Implemnetation of DEGAST3D: Learning Deformable 3D Graph Similarity to Track Plant Cells in Unregistered Time Lapse Images (https://ieeexplore.ieee.org/document/10820545)

3D Plant Cell Segmentation, Tracking, and Cell Division Detection

Overview

This project performs 3D segmentation, registration, tracking, and cell division detection for plant cell images using deep learning-based approaches. It leverages Cellpose for segmentation and implements a 3D Graph Matching approach for tracking plant cells and detecting cell division events.

Features

✅ Automated 3D segmentation using Cellpose ✅ Automated 3D registration
✅ Supports multiple plant datasets for microscopic image analysis
✅ Performs plant cell tracking using 3D graph-based matching
✅ Detects cell division events

---

Installation

1. Install Dependencies

Create the required Conda environment using:

conda env create -f environment.yml

---

Project Structure

project_root/
├── segmentation.py         # 3D segmentation script using Cellpose
├── Registration.py         # Registration script for aligning time-lapse images
├── Inference_Tracking_Celldiv.py  # Combined script for registration, tracking & division detection
├── data_process.py         # Functions for data preprocessing
├── data/                   # Dataset directory (organized by plant)
│   ├── plant_1/
│   │   ├── microscopic_images/     # Raw images
│   │   ├── cellpose/               # Segmentation results
│   │   │   ├── seg/                # Segmented masks
│   │   |	├── stack_info/             # Processed stack information
│   │   |	├── Registration_info/      # Registration details
│   ├── plant_2/
│   ├── plant_4/
│   ├── plant_13/
│   ├── plant_15/
│   ├── plant_18/
│   ├── test_plant/
├── results/                # Output directory for segmentation, tracking, and cell division data
└── README.md               # Project documentation

---

Usage

1. Define Plant Names

Ensure the following plant names are set in segmentation.py and Inference_Tracking_Celldiv.py:

plant_names = ['plant_1', 'plant_2', 'plant_4', 'plant_13', 'plant_15', 'plant_18', 'test_plant']

2. Prepare the Dataset

Place your microscopic image dataset inside the data directory, following the structure:

data/
├── plant_1/
│   ├── microscopic_images/
│   ├── cellpose/
│   │   ├── seg/
│   |	├── stack_info/
│   |	├── Registration_info/
├── plant_2/
├── ...

3. Run Segmentation

Execute the segmentation script by specifying the dataset directory and plant index: This script also includes data processing using STACK_INF0_3D and STACK_INF0_Registration.

python segmentation.py --main_dir "data" --plant_idx 6 --segmentation "cellpose"

📌 Output: Segmented masks will be saved in the seg/ directory inside the respective plant folder. The processed data will also be saved in stack_info and Registration_info sub-folder.

4. Run Registration, Tracking and Cell Division Detection for every pair of timepoint

Perform Registration , Tracking and Cell division detection in one step:

python Inference_Tracking_Celldiv.py --main_dir "data" --plant_idx 6 --segmentation "cellpose"

📌 Output: Tracking and division results will be stored inside results/{segmentation}_{plant_name}/.

---

Customization

• Change --plant_idx to select a different plant dataset.
• The framework supports any segmentation algorithm.

---

Acknowledgments

This project utilizes:

• Cellpose for deep learning-based cell segmentation.
• 3D Graph was constructed based on this reposititory Deep Graphical Feature Learning for the Feature Matching Problem, ICCV2019

---

License

📜 This project is licensed under the MIT License.