U-Net Ensembles for Segmentation of High-Density Cell Populations and Organelles

Ensemble U-Net models achieve state-of-the-art segmentation of high-density platelets and organelles in 3D electron microscopy datasets, enabling detailed morphological analysis.

• Why it matters: Accurate segmentation of complex, intertwined cell structures like activated platelets is crucial for understanding thrombus formation and heterogeneity, yet remains challenging with existing methods.
• What they did: A machine vision pipeline utilizing 2D neural networks with ensemble U-Net architectures was developed to segment and analyze hundreds of platelets and their organelles in volume electron microscopy data, validated against CREMI challenge and focused FIB-SEM datasets.
• The result: The approach delivers high-precision segmentation and morphological measurements, facilitating large-scale single-cell 3D studies and providing insights into thrombus structure, with accessible tools for replication and further analysis.