National Conference of the Italian Group for the Study of Neuromorphology
Vol. 70 No. s2 (2026): 36th National Conference of the Italian Group for the Study of...

P021 | ENCEPHALON-WIDE DEEP LEARNING FOR GRAPH EXTRACTION: A NEW CONNECTOMIC RECONSTRUCTION PIPELINE AND ITS APPLICATION TO THE WHOLE-BRAIN CATECHOLAMINERGIC SYSTEM

F.E. Bellomi1, N.S. Ullman1, F. Petrocchi1, G. Loffredo2, A. Riva1, G. Di Giacomo1, G. Michelotti1, A. Nobili2, L. La Barbera2, F. Mittoni3, C. Caturano1, G. Iannello4, M. D'Amelio2, G. Vivacqua1 | 1Department of Medicine and Surgery, Laboratory of Microscopic and Ultrastructural Anatomy, Campus Bio-Medico University of Roma, Italy; 2Department of Medicine and Surgery, Laboratory of Molecular Neuroscience, Campus Bio-Medico University of Roma, Italy; 3Department of Civil, Computer Science and Aeronautical Technologies Engineering, Università degli Studi Roma Tre, Italy; 4Department of Engineering, Research Unit of Artificial Intelligence and Computing Systems, Campus Bio-Medico University of Roma, Italy

Publisher's note
All claims expressed in this article are solely those of the authors and do not necessarily represent those of their affiliated organizations, or those of the publisher, the editors and the reviewers. Any product that may be evaluated in this article or claim that may be made by its manufacturer is not guaranteed or endorsed by the publisher.
Published: 20 November 2026
0
Views
0
Downloads

Authors

Introduction. The function of the nervous system is tightly linked to the anatomy of its circuits, whose study is the aim of connectomics. In recent years, synaptic-resolution connectomics has produced extremely detailed wiring diagrams, but its cost in time and resources limits it to small volumes or few individuals. Here we present EDGE, a connectomic reconstruction pipeline at cellular scale that trades synaptic detail for a drastic reduction in time and cost.
Methods. Whole mouse brains were immunolabelled for tyrosine hydroxylase (TH), cleared with iDISCO and imaged on a mesoSPIM light-sheet microscope (2.03 × 2.03 × 3 µm voxels). Tiles were stitched with BigStitcher, and an nnU-Net semantic segmentation separated somata, nuclei and filaments from background. EDGE converts the segmentation into a graph: somata are split by nucleus-seeded watershed, filaments are skeletonized, nodes are placed along the skeleton every 3 voxels and linked by geodesic edges constrained inside the labelled tissue. The brain is processed in overlapping blocks and reassembled without seams. Tracing was validated against manual reconstructions: the 166 neurons of the BigNeuron Gold166 benchmark and 34 neurons reconstructed by hand in our data. Graphs were registered to the Allen Mouse Brain Atlas (ABA), enabling region-wise analyses of cell number, fibre density, neuronal morphology and network connectivity, which were compared with values from a systematic review of the literature. The processed data are browsable at multiple resolution levels in Neuroglancer.
Results. We reconstructed the whole TH-positive network of three wild-type mouse brains, covering the dopaminergic and noradrenergic systems. Each brain yields a graph of around 100 million nodes describing about 30,000 TH-positive neurons with their fibres; TH-positive fibres occupy about 16% of the brain volume. Cell numbers, the morphology of each catecholaminergic nucleus, the organisation of the network and the connectivity between nuclei were consistent with literature data.
Conclusions. EDGE combines speed and reliability, making it suitable for the rapid screening of pathological models: it can identify where circuit alterations occur and localise the regions where synaptic-resolution electron microscopy should be targeted.

Downloads

Download data is not yet available.

Citations

How to Cite



1.
P021 | ENCEPHALON-WIDE DEEP LEARNING FOR GRAPH EXTRACTION: A NEW CONNECTOMIC RECONSTRUCTION PIPELINE AND ITS APPLICATION TO THE WHOLE-BRAIN CATECHOLAMINERGIC SYSTEM: F.E. Bellomi1, N.S. Ullman1, F. Petrocchi1, G. Loffredo2, A. Riva1, G. Di Giacomo1, G. Michelotti1, A. Nobili2, L. La Barbera2, F. Mittoni3, C. Caturano1, G. Iannello4, M. D’Amelio2, G. Vivacqua1 | 1Department of Medicine and Surgery, Laboratory of Microscopic and Ultrastructural Anatomy, Campus Bio-Medico University of Roma, Italy; 2Department of Medicine and Surgery, Laboratory of Molecular Neuroscience, Campus Bio-Medico University of Roma, Italy; 3Department of Civil, Computer Science and Aeronautical Technologies Engineering, Università degli Studi Roma Tre, Italy; 4Department of Engineering, Research Unit of Artificial Intelligence and Computing Systems, Campus Bio-Medico University of Roma, Italy. Eur J Histochem [Internet]. 2026 Nov. 20 [cited 2026 Nov. 20];70(s2). Available from: https://www.ejh.it/ejh/article/view/4841