Overview

The ECG Analyzer is a web application that predicts heart diseases from uploaded electrocardiogram images using ResNet50, a deep learning model with 94.84% accuracy and 24.55% loss. It enhances cardiac diagnostics for healthcare professionals and offers a reliable tool for medical students to analyze ECG data.

Dataset

The dataset used in this project is the Ch. Pervaiz Elahi Institute of Cardiology Multan Dataset (Khan & Hussain, 2021). It contains a total of 928 images categorized into four classes:

• ECG Images of Myocardial Infarction Patients
• ECG Images of Patients with Abnormal Heartbeat
• ECG Images of Patients with a History of MI (Myocardial Infarction)
• Normal Person ECG Images

This diverse dataset ensures that the model can effectively distinguish between different types of heart conditions.

You can access the dataset here.

Features

• User Authentication:

  • Sign Up: Users can create a new account.
  • Login: Existing users can log in using their credentials.
  • Password Reset: Users can reset their password via email.

• Role-Based Access:

  • Healthcare Professionals:
    • Diagnosis: Ability to log in, access the upload page, and receive diagnoses for uploaded ECG images.
    • Access: Can only access the upload page for image analysis.
  • Medical Students:
    • Diagnosis: Similar functionality as healthcare professionals for uploading ECG images and receiving diagnoses.
    • Educational Resources: Authorized to access an additional page with explanations on how to read and interpret ECG images.
    • Access: Can view both the upload page and the educational explanation page.

• ECG Image Upload:

  • Upload Page: Allows users to upload ECG images for analysis.
  • Supported File Extensions: Only images with the following extensions are accepted: .png, .jpeg, .webp.

• Educational Page:

  • Explanation of ECG Images: Provides detailed explanations and guidance on how to read ECG images, including descriptions of the four dataset categories:
    • ECG Images of Myocardial Infarction Patients
    • ECG Images of Patients with Abnormal Heartbeat
    • ECG Images of Patients with a History of MI (Myocardial Infarction)
    • Normal Person ECG Images
  • Restricted Access: Only accessible to medical students.

• Error Handling:

  • Page Not Found: Proper error handling for non-existent pages to enhance user experience.

These features ensure that the application provides tailored functionality based on user roles, supports specific image formats, and includes comprehensive error handling.

Technologies

• Deep Learning Model:

  • Model: ResNet50
  • Training: The model is trained on a dataset divided into 70% training, 10% validation, and 20% testing.
  • Image Preprocessing: Conducted using Python and Jupyter Notebook, achieving an accuracy of 94.84% and a loss of 24.55% after extensive experimentation.

• Backend:

  • Framework: Node.js with Express.js
  • Database: MongoDB, utilizing Mongoose for managing user accounts and data.
  • Authentication: Implemented using bcryptjs for password hashing and jsonwebtoken for token-based authentication.
  • Email Verification: Managed with nodemailer for sending verification and confirmation emails.
  • File Upload and Validation: Handled by multer for secure image uploads.
  • Data Validation: Performed using joi to ensure the integrity of incoming data.
  • Verification Code Generation: Utilized nanoid for generating unique verification codes during password resets.
  • Child Process Management: Leveraged spawn from the child_process module to execute prediction functions upon image upload.
  • Environment Configuration: Managed with dotenv for handling environment variables.
  • Cross-Origin Resource Sharing (CORS): Implemented using the cors package to handle cross-origin requests.

• Frontend:

  • Library: React.js
  • Form Management and Validation: Utilized formik for form handling and yup for schema validation prior to backend submission.
  • SEO: Managed with react-helmet for controlling meta tags and improving search engine optimization.
  • Notifications: Implemented with react-hot-toast for displaying real-time errors and messages to users.
  • HTTP Requests: Handled with axios for making API requests.
  • Styling: Applied bootstrap for responsive design and user interface styling.
  • Result Display: Used sweetalert2 for presenting results and alerts to users.
  • Development Workflow: Managed with concurrently for running both frontend and backend simultaneously during development.

This technology stack ensures a robust, secure, and efficient web application, providing a seamless and professional experience for users.

Thesis

For a detailed explanation of the project, including its objectives, methodologies, implementations, and results, please refer to the comprehensive thesis document available here.