LSFM Segmentations

This repository provides tools for mouse brain segmentation and visualization using U-Net architecture. It includes functionalities for dataset preparation, model training, evaluation, and visualization of segmentation results.

Features

• Dataset Preparation: Generate training and testing datasets from 3D brain data.
• Model Training: Train U-Net models for segmentation tasks.
• Evaluation: Evaluate segmentation results using metrics like IoU, F1 score, and confusion matrix.
• Visualization: Visualize brain slices, annotations, and segmentation overlays.
• Export: Save annotated images and segmentation results as PNG files.

Project Structure

├── config.py                # Configuration loader for YAML settings
├── config.yaml              # Configuration file with dataset paths and hyperparameters
├── dataset.py               # Dataset creation and preprocessing utilities
├── dataset_validation.ipynb # Notebook for dataset exploration and validation
├── evaluate_and_export.ipynb # Notebook for model evaluation and export
├── mus_brain_data_mining.ipynb # Notebook for visualization and data mining
├── README.md                # Project documentation
├── requirements.txt         # Python dependencies
├── train_and_save.py        # Script for training and saving U-Net models
├── u_net.py                 # U-Net model implementation

Installation

1. Clone the repository:

   git clone https://github.com/your-repo/LSFM_Segmentations.git
   cd LSFM_Segmentations

2. Set up a virtual environment:

   python3 -m venv .venv
   source .venv/bin/activate

3. Install dependencies:

   pip install -r requirements.txt

4. Install Jupyter kernel:

   python3 -m ipykernel install --user --name=mus_brain

Usage

Dataset Preparation

Use the generate_complete_dataset function in dataset.py to create training and testing datasets from NIfTI files.

Model Training

Run train_and_save.py to train a U-Net model:

python train_and_save.py --model_name unet_model --output_path ./trained_models

Evaluation

Use evaluate_and_export.ipynb to evaluate the trained model and compute metrics like IoU and F1 score.

Visualization

Use mus_brain_data_mining.ipynb to visualize brain slices, annotations, and segmentation overlays.

Dataset Validation

Use dataset_validation.ipynb to explore and validate the dataset.

Configuration

Modify config.yaml to set dataset paths and hyperparameters:

dataset:
  train:
    image: "/path/to/train/images"
    mask: "/path/to/train/masks"
  test:
    image: "/path/to/test/images"
    mask: "/path/to/test/masks"
seed: 42
output_size: 128
batch_size: 8
epochs: 10
lr: 0.001

Dependencies

See requirements.txt for a list of required Python packages.

License

This project is licensed under the MIT License.

Acknowledgments

• U-Net architecture reference: https://arxiv.org/pdf/1505.04597
• Libraries used: PyTorch, NumPy, Matplotlib, scikit-image, scikit-learn

Feel free to contribute or raise issues for improvements!