Understanding how cells reorganize during thrombus formation has been limited by the technical challenges of visualizing and segmenting densely packed structures in three dimensions. This has hindered exploration of the structural basis of excessive platelet activation and clot formation, which underlies major cardiovascular events such as myocardial infarction and stroke. To address this, we combined high-resolution focused ion beam scanning electron microscopy (FIB-SEM) with a tailored deep learning pipeline for automated segmentation. Our approach leverages multi-axis convolutional neural network analysis to integrate information across orthogonal planes, enabling accurate three-dimensional reconstructions of platelets and their organelles. This framework avoids the computational burden of full 3D neural networks while reducing reliance on manual annotation. Importantly, the pipeline is broadly applicable for segmentation of other tightly packed cell types beyond platelets. We validated performance using benchmark electron microscopy data sets and applied the pipeline to platelet-rich thrombi formed in a mouse carotid artery injury model. This enabled large-scale, quantitative analysis of thousands of platelets and organelles, including mitochondria, α-granules, and dense granules, across conditions of varying thrombus induction. Our findings reveal that dense granules undergo substantial depopulation during thrombus formation, while α-granules and mitochondria display a binary depletion pattern dependent on stimulus strength. We also observed spatial heterogeneity in organelle distribution and short-range clustering of degranulation events, challenging models that depict thrombus activation as uniformly coordinated. Together, this work provides both a methodological and biological advance. Methodologically, we establish a practical, generalizable framework for workstation-level segmentation of large volumetric data sets. Biologically, we uncover new principles of platelet activation and clot architecture, with implications for understanding thrombotic disease mechanisms. Ongoing work is extending this pipeline to other cell systems and injury models, underscoring its potential as a broadly useful tool in cell biology and pathology.
Baenen et al. (Sun,) studied this question.