Cellpose

Cellpose segments cell bodies, membranes, and nuclei in microscopy images using a generalist deep learning approach to enable accurate instance segmentation across diverse microscopy modalities and cell types.

Key Features:

• Deep learning-based segmentation: Instance segmentation implemented via a deep learning algorithm for cellular structures.
• Generalist model: Performs segmentation across diverse image types without per-image model retraining or parameter adjustments.
• Training dataset: Trained on a dataset of over 70,000 segmented cellular objects encompassing a wide variety of images.
• 3D adaptation: Adapts a two-dimensional model for three-dimensional imaging without requiring additional 3D-labeled data.
• Manual labeling and curation: Includes capabilities for manual labeling and curation of automated segmentation results to generate validated labels.
• Community-driven retraining: Periodically retrained on aggregated community-contributed labeled datasets to improve performance over time.

Scientific Applications:

• Cellular structure segmentation: Accurate segmentation of cell bodies, membranes, and nuclei for quantitative analysis across different microscopy modalities and cell types.
• Three-dimensional image analysis: Application of adapted 2D models to 3D image stacks to produce 3D segmentations without 3D-labeled training data.
• Dataset generation and curation: Creation and curation of labeled datasets to support community-driven improvements and retraining of segmentation models.

Methodology:

Uses a deep learning-based instance segmentation algorithm trained on over 70,000 segmented cellular objects, adapts a 2D model for 3D contexts without 3D-labeled data, and incorporates manual labeling/curation with periodic retraining on aggregated community-contributed labels.