Skip to content

cell-segmenter

A PyQt-based application for segmenting and analyzing microscopy image sequences, specifically designed for DIC (Differential Interference Contrast) images from TIRF microscopy. The tool provides an interactive interface for training machine learning models to segment cellular structures.

View on GitHub →

Overview

Languages: Python

Documentation

Cell Segmentation Tool

A PyQt-based application for segmenting and analyzing microscopy image sequences, specifically designed for DIC (Differential Interference Contrast) images from TIRF microscopy. The tool provides an interactive interface for training machine learning models to segment cellular structures.

Features

  • Load and view multi-frame TIFF stacks
  • Interactive painting tools for creating training data
  • Multiple segmentation approaches:
  • Random Forest classifier
  • Convolutional Neural Network (CNN)
  • Real-time visualization of segmentation results
  • Batch processing capabilities
  • Training data management:
  • Multiple label classes
  • Undo/redo functionality
  • Label class editing and removal
  • Option to use training data from single or multiple frames

Installation

Prerequisites

  • Python 3.8 or higher
  • pip package manager

Dependencies

pip install -r requirements.txt

Main dependencies include: - PyQt6 - pyqtgraph - numpy - scikit-image - scipy - torch - tifffile

Installation Steps

  1. Clone the repository: 

    git clone https://github.com/yourusername/cell-segmentation-tool.git
cd cell-segmentation-tool

  2. Install in development mode: 

    pip install -e .

Usage

Starting the Application

python main.py

Basic Workflow

  1. Load Data
  2. Click "Load TIFF Stack" to open your image sequence

  3. The application accepts multi-frame TIFF files

  4. Create Labels

  5. Click "Add Label" to create a new label class
  6. Select a color for the label

  7. Create multiple labels for different cellular structures

  8. Add Training Data

  9. Select a label from the list
  10. Switch to "Paint" mode
  11. Paint over regions in the image to mark training data
  12. Use mouse wheel or brush size control to adjust brush size

  13. Navigate through frames using the frame slider

  14. Train Model

  15. Choose between Random Forest and CNN
  16. Select whether to use training

[View full README on GitHub]