HematoVision: Advanced Blood Cell Classification Using Transfer Learning

Team Details

• Team ID: LTVIP2025TMID41359
• Team Size: 4
• Team Leader: Golla Jahnavi
• Team Members:
  • Gujjala Pranay Kumar
  • Irakam Siva Venkata Bhanu Prakash
  • Johan Abhishek

---

Dataset Link: https://www.kaggle.com/datasets/paultimothymooney/blood-cells/data

Project Overview

HematoVision is an innovative project aimed at developing an accurate and efficient model for classifying blood cells. It leverages transfer learning techniques with pre-trained Convolutional Neural Networks (CNNs) to expedite the training process and significantly enhance classification accuracy. The project provides a reliable and scalable tool for pathologists and healthcare professionals, improving the precision and efficiency of blood cell analysis.

Features

• Accurate Blood Cell Classification: Classifies four distinct types of blood cells: Eosinophil, Lymphocyte, Monocyte, and Neutrophil.
• Transfer Learning: Utilizes a pre-trained MobileNetV2 model to achieve high accuracy with reduced training time and computational resources.
• Web Application Interface: A user-friendly Flask-based web application for easy image upload and real-time prediction display.
• Production-Ready: Designed for deployment, with a clear project structure and optimized code.

Project Structure

HematoVision_App/
├── app.py                 # Main Flask application script
├── requirements.txt       # Python dependencies
├── blood_cell.h5          # Trained MobileNetV2 model file
├── templates/             # HTML templates for the web interface
│   ├── home.html
│   └── result.html
└── static/
    └── uploads/           # Directory for temporarily storing uploaded images

Installation and Setup (Local)

Prerequisites

• Python 3.8+
• pip
• Git

Setup Instructions

1. Clone the repository:

   git clone https://github.com/gujjala-pranay/hematovision-app.git
   cd hematovision-app

2. Create and activate a virtual environment:

   python -m venv venv

   • Windows:

     .\venv\Scripts\activate

   • macOS/Linux:

     source venv/bin/activate

3. Install dependencies:

   pip install -r requirements.txt

4. Run the application:

   python app.py

5. Open in browser: Visit http://127.0.0.1:5000/

Usage

• Step 1: Upload a blood cell image (PNG, JPG, JPEG, or GIF)
• Step 2: Click the "Classify Blood Cell" button
• Step 3: View the classification result with the image

Model Details

• Architecture: MobileNetV2 with custom classification layers
• Dataset: 12,500 augmented blood cell images from Kaggle
• Training: 5 epochs, Adam optimizer, categorical cross-entropy
• Accuracy: ~85.3% validation accuracy
• Model File: blood_cell.h5

Deployment

Ready for deployment on platforms like Render, Railway, or Heroku. Refer to the HematoVision_Deployment_Guide.md for full steps.

Contributing

Fork the repository, create pull requests, or submit issues to contribute.