NeoScan-Website

NeoScan AI — Advanced Neonatal Screening 🍼

NeoScan AI is a state-of-the-art diagnostic platform designed for early neonatal screening. Utilizing advanced Deep Learning architectures, the system analyzes images of newborns to detect 15+ medical conditions across four specialized modules, providing immediate risk assessment and clinical advice.

---

🚀 Key Features

• 🧠 Multi-Module AI: Specialized neural networks for Skin, Eye, Face, and Body analysis.
• ⚡ Real-Time Inference: Powered by FastAPI with support for both Live (Model-based) and Demo (Deterministic) modes.
• 📱 Responsive Interface: Modern, mobile-friendly UI for bedside clinical use.
• 🩺 Clinical Decision Support: Provides risk levels (Normal, Moderate, High Risk) and actionable advice for healthcare providers.

---

🧩 Diagnostic Modules

1. Skin Analysis Module

• Model: EfficientNetB3 (PyTorch)
• Detections: Jaundice, Cyanosis, Anemia, Normal.
• Preprocessing: Gray World White Balance & CLAHE for robust performance under phototherapy/hospital lighting.

2. Eye Analysis Module

• Model: EfficientNetB4 (TensorFlow/Keras)
• Detections: Retinopathy of Prematurity (ROP) Stages 1-3, Cataracts/Leukocoria.
• Pipeline: Automated contrast enhancement and Gaussian sharpening for retinal clarity.

3. Face Analysis Module (Genetic/Congenital)

• Model: Ensemble (VGG16 + Vision Transformer - ViT)
• Detections: Down Syndrome, Cleft Lip/Palate, Healthy Features.

4. Body Analysis Module — Skeletal X-Ray (Updated)

• Model: ResNet50 (TensorFlow/Keras) — Transfer Learning with 2-Phase Training
• Dataset: Bone Fracture Multi-Region X-Ray Dataset
• Detections: Fracture Detected, No Fracture (Normal)
• Training: Phase 1 (Frozen Base) + Phase 2 (Fine-Tuned top 30 layers)
• Performance: ~83% Validation Accuracy
• Pipeline: X-Ray image preprocessing → ResNet50 inference → Risk level classification
• Output: Fractured (High Risk) / Not Fractured (Normal) with confidence score

---

🛠️ Tech Stack

• Frontend: HTML5, CSS3 (Vanilla), JavaScript (ES6+)
• Backend: Python 3.10+, FastAPI, Uvicorn
• AI/ML: PyTorch, TensorFlow, Keras, OpenCV, TIMM
• Deployment: GitHub Pages (Frontend), Local/Cloud (Backend)

---

📦 Installation & Setup

1. Clone the Repository

git clone https://github.com/bytecraft17/NeoScan-Website.git
cd NeoScan-Website

2. Setup Backend

cd Backend
pip install -r requirements.txt

3. Model Weights

The pre-trained model weights (.pth, .keras, .h5) are not included in the repository due to their size.

• Download Weights: Google Drive Folder
• Place all weight files in the Backend/ directory.

File	Module	Framework
best_model.pth	Skin (EfficientNetB3)	PyTorch
neoscan_eye_FINAL_effb4.keras	Eye (EfficientNetB4)	TensorFlow
VGG16_best.pth	Face - VGG16	PyTorch
ViT_v2_best.pth	Face - ViT	PyTorch
body_model.h5	Body - ResNet50 (Bone Fracture)	TensorFlow

4. Run the Application

python main.py

The application will be available at http://127.0.0.1:8000.

---

📂 Project Structure

NeoScan_Website/
├── Backend/
│   ├── main.py                        # FastAPI Entry Point & Prediction Logic
│   ├── requirements.txt               # Python Dependencies
│   ├── eye_model_config_FINAL.json    # Eye Module Config
│   └── *.pth / *.keras / *.h5        # AI Model Weights (Download separately)
├── Frontend/
│   ├── index.html                     # Main Application UI
│   ├── css/                           # Styling & Design System
│   └── js/                            # Frontend Logic & API Integration
└── README.md

---

🔄 Changelog

v2.0 — Body Module Overhaul

• Replaced: CTG (Cardiotocography) fetal distress model
• New Model: ResNet50 Bone Fracture X-Ray classifier (TensorFlow/Keras)

• New Classes: Fractured (High Risk)

  Not Fractured (Normal)

• Accuracy: ~83.23% on Bone Fracture Multi-Region X-Ray Dataset
• Backend: Updated /predict/body endpoint in main.py
• Removed: body_model.pkl, debug_tf.py

---

👨‍💻 AI Notebooks & Research

For the full training pipelines, datasets, and architectural experiments, visit the research repository: Neoscan-AI Research Repo

🌍 Live Website: https://bytecraft17.github.io/NeoScan-Website/Frontend/

---

Disclaimer: NeoScan AI is an assistive screening tool and should not replace professional clinical judgment.