🩺 HemaVision AI

Next-Generation Medical Imaging Analysis & Triage System

---

🚨 Medical Disclaimer

CRITICAL WARNING: HemaVision AI is a Research Prototype.

• NOT FDA approved.
• NOT intended for clinical diagnosis, treatment, or patient management.
• No medical decisions should be made based on this software.
• Always verify results with a certified radiologist.

---

� Overview

HemaVision AI is a state-of-the-art framework designed to assist medical researchers in analyzing complex imaging data. moving beyond simple classification, HemaVision integrates Generative AI for semantic reporting and 3D Reconstruction for structural visualization.

🌟 Key Features

• 🤖 AI-Powered Triage: Automated risk assessment using ensembles of CheXNet and DenseNet models.
• 📝 Semantic Reporting: LLM-integrated generation of patient-centric explanations (via Gemini API).
• 🧊 3D Bone Reconstruction: Experimental point-cloud generation from 2D input slices.
• 🛡️ Privacy-First Architecture: "Zero-Retention" design ensures no patient data persists after analysis.
• 🔥 Heatmap Visualization: Grad-CAM integrations to highlight regions of interest.

---

🏗️ System Architecture

The repository uses a clear, decoupled structure to ensure scalability and safety.

graph TD
    User[Clients / Researchers] -->|Upload Scan| API[Flask API Gateway]
    API -->|Pre-process| Inference[Inference Engine]
    Inference -->|2D Analysis| Models[CNN Ensembles]
    Inference -->|Report Gen| LLM[Gemini Integration]
    Inference -->|3D Mesh| Recon[3D Reconstruction Module]
    API -->|JSON Response| User
    API -.->|Audit Log| DB[(Metadata DB)]

Loading

� Directory Map

Directory	Purpose
src/app	Core API server & request routing (Flask)
src/inference	Neural network prediction pipelines
src/models	Model definitions (Weights are strictly ignored)
src/utils	DICOM parsing, image normalization, & helpers
frontend/	Standalone prototypes (3D Viewer)
scripts/	Training routines & audit tools
docs/	Compliance policy & architectural decisions

---

🚀 Getting Started

Prerequisites

• Docker Desktop (Recommended)
• Python 3.9+ (For local dev)
• Git

⚡ Quick Start (Docker)

The easiest way to run HemaVision AI is via Docker Compose.

1. Clone the Repository

   git clone https://github.com/cod-x-prince/med_image.git
   cd med_image

2. Launch Services

   docker-compose up --build

   • API: http://localhost:5000
   • Docs: http://localhost:5000/docs

🔧 Local Installation

1. Initialize Environment

   python -m venv venv
   source venv/bin/activate  # Windows: .\venv\Scripts\Activate

2. Install Dependencies

   pip install -r requirements.txt

3. Model Setup

   Note: This repo does NOT contain model weights (>100MB). Place your .pth or .onnx checkpoints in src/models/ manually.

4. Run Application

   python run.py

---

🛡️ Security & Compliance

We adhere to strict DevSecOps principles for medical software:

1. No Secrets In Code: All credentials are managed via environment variables.
2. Data Isolation: Medical images are processed in-memory and immediately discarded.
3. Audit Trails: Non-PII logging tracks system performance and error rates.

See Data Governance Policy for details.

---

🗺️ Roadmap

• Phase 1: Core Architecture & Security Audit
• Phase 2: React/Next.js Client Migration
• Phase 3: OHIF Viewer Integration
• Phase 4: DICOMweb Standard Support

---

🤝 Contributing

Contributions are welcome! Please read docs/architecture.md before submitting a PR. Rule #1: NEVER commit medical data (DICOM/NIfTI) or Secrets.

� License

Distributed under the MIT License. See LICENSE for more information.

---

Built with ❤️ by the HemaVision AI Team