AI Hackathon: Wound Image Classification with MongoDB Vector Search

Developed for the VOXEL51 & MongoDB AI Hackathon

This project demonstrates an end-to-end AI application for medical image classification, leveraging MongoDB's vector search to identify wound types from images. The stack is containerized using Docker Compose and integrates Flask, Jupyter, and MongoDB for scalable, reproducible development and deployment.

Project Report (PDF)

You can view the detailed project report or presentation by clicking the link below:

📄 View Project Report (PDF)

Project Overview

This application was created as part of the VOXEL51 & MongoDB AI Hackathon. It enables users to upload wound images and receive automated wound type predictions using a trained deep learning model. The system stores image embeddings in MongoDB and utilizes vector search to efficiently retrieve the most similar cases, supporting explainable AI in medical diagnostics.

Key Features:

• Wound Image Classification: Predicts wound types from uploaded images using a trained neural network.
• MongoDB Vector Search: Stores image embeddings and retrieves similar cases for explainable results.
• Interactive Web UI: Built with Flask for easy image upload, prediction, and result visualization.
• Jupyter Integration: For model training, experimentation, and data exploration.
• Containerized Stack: All services run in isolated Docker containers for easy setup and reproducibility.

Technologies Used

• MongoDB: Stores image data and embeddings, provides vector search capabilities.
• Flask: Serves the web application and API endpoints.
• Jupyter: Supports interactive development and model training.
• Docker Compose: Orchestrates multi-container deployment.

My Contributions

• Designed and implemented the end-to-end pipeline for wound image classification.
• Integrated MongoDB vector search for efficient and explainable image retrieval.
• Developed the Flask web interface for user interaction.
• Automated the environment setup using Docker Compose.

Building & Running

# Clone the repository
git clone https://github.com/hyper07/AI_Hackathon.git

# Move to the project directory
cd AI_Hackathon/

# Build and run the containers
docker-compose up -d

# Stop and remove the containers
docker-compose down

Accessing Services

MongoDB

• Connection String:

  from pymongo import MongoClient
  client = MongoClient('mongodb://user:pass@hackathon-mongo:27017/')

Jupyter

• Web Interface: http://localhost:8899

Flask App

• Web Interface: http://localhost:5010

Dataset

Wound image dataset from Kaggle:
https://www.kaggle.com/datasets/ibrahimfateen/wound-classification/data

curl -L -o ./wound-classification.zip\
  https://www.kaggle.com/api/v1/datasets/download/ibrahimfateen/wound-classification

References

• MongoDB Vector Search Documentation
• Flask Documentation
• Docker Compose Documentation

---

For more detailed information on each service, please refer to the respective documentation.