🧠 HybridDualPathNet – Brain MRI Disease Classifier

🚀 A hybrid AI architecture for medical image classification combining CNN + Transformer fusion.

A deep learning-based system for classifying brain MRI scans into:

• Alzheimer’s Disease
• Parkinson’s Disease
• Healthy Control

🚀 Achieved 98.46% test accuracy using advanced deep learning techniques.

---

🔍 Overview

HybridDualPathNet is a hybrid deep learning framework that combines EfficientNetV2 and Swin Transformer using cross-attention fusion to accurately classify neurological conditions from MRI scans.

It also integrates Grad-CAM explainability to highlight important brain regions influencing predictions.

---

🧠 Model Architecture

• EfficientNetV2 (Local feature extraction)
• Swin Transformer (Global context learning)
• Cross-Attention Fusion Module
• Focal Loss + Label Smoothing
• OneCycle Learning Rate Scheduler
• Test-Time Augmentation (TTA)

---

🚀 Features

• Hybrid Deep Learning Model (CNN + Transformer)
• Cross-Attention based feature fusion
• Grad-CAM++ for explainability
• Test-Time Augmentation (TTA)
• High-performance classification with strong generalization

---

📊 Model Performance

• ✅ Test Accuracy (TTA): 97.84%
• ✅ Macro AUC-ROC: 0.9981
• ✅ F1 Score: ~0.98 across all classes

---

📸 Results

🔹 Confusion Matrix

🔹 Training Curves

🔹 ROC Curve

---

🛠 Tech Stack

• Python
• PyTorch
• OpenCV
• NumPy
• Matplotlib
• Scikit-learn
• timm

---

📂 Project Structure

NeuroFuseNet/
│
├── notebook/
│   └── NIRJALA_13.ipynb
│
├── outputs/
│   ├── confusion_matrix.png
│   ├── training_curves.png
│   ├── roc_curves.png
│   ├── gradcam_visualizations.png
│   ├── sample_images.png
│   └── results_summary.json
│
├── model/
│   └── (model available via Google Drive)
│
├── README.md
├── requirements.txt

---

▶️ How to Run

1. Clone the repository

git clone https://github.com/Jarpula-Nirjala/NeuroFuseNet.git
cd NeuroFuseNet

2. Install dependencies

pip install -r requirements.txt

3. Run the notebook

Open:

notebook/NIRJALA_13.ipynb

---

⚠️ Dataset

Due to size limitations, the dataset is not included.

👉 Download Dataset: https://drive.google.com/file/d/1AnHbNwv5rBtxYwCBDUwXS_dIGAs2FX1F/view?usp=sharing

---

📥 Model Download

Due to GitHub file size limitations (~195MB), the trained model is hosted on Google Drive:

👉 Download Trained Model: https://drive.google.com/file/d/1rDF-vTOgMSIrpE3rV1OXukyhXiVNxxRq/view?usp=sharing

---

📌 Future Improvements

• Deploy as a web application (Streamlit / Flask)
• Add real-time MRI prediction interface
• Expand dataset for better generalization
• Optimize model for faster inference

---

✨ Author

Jarpula Nirjala 📧 nirjala8462@gmail.com 🔗 https://www.linkedin.com/in/nirjala-jarpula-749346321/

---

⭐ Support

If you like this project, give it a ⭐ on GitHub!