🧠 NeuroScan AI - Brain Tumor Detection & Neural Explainability

NeuroScan AI is an end-to-end deep learning web application for brain tumor classification from MRI scans. It features a fine-tuned VGG-19 model, Grad-CAM (Gradient-Weighted Class Activation Mapping) visualization for explainable AI, and a modern dark-themed medical dashboard.

---

📊 Fine-Tuning Performance & Results

Model Stage	Train Accuracy	Validation Accuracy	Test Accuracy	Test Loss
Before Fine-Tuning	84.22%	86.13%	84.52%	0.3130
After Fine-Tuning	95.57%	93.23%	93.55%	0.1785

Fine-Tuning Improvements Applied:

1. Top-Block Unfreezing: Unfroze Block 5 layers (block5_conv1 to block5_conv4) of VGG-19 to adapt high-level texture features specifically to brain MRI tissue contrast.
2. Optimized Learning Rate: Used Adam(lr=5e-5) with ReduceLROnPlateau for stable convergence.
3. Batch Size Tuning: Reduced batch size from 64 to 32 on 1,445 training samples to double gradient update frequency per epoch.
4. Regularization: Increased Dropout to 0.3 to eliminate overfitting and boost test set generalizability.

---

✨ Features

• ⚡ Dynamic Model Hot-Reloading: Flask backend monitors model.h5 modification timestamp on disk and hot-reloads model weights seamlessly without server downtime.
• 🎯 Grad-CAM Visual Heatmaps: Generates neural activation heatmaps from block5_conv4 overlaid onto brain MRI scans using OpenCV JET colormaps.
• 🎨 Modern Medical Dashboard:
  • Analysis Tab: Drag-and-drop MRI scan loading, preview, SVG confidence gauge ring, and Grad-CAM heatmap visualization.
  • Reports Tab: Interactive diagnostic log table with search filtering, summary stats (Total Scans, Tumor Count, Avg Confidence), and side-by-side report detail modal.
  • Settings Tab: Customizable Grad-CAM target layers, colormaps (JET, VIRIDIS, HOT, INFERNO), alpha transparency slider, and manual weight reload triggers.

---

🛠️ Project Structure

Brain Tumor/
├── app.py                      # Flask Server with Grad-CAM & Weight Hot-Reloading
├── brain_tumor_dataset/        # Training script (train_brain_tumor.py)
├── templates/
│   └── index.html              # NeuroScan AI Dashboard UI
├── static/
│   ├── css/style.css           # Vanilla CSS3 Dark Medical Theme
│   └── js/app.js               # Frontend Controller & REST API Handlers
├── Detection.py                # Preprocessing & evaluation utilities
├── training_performance.png    # Training vs Validation Accuracy/Loss plot
├── README.md                   # Project documentation
└── .gitignore                  # Git ignore specifications

---

🚀 How to Run Locally

1. Prerequisites

Ensure Python 3.10+ and TensorFlow are installed:

pip install tensorflow opencv-python flask numpy matplotlib imutils requests

2. Launch Backend Application

Run the Flask backend server:

python app.py

3. Open Dashboard

Open your browser and navigate to:

http://127.0.0.1:5000

Upload any brain MRI scan (JPG/PNG) to analyze and visualize Grad-CAM activation heatmaps!