Z-Ray: Tri-Modal Medical Diagnostic Engine 🩺

Z-Ray is a high-performance, multi-modal AI diagnostic suite developed by Zeta Coders. It integrates X-Ray and ECG analysis into a unified "Glass-Box" dashboard, providing clinicians with high-accuracy predictions backed by visual heatmaps and clinical fusion.

---

---

✨ Key Features

• Multi-Modal Analysis: Unified interface for Chest X-Ray and 12-Lead ECG.
• Explainable AI (XAI): Real-time Grad-CAM heatmaps for X-Rays and dynamic human-readable clinical reasoning for ECGs.
• Intelligent Signal Processing: Automated lead mapping and sampling rate standardization (Scipy-powered interpolation).
• Clinical Fusion: Random Forest layers that combine neural features with patient demographics (Age/Gender) for superior diagnostic accuracy.
• Edge Optimized: 75% size reduction via INT8 Quantization and ONNX Runtime for sub-100ms inference on standard CPUs.

---

🚀 Core Engine Architecture

🌍 Data Flow Diagram (DFD)

🗄️ Entity-Relationship (ER) Diagram

---

💻 Engine Specifications

1. Vision Engine (Chest X-Ray)

• Backbone: MobileNetV3-Large (Optimized for edge deployment).
• Dataset: NIH Chest X-ray 14 (112,120 clinical images).
• Explainability: Grad-CAM heatmaps highlight acute pathology regions (e.g., Pneumothorax, Effusion).

2. Signal Engine (12-Lead ECG)

• Architecture: 1D-Residual Network (1D-ResNet).
• Feature Engineering:
  • Auto-standardization: Automatically resamples signals to 1000Hz using Cubic Spline Interpolation.
  • Lead Mapping: Intelligent parsing of CSV headers to align disparate lead orderings.
• Reasoning Layer: A dynamic template engine generating varied, context-aware clinical interpretations based on neural confidence and patient age/gender.

---

🛠️ Installation & Setup

1. Clone & Environment

git clone https://github.com/sriramxdev/Z-Ray.git
cd Z-Ray
# Recommendation: use a virtual environment
python -m venv venv
source venv/bin/activate  # Linux/macOS

2. Install Dependencies

pip install -r webui/requirements.txt

3. Launch Platform

cd webui
python server.py
# Default access: http://localhost:5000 (admin / admin)

---

📊 Performance & Optimization

As a project designed for real-world utility on constrained devices, Z-Ray employs advanced optimization techniques:

• INT8 Quantization: Models are compressed from FP32 to INT8, reducing file sizes by ~75% while maintaining >98% of original accuracy.
• ONNX Runtime: Unified cross-platform inference that allows the backend to run on non-dGPU hardware with millisecond latency.
• FOSS Priority: Built entirely using Free and Open Source Software (Fedora, PyTorch, MONAI, FastAPI).

---

📂 Project Structure

Z-Ray/
├── Diagrams/          # SVG/PNG System Architecture Diagrams
├── Notebooks/         # Model Training & Exploratory Notebooks
├── web-backend/       # Core Inference Assets
│   └── deployment/
│       ├── onnx_assets/       # Quantized INT8/FP32 Models
│       └── fusion_assets/     # RF Fusion & Signal Weights
└── webui/             # Unified Dashboard & API Gateway
    ├── server.py              # Flask API & Diagnostic Server
    ├── *.html                 # Multimodal Analysis UI Modules
    └── Diagrams/              # Local Diagram Cache for Portal

Contributions

• Team Zeta Minds, Uhack 4.0