BBBC018 Cell Segmentation

A comprehensive cell segmentation project using pre-trained Cellpose models on the BBBC018 dataset. This repository demonstrates cell nuclei and actin segmentation from microscopy images with detailed performance metrics and comparative analysis.

Overview

This project provides a complete pipeline for segmenting cells in microscopy images using the Cellpose deep learning framework. It includes:

• Image preprocessing with normalization and CLAHE enhancement
• Multiple model comparison (cyto and nuclei models)
• Segmentation metrics including IoU, AP, AR, and F1 scores
• Detailed performance analysis across different image modalities (DNA and actin)

Dataset

• Dataset: BBBC018 - A publicly available benchmark dataset
• Content: Microscopy images with corresponding ground-truth segmentation masks
• Image Types:
  • DNA images (nuclei staining)
  • Actin images (cytoplasm/filament staining)

Features

Image Preprocessing

• Normalization: Applied to all images for consistent intensity ranges
• CLAHE (Contrast Limited Adaptive Histogram Equalization): Applied only to actin images to enhance contrast
• Note: CLAHE is intentionally excluded from DNA images as it enhances cytoplasm signals, which negatively impacts nuclei segmentation accuracy

Model Evaluation

• Implements custom segmentation metrics computation
• Outlines are converted to masks for accurate metric calculation
• Evaluated metrics @ IoU threshold 0.5:
  • Average Precision (AP)
  • Average Recall (AR)
  • F1 Score

Comparative Analysis

• Performance comparison across multiple Cellpose models
• Side-by-side visualization of results
• Metric comparison charts for DNA and actin segmentation

Requirements

Environment Setup

This project requires two separate virtual environments due to dependency conflicts:

Environment 1: Cellpose SAM (Latest Version)

For the latest Cellpose SAM models:

conda create -n cellpose-sam python=3.10
conda activate cellpose-sam
pip install cellpose torch torchvision

Environment 2: Cellpose Standard (Slightly Older Version)

For cyto and nuclei models:

conda create -n cellpose-standard python=3.10
conda activate cellpose-standard
pip install cellpose==3.1.1.2 torch torchvision

Dependencies

Common dependencies (install in both environments):

• NumPy
• Pandas
• Matplotlib
• OpenCV (cv2)
• PyTorch/CUDA (for GPU acceleration)

Usage

1. Prepare Data

Organize your microscopy images and corresponding ground-truth masks in the following structure:

data/
├── images/
│   ├── dna/
│   └── actin/
└── labels/
    ├── dna/
    └── actin/

2. Run the Notebook

# Activate the appropriate environment
conda activate cellpose-sam

# Launch Jupyter
jupyter notebook sample.ipynb

3. Execute Cells in Order

The notebook is organized into logical sections:

• Setup & Imports
• Image preprocessing
• Model inference
• Metric calculation
• Result visualization

Results

The notebook generates:

• Segmentation visualizations for both DNA and actin images
• Performance metrics for each model
• Comparative bar charts showing:
  • Average Precision (AP)
  • Average Recall (AR)
  • F1 Scores
• Per-image and per-model statistics

Model Comparison

The project evaluates multiple Cellpose models:

• Nuclei Model: Optimized for DNA/nuclei segmentation
• Cyto Model: Optimized for cytoplasm and actin filament segmentation
• SAM-based Models: Latest Segment Anything Model variants

Citation

If you use this work, please cite:

• Cellpose: Stringer et al., 2021
• BBBC018 Dataset: Broad Bioimage Benchmark Collection

References

• Cellpose GitHub Repository
• BBBC018 Dataset
• Broad Bioimage Benchmark Collection

Author

Marouane Barati

Contributing

Contributions are welcome! Please feel free to submit issues or pull requests.

Virtual Environment Conflicts

If models fail to load, ensure you're using the correct environment for the model type:

• SAM models → cellpose-sam environment
• Cyto/Nuclei models → cellpose-standard environment

Memory Issues

If you encounter out-of-memory errors, try:

• Reducing batch size
• Using smaller image patches
• Closing other GPU-consuming applications

Contact

For questions or issues, please open an issue on GitHub or contact marwanbrt12@gmail.com.