This project focuses on Brain Tumor Classification from MRI images using a Custom Convolutional Neural Network (CNN) built from scratch with TensorFlow and Keras.
The model classifies MRI scans into four categories:
- Glioma
- Meningioma
- Pituitary Tumor
- No Tumor
The goal of this project is to build an accurate, well-generalized, and explainable deep learning model and deploy it using Streamlit for real-time predictions.
- ✅ Custom CNN architecture (no transfer learning)
- ✅ Proper train–validation–test split
- ✅ Data augmentation tuned for medical MRI images
- ✅ Learning rate scheduling and early stopping
- ✅ Achieved 85.2% accuracy on unseen test data
- ✅ Deployed using Streamlit with live predictions
- Source: Kaggle – Brain Tumor MRI Dataset
- Image Type: MRI scans (visually grayscale, stored as RGB)
- Classes: 4
Dataset/
├── Training/
│ ├── glioma/
│ ├── meningioma/
│ ├── pituitary/
│ └── notumor/
└── Testing/
├── glioma/
├── meningioma/
├── pituitary/
└── notumor/
- Training data: Used for model learning
- Validation data: 20% split from training set
- Test data: Completely unseen during training
- Python
- TensorFlow / Keras
- NumPy
- Matplotlib
- Google Colab
- Streamlit
The model was designed from scratch using convolutional blocks optimized for medical images:
- Convolution + ReLU activation
- Batch Normalization
- MaxPooling layers
- Dropout for regularization
- Dense layers for classification
- Softmax output layer
(224, 224, 3)
MRI images appear grayscale but are stored and processed as RGB images with identical channels.
MRI-safe augmentation techniques were applied:
- Image resizing and normalization
- Small rotations
- Zooming
- Width and height shifting
- Horizontal flipping
Aggressive augmentation was avoided to preserve medical image integrity.
-
Optimizer: Adam
-
Learning Rate: 3e-4
-
Loss Function: Categorical Crossentropy
-
Callbacks:
- EarlyStopping (restore best weights)
- ReduceLROnPlateau
The best model was automatically restored based on validation loss.
| Metric | Value |
|---|---|
| Training Accuracy | ~92% |
| Best Validation Accuracy | 82.3% |
| Final Test Accuracy | 85.2% |
| Test Loss | 0.6078 |
The following plot shows training and validation accuracy during model training:
The trained model is deployed using Streamlit, providing an interactive interface for real-time MRI classification.
- Upload MRI images
- Real-time tumor classification
- Clear display of predicted class
Click the link below to watch the Streamlit app running live:
ℹ️ GitHub does not play videos inline. The link is clickable and downloadable.
The trained model is saved in both formats:
brain_tumor_model.h5
brain_tumor_model.keras
.h5→ Deployment & compatibility.keras→ Future-proof format
git clone https://github.com/Sayeem3051/MRI-Classification1.git
cd MRI-Classification1pip install -r requirements.txtstreamlit run app.py- Built a CNN from scratch for medical imaging
- Applied correct validation strategies
- Controlled overfitting using callbacks
- Deployed a deep learning model using Streamlit
- Use transfer learning (MobileNetV2 / ResNet)
- Add Grad-CAM for model explainability
- Improve class-wise performance
- Deploy as a public web application
This project demonstrates a complete deep learning pipeline from data preprocessing and model training to evaluation and deployment. The achieved performance validates the effectiveness of a well-designed Custom CNN for brain tumor classification.
⭐ If you find this project useful, consider starring the repository!