Brain-Tumor-TransferLearning

Dataset:

• The dataset utilized for this study is the Brain Tumor MRI Dataset sourced from Kaggle.
• It consists of a carefully curated collection of brain MRI scans specifically chosen to facilitate research in automated brain tumor detection and classification using the Keras library.
• The dataset aims to enhance diagnostic accuracy and includes a randomized subset (20% of the original data) categorized into 'yes' (tumor present) and 'no' (healthy) tumor classes for both training and validation purposes.

Investigated Approaches:

1. VGG16 + FC (VGG16 with Fully Connected Layers):

   • VGG16 is a deep CNN architecture comprising 16 weight layers.
   • It was pre-trained on the ImageNet dataset.
   • Additional fully connected layers are incorporated for classification.

2. VGG + CNN2D (VGG16 with Additional Convolutional Layers):

   • This approach extends VGG16 by introducing more convolutional layers for enhanced feature extraction.

3. ResNet50 + FC (ResNet50 with Fully Connected Layers):

   • ResNet50, a deep residual network with 50 layers, addresses the vanishing gradient problem using skip connections.
   • Fully connected layers are added for classification.

4. ResNet50 + CNN2D (ResNet50 with Additional Convolutional Layers):

   • Building upon ResNet50, this variant incorporates additional convolutional layers.

5. ViT (Vision Transformer) + FC (Fully Connected Layers):

   • Originally designed for natural language processing, ViT is a transformer-based architecture.
   • It has been adapted for image classification by treating images as sequences of patches.
   • Fully connected layers are employed for classification⁷.

Investigation Objectives:

• Evaluate the effectiveness of transfer learning using pre-trained models (VGG16, ResNet50, and ViT) for brain tumor classification.
• Compare the performance of different architectures in terms of accuracy, sensitivity, specificity, and other relevant metrics.
• Gain insights into how transfer learning impacts model convergence, generalization, and robustness.

Instructions for Kaggle API

1. Download Kaggle API:

   • Install the Kaggle API by running pip install kaggle.

2. Kaggle API Token:

   • Go to your Kaggle account settings and generate an API token.
   • Save the token as kaggle.json in the root directory of this repository.

3. Download Dataset:

   • Use the Kaggle API to download the dataset:

     kaggle datasets download -d  preetviradiya/brian-tumor-dataset
     

4. Upload Kaggle API Token to Colab/Notebook:

   • If using Colab or Jupyter Notebook, upload the kaggle.json token to your environment.
   • Use the following code snippet:

     from google.colab import files
     files.upload()

5. Unzip Dataset:

   • Unzip the downloaded dataset:

     unzip brian-tumor-dataset.zip