MRI Brain Tumor Classification (CNN + Transfer Learning)

A reproducible workflow for classifying brain MRI scans into four categories:

• Glioma Tumor
• Meningioma Tumor
• Pituitary Tumor
• No Tumor

This repository centers around a single notebook: MRIBrainTumorDetection.ipynb, which walks through:

• S1: Image preprocessing and loading with Keras ImageDataGenerator
• S2: A baseline CNN built from scratch
• S3: Transfer learning with VGG16
• S4: Fine-tuning, evaluation, and visualization

Medical disclaimer: This project is for research and educational purposes only and is not intended for clinical use.

Features

• End-to-end pipeline in one notebook
• Augmented training with Keras generators
• Transfer learning using VGG16 with optional fine-tuning
• Early stopping and training curves
• Confusion matrix, classification report, and sample predictions

Dataset

• Source: Brain Tumor MRI Dataset (Kaggle)
• URL: https://www.kaggle.com/datasets/sartajbhuvaji/brain-tumor-classification-mri

Expected directory structure:

datasets/
  Training/
    glioma_tumor/
    meningioma_tumor/
    pituitary_tumor/
    no_tumor/
  Testing/
    glioma_tumor/
    meningioma_tumor/
    pituitary_tumor/
    no_tumor/

In the notebook, set:

• train_path = '/content/drive/MyDrive/datasets/Training'
• test_path = '/content/drive/MyDrive/datasets/Testing'

Update paths if running locally.

Environment

• Python 3.9+
• TensorFlow 2.10+ (GPU recommended)
• NumPy, Matplotlib, scikit-learn
• Jupyter or VS Code with Python extension

Install dependencies:

python -m venv .venv
. .venv/Scripts/activate    # Windows
# source .venv/bin/activate # macOS/Linux
pip install --upgrade pip
pip install tensorflow numpy matplotlib scikit-learn jupyter

Quick Start

Option A: Google Colab (recommended)

1. Upload this repo or open the notebook in Colab.
2. Mount Google Drive in the notebook:

   from google.colab import drive
   drive.mount('/content/drive', force_remount=True)

3. Place the dataset under /content/drive/MyDrive/datasets/.
4. Run all cells in MRIBrainTumorDetection.ipynb.

Option B: Local (VS Code/Jupyter)

1. Place the dataset under datasets/Training and datasets/Testing.
2. Open MRIBrainTumorDetection.ipynb in VS Code or Jupyter.
3. Update train_path and test_path to your local folders.
4. Run the notebook cells in order.

Training

The notebook contains two training flows:

• Baseline CNN:

  • Small 3-block CNN
  • Categorical cross-entropy, Adam optimizer
  • 224×224 image size

• Transfer Learning (VGG16):

  • ImageNet weights, include_top=False
  • GlobalAveragePooling + Dense classifier head
  • Phase 1: freeze base, train head
  • Phase 2: unfreeze top layers, fine-tune with lower LR
  • Early stopping on validation loss

Adjust epochs, batch_size, and augmentation params as needed.

Evaluation

The notebook provides:

• Test accuracy via model.evaluate()
• Confusion matrix and classification report (precision, recall, F1)
• Visualization of predictions vs. ground truth

Example metric definition:

• Accuracy: $Accuracy=\frac{TP+TN}{TP+TN+FP+FN}$

Inference and Saving

Add this at the end of the notebook to save your best model:

model.save("brain_tumor_vgg16.h5")

To run inference on a single image, add a helper cell:

import numpy as np
import tensorflow as tf
from tensorflow.keras.applications.vgg16 import preprocess_input
from tensorflow.keras.preprocessing import image

class_names = list(test_set.class_indices.keys())

def predict_image(img_path):
    img = image.load_img(img_path, target_size=(224, 224))
    x = image.img_to_array(img)
    x = np.expand_dims(x, axis=0)
    x = preprocess_input(x)  # match preprocessing used in training
    probs = model.predict(x)[0]
    pred_idx = np.argmax(probs)
    return class_names[pred_idx], float(probs[pred_idx])

label, confidence = predict_image("path/to/image.jpg")
print(f"Predicted: {label} ({confidence:.3f})")

Repository Structure

.
├─ MRIBrainTumorDetection.ipynb   # Main end-to-end notebook
└─ README.md                      # This file

Roadmap

• Add Grad-CAM visualizations for interpretability
• Export SavedModel/TF Lite for deployment
• Hyperparameter search (KerasTuner/Optuna)
• Add ResNet50 and EfficientNet baselines

Acknowledgments

• Dataset by Sartaj Bhuvaji on Kaggle
• VGG16 from Keras Applications (ImageNet weights)

License

This project is released under the MIT License. See LICENSE for details.

Responsible AI/Medical Disclaimer

• Do not use this model for diagnosis or clinical decision-making.
• Performance can vary across institutions and imaging protocols.
• Always consult certified medical professionals for clinical use.