neuro-sam 0.1.8

Neuro-SAM: Foundation Models for Dendrite and Dendritic Spine Segmentation

Neuro-SAM

Foundation Models from Dendrite and Dendritic Spine Segmentation

This project demonstrates an interactive UI to segment dendrites and dendritic spines. The model of choice is SAMv2 and the framework used is pytorch.

📝 Table of Contents

• Neuro-SAM
  • 📑 Table of Contents
  • 🧠 Overview
  • 📦 Built With
  • 📂 Repository Structure
  • 🚀 Installation
  • 📊 Usage
  • 🛠 Workflow
  • 🧑‍💻 Model Training
  • 📁 Data Format
  • 📄 License
  • 📬 Contact

🧠 Overview

Neuro-SAM provides an end-to-end pipeline for analyzing neural structures from 3D microscopy data, featuring:

• Path Tracing: Waypoint-based A* pathfinding
• Dendrite Segmentation: SAM2-based dendrite segmentation
• Smart Spine Detection: Multi-view analysis for spine detection
• Spine Segmentation: Individual spine segmentation using trained SAM2 model

📦 Built With

PyTorch - an open-source machine learning library for Python, widely used for deep learning applications.

Segment Anything Model - a foundation model used for segmentation built by Meta AI.

Weights and Biases - a tool for tracking and visualizing machine learning experiments.

Visual Studio Code - a code editor redefined and optimized for building applications.

FAU High Performance Computing - a high-performance computing cluster at Friedrich-Alexander-Universität Erlangen-Nürnberg.

📁 Repository Structure

Neuro-SAM/
├── Train-SAMv2/                    # SAM2 training infrastructure
│   ├── sam2/                       # SAM2 model implementation
│   ├── checkpoints/                # Model checkpoints
│   ├── results/                    # Trained model outputs
│   ├── utils/                      # Training utilities
│   ├── train_dendrites.py         # Dendrite model training
│   └── train_spines.py            # Spine model training
├── brightest_path_lib/             # Advanced pathfinding algorithms
│   ├── algorithm/                  # A* and waypoint search implementations
│   ├── cost/                       # Cost function definitions
│   ├── heuristic/                  # Heuristic functions
│   ├── visualization/              # Path visualization tools
│   └── ...
├── napari_utils/                   # Napari plugin components
│   ├── main_widget.py             # Main interface with anisotropic scaling
│   ├── path_tracing_module.py     # Interactive path tracing
│   ├── segmentation_module.py     # Dendrite segmentation interface
│   ├── spine_detection_module.py  # Spine detection with smart tracking
│   ├── spine_segmentation_module.py # Individual spine segmentation
│   └── visualization_module.py    # Path management and visualization
└── neuro_sam_plugin.py            # Main plugin entry point

🚀 Installation

Prerequisites

• Python 3.10+
• CUDA-compatible GPU (recommended)
• Conda/Miniconda

Environment Setup

1. Clone the repository:

git clone https://github.com/nipunarora8/Neuro-SAM.git
cd Neuro-SAM

2. Create local environment:

conda create -p ./.venv python=3.10 -c conda-forge
conda activate ./.venv

3. Install dependencies:

pip install uv
uv sync

4. Download SAM2 checkpoints:

cd Train-SAMv2/checkpoints
bash download_ckpts.sh

📊 Usage

Quick Start

from neuro_sam_plugin import run_neuro_sam

# Launch with default spacing (94nm x 94nm x 500nm)
viewer = run_neuro_sam(image_path="your_image.tif")

# Launch with custom voxel spacing
viewer = run_neuro_sam(
    image_path="your_image.tif",
    spacing_xyz=(100.0, 100.0, 300.0)  # X, Y, Z spacing in nm
)

Command Line Interface

# Basic usage
python neuro_sam_plugin.py --image_path /path/to/your/image.tif

# Custom spacing
python neuro_sam_plugin.py --image_path image.tif \
    --x-spacing 100.0 --y-spacing 100.0 --z-spacing 300.0

# Load benchmark dataset
python neuro_sam_plugin.py

🔬 Workflow

1. Configure Voxel Spacing

Set accurate X, Y, Z voxel spacing in the "Path Tracing" tab for proper anisotropic scaling:

• Typical two-photon: 94nm × 94nm × 500nm
• Confocal: varies by objective and zoom

2. Trace Dendritic Paths

• Click waypoints along dendrite structures
• Algorithm automatically finds optimal brightess paths

3. Segment Dendrites

• Load pre-trained SAMv2 dendrite model
• Segment individual path with SAMv2

4. Detect Spines

• Smart multi-view detection using tube data generation
• Angle-based matching between 2D and tubular views

5. Segment Spines

• Fine-grained spine segmentation using specialized SAMv2 model
• Dendrite mask overlay to suppress background signal
• Manual point extension across frames
• Contrasting color system for visualization

🔧 Model Training

Dendrite Model

cd Train-SAMv2
python train_dendrites.py --ppn 20 --pnn 10 --batch_size 32 --model_name "small"

Spine Model

python train_spines.py --model_name "small" --batch_size 16

📁 Data Format

Input Requirements

• Image Format: TIFF, .d3set (to reproduce training results)
• Dimensions: 3D volumes (Z×Y×X)
• Bit Depth: 8-bit or 16-bit grayscale
• Size: Tested up to 2048×2048×500 voxels

Output Formats

• Paths: NumPy arrays with coordinates
• Masks: Binary TIFF volumes

📄 License

This project is licensed under the MIT License - see the LICENSE file for details.

Useful VSCode Extensions

• Remote Explorer - Open projects on remote servers.
• Log Viewer - A log monitoring extension.
• Black Formatter - Python auto code formatter.
• Markdown All in One - Markdown preview and editing.

📬 Contact

Nipun Arora - [e-mail]

---

Made with ♥️ at Anki Lab 🧠✨