MediTriage

From chaos to clarity: an AI-powered triage system that ranks patients by urgency, helping hospitals save lives faster when resources are limited.

Comment

Inspiration

This project is an AI-powered triage system designed to improve the classification of patients in the Emergency Room (ER). Since hospital resources are often limited, the system helps prioritize critical cases and ensure that patients who need urgent care receive it first. By doing so, it contributes to saving as many lives as possible while optimizing hospital workflows.

What it does

MediTriage is an AI-powered triage system that helps classify emergency patients based on their severity and priority level.
By entering patient details such as age, gender, symptoms, vital signs, and medical history, the system predicts:

  • Severity level %
  • Triage category (critical, urgent, stable)

This allows emergency room staff to prioritize critical cases quickly and allocate hospital resources more effectively.
The goal is to reduce waiting time for high-risk patients and ultimately save more lives in overcrowded ER environments.

How I built it

I built MediTriage as a web-based system with a focus on both accuracy and usability.

  1. Interface Design
  • Developed the user interface using HTML, CSS, and Tailwind CSS for a clean, responsive design.
  • Built interactive forms and dashboard pages for seamless patient data entry and triage visualization.
  1. Backend & Database
  • Used Flask to handle requests, manage routes, and serve the application.
  • Integrated SQLAlchemy as the ORM to manage patient data and triage results efficiently.
  1. Machine Learning Model
  • Trained a Random Forest classifier to predict both severity level and triage category.
  • Preprocessed patient data with feature encoding and normalization.
  • Saved and reloaded the trained model for consistent predictions.
  1. System Integration
  • Connected the frontend with the backend to allow real-time predictions upon patient registration.
  • Stored all patient information, including predictions, in the database.
  • Displayed results on a dashboard, automatically sorting patients from most severe to least severe.
  1. Testing & Iteration
  • Ran multiple test cases with diverse patient data.
  • Adjusted the model and refined the dashboard interface for better usability.

Challenges I ran into

  • I couldn’t find an accurate real-world dataset, so I had to generate synthetic patient data myself. The predictions would be more reliable with actual hospital data.
  • Managing data encoding to ensure the same transformations are applied during training and inference.
  • Balancing simplicity and usability in the interface while handling many patient features and predictions.

Accomplishments that I'm proud of

  • Successfully building a full-stack AI-powered triage system by myself.
  • Integrating a machine learning model with a functional Flask + SQLAlchemy web application.
  • Designing a dashboard that automatically prioritizes patients based on severity.

What I learned

  • How to connect machine learning workflows with real-world web applications.
  • The importance of consistent preprocessing and encoding in ML pipelines.
  • How to build a scalable and user-friendly system for critical healthcare use cases.

What's next for MediTriage

  • Add real hospital datasets for higher accuracy.
  • Implement role-based access for doctors, nurses, and admins.
  • Add interactive graphs and statistics to track patient flow and outcomes.
  • Integrate with IoT medical devices for live vital sign monitoring.
  • Link MediTriage with hospital ER management systems to streamline operations.
Share this project:

Updates