Speaker
Description
Accurate artery–vein labelling is essential for the analysis of whole-brain vasculature and blood flow simulations to understand cerebrovascular organisation and function. However, most datasets lack artery–vein annotations~\cite{todorov}, since labeling is generally possible only at acquisition using specific staining protocols~\cite{kirst}.
Despite reported topological differences between penetrating arteries and veins~\cite{schmid}, whole‑brain manual labeling is impractical, motivating automated mathematical approaches. Using graph‑based representations of penetrating trees extracted from whole‑brain mouse vasculature with acquisition‑derived artery–vein labels~\cite{kirst}, we apply computational topology to compute a topological morphological descriptor (TMD)~\cite{kanari, beers}. TMD captures topological and geometric features from vascular trees and encodes them as persistence barcodes, which are then transformed into persistent images.
Using persistent images to train a convolutional neural network, we achieve classification accuracies exceeding 80\%, demonstrating that persistent homology provides discriminative power for vascular labeling. A Random Forest classifier trained on alternative features reaches similar accuracy, showing that artery–vein separation is learnable across different modeling choices. Finally, the framework generalizes to graph‑based unlabeled datasets, enabling scalable, mathematically grounded artery–vein classification.
Bibliography
@article{kirst,
title={Mapping the fine-scale organization and plasticity of the brain vasculature},
author={Kirst, Christoph and Skriabine, Sophie and Vieites-Prado, Alba and Topilko, Thomas and Bertin, Paul and Gerschenfeld, Gaspard and Verny, Florine and Topilko, Piotr and Michalski, Nicolas and Tessier-Lavigne, Marc and others},
journal={Cell},
volume={180},
number={4},
pages={780--795},
year={2020},
publisher={Elsevier}
}
@article{todorov,
title={Machine learning analysis of whole mouse brain vasculature},
author={Todorov, Mihail Ivilinov and Paetzold, Johannes Christian and Schoppe, Oliver and Tetteh, Giles and Shit, Suprosanna and Efremov, Velizar and Todorov-V{\"o}lgyi, Katalin and D{\"u}ring, Marco and Dichgans, Martin and Piraud, Marie and others},
journal={Nature methods},
volume={17},
number={4},
pages={442--449},
year={2020},
publisher={Nature Publishing Group US New York}
}
@article{schmid,
title={Vascular density and distribution in neocortex},
author={Schmid, Franca and Barrett, Matthew JP and Jenny, Patrick and Weber, Bruno},
journal={Neuroimage},
volume={197},
pages={792--805},
year={2019},
publisher={Elsevier}
}
@article{kanari,
title={A topological representation of branching neuronal morphologies},
author={Kanari, Lida and D{\l}otko, Pawe{\l} and Scolamiero, Martina and Levi, Ran and Shillcock, Julian and Hess, Kathryn and Markram, Henry},
journal={Neuroinformatics},
volume={16},
number={1},
pages={3--13},
year={2018},
publisher={Springer}
}
@article{beers,
title={Barcodes distinguishing morphology of neuronal tauopathy},
author={Beers, David and Goniotaki, Despoina and Hanger, Diane P and Goriely, Alain and Harrington, Heather A},
journal={Physical Review Research},
volume={5},
number={4},
pages={043006},
year={2023},
publisher={APS}
}
