KITE: Web-Based Semi-Automated Retinal Fluid Segmentation Tool

An interactive web application for segmenting retinal fluid regions in OCT images

Screenshots • Installation • Usage

🩺 Overview

KITE is a web-based tool designed to assist medical professionals in segmenting retinal fluid regions in Optical Coherence Tomography (OCT) images. The tool combines traditional image processing techniques with state-of-the-art deep learning models to provide accurate, efficient, and user-friendly medical image annotation.

Key Features

• 🖼️ Interactive Annotation Tools: Point, line, box, and fill tools for precise manual annotation
• 🤖 Multiple Segmentation Algorithms:
  • KITE Algorithm: Custom region-growing approach using OpenCV
  • U-Net: Deep learning model trained on DUKE OCT dataset
  • MedSAM: Foundation model for medical image segmentation
• 🌐 Web-Based Interface: Django based
• 📊 Real-Time Processing: Fast segmentation with immediate visual feedback
• 💾 Export Capabilities: Download results in PNG and NPY formats
• 🎨 Layer Management: Organize and edit multiple segmentation layers

🎯 Applications

• Clinical Research: Generate annotated datasets for training ML models
• Medical Education: Teaching tool for understanding retinal pathology
• Ophthalmology Practice: Assist in diagnosis and treatment planning

🏗️ System Architecture

┌─────────────────┐    ┌──────────────────┐    ┌─────────────────┐
│   Frontend      │    │     Backend      │    │   AI Models     │
│                 │    │                  │    │                 │
│ • HTML/CSS/JS   │◄──►│ • Django REST    │◄──►│ • U-Net         │
│ • Konva.js      │    │ • Image Pipeline │    │ • MedSAM        │
│ • Annotation    │    │ • File Handling  │    │ • Traditional   │
│   Tools         │    │                  │    │   CV Methods    │
└─────────────────┘    └──────────────────┘    └─────────────────┘

🚀 Installation

Prerequisites

• Python 3.8+
• Node.js (for frontend dependencies)
• CUDA-compatible GPU (optional, for faster deep learning inference)

1. Clone the Repository

git clone https://github.com/your-username/KITE.git
cd KITE

2. Set Up Python Environment

Using conda (recommended):

conda create --name kite-env python=3.10
conda activate kite-env

Using venv:

python -m venv venv
source venv/bin/activate  # On Windows: venv\Scripts\activate

3. Install Dependencies

pip install -r requirements.txt

4. Download Model Weights

To use the MedSAM segmentation model, you need to manually download the model weights.

Instructions are provided in: models/README.md

5. Database Setup

cd website
python manage.py migrate
python manage.py collectstatic

6. Run the Application

python manage.py runserver

Visit http://localhost:8000 to access the application.

📖 Usage

Quick Start

1. Select Algorithm: Choose between KITE, U-Net, or MedSAM
2. Upload Image: Upload your OCT image (JPEG format) or choose provided the demo image
3. Annotate (if using KITE / MedSAM): Use annotation tools to mark regions of interest
4. Segment: Click "Ready to Segment!" to process the image
5. Edit: Refine results using manual editing tools
6. Export: Download segmentation masks in your preferred format

Note: It is possible to upload previous annotations or mask results and edit them in KITE mode.

Detailed Workflows

KITE Algorithm Workflow

Upload Image → Annotate ROI → Traditional Segmentation → Manual Refinement → Export

U-Net Workflow

Upload Image → Automatic Segmentation → Manual Refinement → Export

MedSAM Workflow

Upload Image → Draw Bounding Boxes → AI Segmentation → Manual Refinement → Export

🛠️ Technology Stack

Frontend

• HTML5/CSS3: Modern web standards
• JavaScript: Interactive functionality
• Konva.js: Canvas-based annotation tools

Backend

• Django: Web framework and REST API
• Python: Core programming language
• OpenCV: Traditional image processing
• NumPy/SciPy: Scientific computing

Deep Learning

• PyTorch: Deep learning framework
• U-Net: Custom trained model on DUKE dataset
• MedSAM: Foundation model integration

📷 Screenshots

(from KITE mode)

• Annotation

• Segmentation

• Mask Layers

📄 License

This project is licensed under the MIT License.

👥 Team

Koç University COMP 491 Capstone Project Team

• Duru Tandoğan
• Mislina Akça
• Sinemis Toktaş
• Yamaç Ömür

Project Advisor: Çiğdem Gündüz Demir – Koç University

🙏 Acknowledgments

• DUKE University OCT Dataset
• U-Net and MedSAM research communities
• OpenCV and PyTorch communities

📈 Future Work

• Integration with additional clinical data (text, metadata)
• Support for other medical imaging modalities (MRI, CT)
• Real-time collaborative annotation
• Advanced AI model fine-tuning capabilities
• Clinical validation studies

📚 References

1. Ronneberger, O., Fischer, P., & Brox, T. (2015). U-Net: Convolutional networks for biomedical image segmentation.
2. Ma, J., He, Y., Li, F., et al. (2024). Segment anything in medical images. Nature Communications.

---

⬆ Back to Top