CebraEM

CebraEM implements a workflow for the segmentation of cellular organelles in volume SEM datasets using a transferable CNN-based membrane prediction combined with machine learning-based post-processing.

Key Features:

• Transferable CNN-based membrane prediction: Uses a generic convolutional neural network for membrane prediction that does not require dataset-specific training.
• Random forest classification: Classifies over-segments to enable semantic instance segmentation.
• Graph multi-cut grouping: Groups over-segments into final organelle instances via graph-based optimization.
• Modular architecture: Includes modules for core functionality, annotation, and network analysis.
• Cross-platform support: Supports execution in Linux and Windows environments via conda.

Scientific Applications:

• Organelle segmentation in volume SEM: Produces state-of-the-art semantic instance segmentations of cellular organelles across diverse volume SEM datasets without additional post-processing.
• Large-scale data processing: Enables efficient segmentation of hundreds of gigabytes of volume EM data by single users.
• Multi-modal volume SEM analysis: Applicable to datasets from various sample preparations and imaging modalities in volume SEM.

Methodology:

CebraEM relies on a generic CNN-based membrane prediction, followed by over-segmentation, random forest classification, and graph multi-cut grouping.