ICU Patient Monitoring System

An actor-based, real-time patient monitoring system for the ICU, designed to simulate a hospital environment. It uses Apache Kafka for streaming patient vital signals and leverages LLMs (Gemini and Groq) for intelligent alert analysis, all presented on a Flask-based real-time dashboard.

✅ Project Overview

• A real-time ICU patient monitoring system.
• Streams ECG, SpO₂, ABP signals etc.
• Uses Pykka actors + Kafka for scalable processing.
• Uses LLMs (Gemini + Groq) for intelligent alert generation.
• Frontend built with Flask + Socket.IO.
• Simulates real-world hospital data environment.

🧠 Core Features

• Actor-based architecture (Supervisor > PatientActor > SignalActor).
• Kafka-driven streaming of patient vitals with 500ms delay.
• LLM actors (Gemini + Groq) for analyzing medical alerts.
• WebSocket-based real-time frontend with signal charts + alerts.
• Rule-based anomaly detection (e.g., flatline, threshold breaches).
• Multi-patient management and signal isolation per patient.

Overview

Patient Monitoring Actor Subsytem Component Diagram

Groq Data Flow diagram

Gemini Data Flow diagram

🛠️ Setup Instructions

1. Clone & Setup Environment

git clone https://github.com/IOutis/ICU_Patient_monitoring_system.git
cd ICU_Patient_monitoring_system
python -m venv monitoring_venv
source monitoring_venv/bin/activate  # On Windows: monitoring_venv\Scripts\activate
pip install -r requirements.txt

2. Prepare Data

• Run these files in the given order

1. data_preparation/data/load_data.py
2. data_preparation/data_download/tracks_data_download.py

3. Ensure Kafka Is Running

• Kafka must be running on default ports (localhost:9092).
• Required topics must be created beforehand.
• You can use Docker or a local installation to run Kafka.

🚀 How to Run the System (in 5 Terminals)

Run the following in separate terminals from the root directory:

# 1. Gemini LLM Actor
python -m monitoring_system.gemini_llm_class

# 2. Groq LLM Actor
python -m monitoring_system.groq_llm_class

# 3. Flask + Socket.IO Webserver
python -m webserver.server

# 4. Kafka Producer (streams vitals every 500ms)
python monitoring_system/pykka_producer.py

# 5. Kafka Consumer + Actor System
python -m monitoring_system.pykka_consumer

These processes are designed to run concurrently and communicate via Kafka + actor messaging.

🌐 Access the Dashboard

Open in your browser:

http://localhost:5000

You'll see:

• Live patient signal graphs (ECG, SpO₂, ABP etc.)
• Real-time anomaly alerts
• Streamed via WebSocket + actor signals

🎥 Demo Video

Demo video

🧠 Technologies Used

• Python: Main backend language
• Pykka: Actor model framework
• Kafka: Message queue for real-time signal streaming
• Flask + Socket.IO: Real-time web frontend
• LLMs: Gemini + Groq for intelligent alert handling
• HTML/JS: Dashboard frontend templates

⚙️ Customization & Environment

• A .env file can store:
  • GEMINI_API
  • GEMINI_API_2
  • GROQ_API_1
  • GROQ_API_2
  • GROQ_API_3
  • Kafka topic names
• You can modify signal file paths or thresholds in pykka_producer.py and the consumer logic.

✅ Future Plans

• Add model-based anomaly detection (if labeled data is available)
• Add patient authentication/login
• Switch to a WebSocket backend with FastAPI for scalability
• Dockerize the full pipeline for easier deployment

🧑 Author

Mohd Mushtaq
Aspiring AI Agent Engineer | Full-Stack Developer
GitHub: https://github.com/IOutis