🧠 Brain Tumor Detection Using CNN with Explainable AI and ViT Comparison

📌 Overview

This project focuses on detecting brain tumors from MRI scans using a Convolutional Neural Network (CNN). To make the model predictions more transparent and interpretable, we integrate Explainable AI (XAI) tools such as:

• 🔍 SHAP (SHapley Additive Explanations)
• 🧩 LIME (Local Interpretable Model-Agnostic Explanations)
• 🔥 Grad-CAM (Gradient-weighted Class Activation Mapping)

To benchmark performance, we also compare the CNN model against a Vision Transformer (ViT) architecture. All experiments are run on Google Colab, utilizing GPU for fast training.

---

✅ Key Features

• 🎯 High-accuracy brain tumor classification using CNN
• 🧠 Interpretability via SHAP, LIME, and Grad-CAM
• 📊 Model comparison with Vision Transformer
• 📸 Visual outputs for prediction and explanation
• ☁️ 100% compatible with Google Colab

---

---

🧠 Models Used

1. CNN (Convolutional Neural Network)

• Built using Keras
• Includes Conv2D, MaxPooling, Flatten, Dense, Dropout layers
• Classifies MRI images as Tumor or No Tumor

2. Vision Transformer (ViT)

• Transformer-based image classification model
• Operates on image patches instead of convolutions
• Used for performance comparison with CNN

3. Explainable AI Tools

Method	Purpose	Output
SHAP	Global + local feature importance	Value-based overlays
LIME	Local explanation using superpixels	Segmented masks
Grad-CAM	Highlights most important image regions	CNN-based heatmaps

---

🧪 Results

🔢 Model Performance

Model	Accuracy	Precision	Recall	F1-Score
CNN (Baseline)	98.5%	98.7%	98.4%	98.5%
CNN + SHAP	98.5%	98.7%	98.4%	98.5%
CNN + LIME	98.5%	98.6%	98.3%	98.4%
CNN + Grad-CAM	98.5%	98.8%	98.4%	98.6%
Vision Transformer	97.2%	97.4%	97.0%	97.2%

📊 Explainability Summary

Tool	Pros	Cons
SHAP	Pixel-level importance, model-agnostic	Slower with large images
LIME	Simple visual output, works with any model	Inconsistent with different segmentations
Grad-CAM	Effective for CNN heatmaps	Not compatible with ViT

✅ Conclusion: The CNN + Grad-CAM combination provided the most interpretable and accurate results.

---

📂 Dataset

• Name: Brain MRI Images for Brain Tumor Detection
• Source: Kaggle
• Classes: Tumor, No Tumor
• Preprocessing:
  • Image resizing to uniform shape
  • Normalization
  • One-hot encoding of labels

---

🔧 How to Run (in Google Colab)

1. Open Google Colab
2. Mount Google Drive or upload dataset directly
3. Run the following notebooks in order:
   • BrainTumorDetection_CNN.ipynb
   • shap_explanation.ipynb
   • lime_explanation.ipynb
   • gradcam_explanation.ipynb
   • ViT_Comparison.ipynb
4. View performance metrics and explanation visualizations in the output cells.

---

📚 References

• 🔗 SHAP
• 🔗 LIME
• 🔗 Grad-CAM
• 🔗 Vision Transformer (ViT)
• 🔗 Brain MRI Dataset (Kaggle)

📦 Requirements

Most packages are pre-installed in Google Colab. For local use:

pip install tensorflow keras numpy matplotlib opencv-python shap lime