Skip to content

README.md

Latest commit

 

History

History
54 lines (41 loc) · 3.24 KB

README.md

File metadata and controls

54 lines (41 loc) · 3.24 KB
Raw

PDAC Sentinel v3.0: Metabolic Decoupling Detection Engine

Streamlit App Python 3.10+ License: MIT

🔬 Mission Statement

Our primary goal is to save lives through the early detection of Pancreatic Ductal Adenocarcinoma (PDAC). By identifying the "Metabolic Decoupling" signature—where blood glucose rises while body weight falls—Sentinel provides a critical window for intervention before the disease reaches advanced stages.

🚀 Access the Live Clinical Dashboard

🚀 Overview

PDAC Sentinel is a Clinical Decision Support System (CDSS) designed for General Practitioners (GPs). It bridges the gap between raw longitudinal data and clinical action by integrating deterministic medical protocols with state-of-the-art Large Language Models (LLMs).

🛠️ System Architecture

The platform follows a robust three-tier architecture designed for reliability and safety:

  1. Clinical Engine (Deterministic): Evaluates longitudinal data (HbA1c and BMI deltas) to calculate a risk score (0-5) based on validated oncology protocols (END-PAC & UK-EDI).
  2. AI Briefing Agent (LLM): Utilizes Qwen2.5-Coder-32B via Hugging Face Inference API to synthesize complex metabolic flags into professional executive summaries.
  3. Safety Gate: A hybrid validation layer that prevents AI hallucinations by forcing deterministic outputs in low-risk cases.

📊 Clinical Intelligence

The system monitors specific biomarkers over a 12-month horizon to detect PDAC's paraneoplastic effects:

  • Hyperglycemia: HbA1c increase ≥ 1.0%.
  • Involuntary Weight Loss: BMI decrease ≥ 1.0 point.
  • New-Onset Diabetes (NOD): Focus on patients diagnosed within <36 months.
  • Treatment Resistance: Failure of Metformin to stabilize glucose levels in the presence of weight loss.

💻 Technical Stack & Engineering Highlights

  • Core: Python 3.10+ (Clean Code, Modular Design).
  • AI Integration: Hugging Face Inference API with smolagents for structured reasoning.
  • Data Integrity: Pydantic models for strict validation of clinical inputs.
  • Frontend: Streamlit for real-time data visualization and UX.
  • Security: Environment-based secret management (No hardcoded API keys).

🌟 Why this matters for Applied AI:

  • Hybrid AI: Combines the reliability of hard-coded medical rules with the synthesis power of LLMs.
  • Production-Ready: Developed with CI/CD compatibility and cloud-native deployment.
  • Ethical AI: Implements "Guardrails" to ensure high-stakes medical decisions are always backed by data-driven flags.

📂 Project Structure

├── dashboard_sentinel.py  # Streamlit UI & Orchestration
├── agent_logic.py        # LLM integration & Prompt Engineering
├── clinical_logic.py     # Deterministic Risk Scoring Engine
├── schemas.py           # Pydantic data models
└── requirements.txt      # Dependency management