Digital Patient Sentinel 🛡️🏥

📖 Project Overview

Digital Patient Sentinel is an end-to-end Machine Learning pipeline for intensive care units (ICU). The system integrates static clinical data (EHR) with dynamic sensor signals (IoT) to predict acute critical risks, such as Sepsis.

"The goal is not just to train a model, but to engineer a trustworthy system that bridges the gap between raw data and clinical decision support."

---

🎯 Key Features (Aligned with Industry Standards)

1. 🔒 Privacy & GDPR Compliance

Privacy is not an afterthought; it is baked into the ingestion phase (src/privacy.py).

• Pseudo-Anonymization: Automatically replaces raw patient IDs with SHA-256 cryptographic hashes.
• Data Minimization: Generalizes sensitive dates to prevent re-identification, ensuring ethical handling of patient data from step one.

2. 🧬 Biomedical Knowledge Integration

The model moves beyond "black-box" approaches by incorporating physiological domain knowledge:

• Shock Index (SI): Calculates Heart Rate / Systolic BP as an early indicator of hemodynamic instability.
• Pulse Pressure: Derived feature to monitor cardiovascular health.
• Vitals Logic: Automatic flagging of hypoxia and fever events based on sensor thresholds.

3. ⚙️ Modular Data Pipeline

Built as a structured Python package to simulate a production environment:

• Ingestion: Simulates heterogeneous data sources (Clinical demographics + High-frequency vitals).
• Cleaning & Encoding: Robust handling of categorical data and missing values.
• Modeling: Random Forest classifier optimized for clinical interpretability (Recall/Precision).

4. 🩺 Clinical Validation & Explainability

Trust is key in healthcare. The pipeline outputs a "White-box" validation report, prioritizing Feature Importance to prove that the model relies on physiological signals (e.g., Lactate, Blood Pressure) rather than biased demographic proxies.

---

🏗️ System Architecture

The data flows through a linear, audit-ready pipeline:

graph LR
    A[Raw ICU Data] -->|Ingestion| B(Privacy Module)
    B -->|SHA-256 Hashing| C{Biomedical Feature Eng.}
    C -->|Calculation| D["ML Model (Random Forest)"]
    D -->|Prediction| E[Clinical Validation Report]

Loading

📂 Repository Structure

The project is organized as a maintainable Python package.

digital_patient_sentinel/
│
├── src/                        # Source Code (Python Package)
│   ├── __init__.py             # Package initializer
│   ├── data_loader.py          # Synthetic Data Generator (Mocking Hospital DB)
│   ├── privacy.py              # GDPR Logic & Hashing
│   ├── features.py             # Biomedical Feature Engineering
│   └── model.py                # ML Model Wrapper & Validation Reports
│
├── assets/                     # Images for documentation
├── main.py                     # Pipeline Orchestrator (Entry Point)
├── requirements.txt            # Project dependencies
└── README.md                   # Technical Documentation