Brain Tumor Classification with CNNs (Keras)

This project implements convolutional neural networks (CNNs) for binary brain tumor classification using Keras with a TensorFlow backend.
The goal is to compare different CNN architectures and evaluate their performance on an imbalanced medical image dataset.

---

Dataset

• Images are resized to 256×256
• Binary classification: tumor / no tumor
• Class distribution is checked before training

---

Models

Three CNN architectures are implemented and trained:

1. Simple CNN – baseline model
2. Deeper CNN – more convolutional layers and parameters
3. BatchNorm CNN – uses batch normalization for more stable training

All models use:

• Adam optimizer
• Binary cross-entropy loss
• Metrics: accuracy, recall, precision

Training

• Each model is trained for 20 epochs
• Learning curves are plotted:
  • loss
  • accuracy
  • recall

Because the dataset is imbalanced, recall is treated as the main metric.

---

Libraries

• Python
• TensorFlow / Keras
• NumPy
• Matplotlib
• Seaborn
• scikit-learn