Healthcare Triage Assistant

End-to-end ML deployment project predicting patient triage levels using FastAPI, Streamlit, and Docker followed by a short GenAI explanation of the predictions.

Try the demo!: Healthcare Triage Assistant Frontend

---

Overview

The Healthcare Triage Assistant is a ML project designed to support emergency departments by predicting patient triage levels based on vital signs, symptoms, and arrival mode. It provides a user friendly interface for healthcare staff to quickly assess patient urgency and prioritize care effectively. Also includes a GenAI explanation feature to help non-clinical staff understand the model's predictions in an intuitive way.

The project demonstrates the full ML lifecycle:

• Data wrangling & feature engineering
• Model training & evaluation
• FastAPI backend for serving predictions
• Streamlit frontend with intuitive hospital friendly UI
• Dockerized deployment for portability
• GenAI integration for explainability

---

Dataset

This project uses the Synthetic Medical Triage Priority Dataset from Kaggle for model development and evaluation.

---

Project Features

• Predicts triage levels (Routine, Urgent, Emergency, Self-care/Non-urgent)
• FastAPI backend with RESTful API endpoints
• Streamlit frontend with segmented controls and color coded outputs
• Dockerized setup with docker-compose for easy deployment
• Interactive API documentation (/docs) for API exploration
• GenAI integration for explainability

---

Project Structure

Healthcare-Triage-Assistant/
├── app/
│   └── streamlit_app.py                    # Streamlit frontend
├── data/
│   └── raw/
│       └──synthetic_medical_triage.csv     # Raw dataset
├── notebooks/
│   └── cleandata.ipynb                    # Data cleaning & Model training
├── src/
│   ├── api.py                              # FastAPI backend
│   └── model.pkl                           # Trained ML model
├── .dockerignore
├── docker-compose.yml                      # Docker Compose configuration
├── Dockerfile.backend                      # Docker definition for backend
├── Dockerfile.frontend                     # Docker definition for frontend
├── example.env
├── LICENSE
├── README.md
├── requirements-backend.txt                # For docker backend
├── requirements-frontend.txt               # For docker frontend
└── requirements.txt                        # For local deployment

---

Installation

Clone the repo:

git clone https://github.com/rishabhpatel9/Healthcare-Triage-Assistant.git
cd Healthcare-Triage-Assistant

Environment Variables

Before running the application, you need to configure your environment variables:

1. Copy example.env to .env:

cp example.env .env

2. Open .env and fill in the required variables (specifically OPENROUTER_API_KEY for the GenAI explanations).

Install dependencies:

pip install -r requirements.txt

Run FastAPI backend:

uvicorn src.api:app --reload

Run Streamlit frontend:

streamlit run app/streamlit_app.py

---

Deployment with Docker

Build and run with docker-compose:

docker-compose up --build

Use Streamlit app frontend → http://localhost:8501

---

API Documentation

Once backend is running, visit:

• API Documentation → http://localhost:8000/docs
• Redoc → http://localhost:8000/redoc

---

Model Training Features

The triage classifier was trained on patient intake data with the following features:

• Age → Patient’s age in years
• Heart Rate → Beats per minute
• Systolic Blood Pressure → mmHg
• Oxygen Saturation (SpO₂) → Percentage
• Body Temperature → Patient's body temperature (°C)
• Chronic Disease Count → Number of chronic conditions
• Previous ER Visits → Number of prior emergency room visits
• Pain Level → Self-reported scale (0–10)
• Arrival Mode → Walk-in, Ambulance, Referral

Target Variable - Triage Level

Ouput shown as:

• 🟢 Routine
• 🟡 Urgent
• 🔴 Emergency
• 🔵 Self-care / Non-urgent

along with a short GenAI explanation of the prediction.

---

Future Work

• Improve model accuracy with larger datasets
• Integrate hospital database for real patient records
• Add authentication & role-based access control

---

Contributing

Contributions are welcome!

• Fork the repo
• Create a feature branch
• Submit a pull request