Zheng‑Yu Lv1,2,3  · Zhen‑Hang Lu2,3  · Shuo Liu3,4  · Pengcheng Zhou3,4  · Pak‑Ming Lau2,3  · Guo‑Qiang Bi1,2,3  · Chang‑Lu Tao2,3

1 School of Artificial Intelligence and Data Science, University of Science and Technology of China, Hefei 230027, China 

2 Center for Integrative Imaging, Hefei National Research Center for Physical Sciences at the Microscale, MOE Key Laboratory for Membraneless Organelles and Cellular Dynamics, School of Life Sciences, Division of Life Sciences and Medicine, University of Science and Technology of China, Hefei 230027, China 

3 Interdisciplinary Center for Brain Information, the Brain Cognition and Brain Disease Institute, Shenzhen–Hong Kong Institute of Brain Science, Shenzhen Institutes of Advanced Technology, Chinese Academy of Sciences, Shenzhen 518055, China 

4 Faculty of Life Sciences, Shenzhen University of Advanced Technology, Shenzhen, China

Abstract

Synaptic vesicles (SVs) are essential components of neurotransmission, and their structure and organization are critical determinants of synaptic function. Cryo-electron tomography (cryo-ET) enables the in situ visualization of SVs in a near-native state. However, their accurate segmentation remains challenging owing to the low signal-to-noise ratio, missing-wedge artifacts in synaptic tomograms, and the inherent morphological heterogeneity of SVs. Herein, we present Vesicle3D, an integrated platform that combines: (1) a vesicle-specific denoising and missing-wedge restoration model; (2) a dual-pathway 3D Res-UNet with prediction fusion; (3) ellipsoid-aware post-processing to preserve authentic vesicle morphology; and (4) an interactive, Napari-based graphical user interface for correction and fine-tuning. This method outperformed existing methods, particularly in the detection of ellipsoidal SVs, and demonstrated robust generalization across diverse datasets, including chemically fixed neurons, isolated synaptosomes, and cryo-FIB lamellae. Furthermore, fine-tuning with approximately 200 annotations enabled rapid adaptation to new datasets. Vesicle3D provides a scalable and versatile framework for the large-scale quantitative analysis of vesicles in cryo-tomograms.

Keywords

Synaptic vesicle; Cryo-electron tomography; Deep learning; 3D segmentation

[SpringerLink]