Brain Tumor MRI Classification Dataset (Tumor vs. No Tumor)

Authors: Ahta Shamul Hoque Emran, Hafija Akter, Farhan Masud Nayem
Supervisor: Abdullah Al Shiam

Description:
This dataset has been developed as part of an academic thesis focused on brain tumor detection and classification using deep learning methods. It contains MRI (Magnetic Resonance Imaging) brain scans categorized into two classes: Tumor and No Tumor.
The initial dataset was imbalanced, with a significantly higher number of "No Tumor" images compared to the "Tumor" class. To address this imbalance and improve model performance, we applied data augmentation techniques to the Tumor class.

After augmentation, the dataset contains a nearly balanced number of images across both classes:
-Tumor (Original): 3,671 images
-No Tumor: 13,273 images
-Tumor (After Augmentation): 13,252 images

The total dataset used for training and analysis includes:
Tumor: 13,252 images
No Tumor: 13,273 images
All images are stored in JPG format. The size of the images in this dataset is different.

Augmentation Details:

We used TensorFlow's ImageDataGenerator for on-disk augmentation. The following parameters were applied:
- rotation_range=90: Randomly rotates images up to 90 degrees.
- rescale=1./255: Normalizes pixel values to [0, 1].
- fill_mode='nearest': Fills in missing pixels after transformation.
- Other transformations like shift, shear, zoom, and flipping were not used to retain medical image structure.

This dataset can be used for:
- Training and testing machine learning and deep learning models
- Research on medical image classification
- Developing automated systems for early detection of brain tumors