M. Knudsen, O. H. Ramstad, R. van de Wijdeven, S. Nichele, I. Sandvig, G. Tufte, U. Stefan Bauer, Ø. Halaas, S. Hendseth, et al. , “Towards making a cyborg: A closed-loop reservoir-neuro system,” in Artiﬁcial Life Conference Proceedings 14 , pp. 430–437, MIT Press, 2017

work page 2017

[5] [5]

Deep Reservoir Computing Using Cellular Automata

S. Nichele and A. Molund, “Deep reservoir computing using cellular automata,” arXiv preprint arXiv:1703.02806 , 2017

work page internal anchor Pith review Pith/arXiv arXiv 2017

[6] [6]

Neuroplastic- ity in stroke recovery. the role of microglia in engaging and modifying synapses and networks,

I. Sandvig, I. L. Augestad, A. K. H ˚aberg, and A. Sandvig, “Neuroplastic- ity in stroke recovery. the role of microglia in engaging and modifying synapses and networks,” European Journal of Neuroscience , vol. 47, no. 12, pp. 1414–1428, 2018

work page 2018

[7] [7]

Neﬁ: Network extraction from images,

M. Dirnberger, T. Kehl, and A. Neumann, “Neﬁ: Network extraction from images,” Scientiﬁc reports, vol. 5, p. 15669, 2015

work page 2015

[8] [8]

E. R. Kandel, J. H. Schwartz, T. M. Jessell, D. of Biochemistry, M. B. T. Jessell, S. Siegelbaum, and A. Hudspeth, Principles of neural science , vol. 4. McGraw-hill New York, 2000

work page 2000

[9] [9]

Opencv is an open-source, computer-vision library for extract- ing and processing meaningful data from images.,

B. Gary, “Opencv is an open-source, computer-vision library for extract- ing and processing meaningful data from images.,” Dr. Dobbs, no. 25, pp. 120–126, 2000

work page 2000

[10] [10]

scikit-image: image processing in python,

S. Van der Walt, J. L. Sch ¨onberger, J. Nunez-Iglesias, F. Boulogne, J. D. Warner, N. Yager, E. Gouillart, and T. Yu, “scikit-image: image processing in python,” PeerJ, vol. 2, p. e453, 2014

work page 2014

[11] [11]

Automatic differentiation in py- torch,

A. Paszke, S. Gross, S. Chintala, G. Chanan, E. Yang, Z. DeVito, Z. Lin, A. Desmaison, L. Antiga, and A. Lerer, “Automatic differentiation in py- torch,” Open review preprint https://openreview.net/pdf?id=BJJsrmfCZ, 2017

work page 2017

[12] [12]

A Neural Algorithm of Artistic Style

L. A. Gatys, A. S. Ecker, and M. Bethge, “A neural algorithm of artistic style,” arXiv preprint arXiv:1508.06576 , 2015

work page internal anchor Pith review Pith/arXiv arXiv 2015

[13] [13]

Image inpainting for irregular holes using partial convolutions,

G. Liu, F. A. Reda, K. J. Shih, T.-C. Wang, A. Tao, and B. Catanzaro, “Image inpainting for irregular holes using partial convolutions,” in Proceedings of the European Conference on Computer Vision (ECCV) , pp. 85–100, 2018

work page 2018

[14] [14]

Very Deep Convolutional Networks for Large-Scale Image Recognition

K. Simonyan and A. Zisserman, “Very deep convolutional networks for large-scale image recognition,” arXiv preprint arXiv:1409.1556 , 2014

work page internal anchor Pith review Pith/arXiv arXiv 2014

[15] [15]

YOLOv3: An Incremental Improvement

J. Redmon and A. Farhadi, “Yolov3: An incremental improvement,” arXiv preprint arXiv:1804.02767 , 2018

work page internal anchor Pith review Pith/arXiv arXiv 2018

[16] [16]

An improved watershed algorithm based on efﬁcient computation of shortest paths,

V . Osma-Ruiz, J. I. Godino-Llorente, N. S ´aenz-Lech´on, and P. G ´omez- Vilda, “An improved watershed algorithm based on efﬁcient computation of shortest paths,” Pattern Recognition, vol. 40, no. 3, pp. 1078–1090, 2007

work page 2007

[17] [17]

W-Net: A Deep Model for Fully Unsupervised Image Segmentation

X. Xia and B. Kulis, “W-net: A deep model for fully unsupervised image segmentation,” arXiv preprint arXiv:1711.08506 , 2017

work page internal anchor Pith review Pith/arXiv arXiv 2017

[18] [18]

Improved zhang-suen thinning algorithm in binary line drawing applications,

W. Chen, L. Sui, Z. Xu, and Y . Lang, “Improved zhang-suen thinning algorithm in binary line drawing applications,” in 2012 International Conference on Systems and Informatics (ICSAI2012) , pp. 1947–1950, IEEE, 2012

work page 2012

[19] [19]

Exploring network structure, dy- namics, and function using networkx,

A. Hagberg, P. Swart, and D. S Chult, “Exploring network structure, dy- namics, and function using networkx,” tech. rep., Los Alamos National Lab.(LANL), Los Alamos, NM (United States), 2008

work page 2008

[20] [20]

Functional connectivity estimation over large networks at cellular resolution based on electrophysiological recordings and structural prior,

S. Ullo, T. R. Nieus, D. Sona, A. Maccione, L. Berdondini, and V . Murino, “Functional connectivity estimation over large networks at cellular resolution based on electrophysiological recordings and structural prior,” Frontiers in Neuroanatomy, vol. 8, p. 137, 2014

work page 2014

[21] [21]

Microsoft coco: Common objects in context,

T.-Y . Lin, M. Maire, S. Belongie, J. Hays, P. Perona, D. Ramanan, P. Doll ´ar, and C. L. Zitnick, “Microsoft coco: Common objects in context,” in European conference on computer vision , pp. 740–755, Springer, 2014

work page 2014