Brain Tumor Detection System

Tumor Classification & Segmentation using Deep Learning

Overview

1. Brain Tumor Classification
   • Classifies MRI images into categories (e.g., glioma, meningioma, pituitary tumor, or healthy).
   • Uses Convolutional Neural Networks (CNNs) for automated medical image analysis.
2. Brain Tumor Segmentation
   • Identifies and highlights tumor regions within MRI scans.
   • Employs U-Net (and/or other segmentation architectures) for pixel-wise classification.

Project Structure

Brain-Tumor-Detection-System
 ├── ML Project 1 Brain_Tumor_classification.ipynb   # Classification notebook
 ├── ML Project 2 brain_tumor_segmentation.ipynb     # Segmentation notebook
 ├── data/                                          # Dataset (MRI images)
 ├── models/                                        # Saved models
 ├── results/                                       # Predictions, segmented masks, visualizations
 ├── requirements.txt                               # Dependencies
 └── README.md                                      # Project documentation

Installation & Setup

1. Clone the repository:
   bash

   1git clone https://github.com/ayelefransi/Brain-Tumor-Detection-System.git 2cd Brain-Tumor-Detection-System
2. Install dependencies:
   pip install -r requirements.txt
3. Datasets:
   • Classification: MRI datasets with labeled tumor classes.
   • Segmentation: MRI scans with annotated masks. (Kaggle datasets such as Brain Tumor MRI Dataset and Brain Tumor Segmentation (BraTS) can be used.)

Usage

1. Classification

• Loads MRI images
• Preprocesses data (resizing, normalization, augmentation)
• Trains a CNN model
• Evaluates performance (accuracy, precision, recall, F1-score)

2. Segmentation

• Loads MRI scans and masks
• Preprocesses data (contrast enhancement, normalization)
• Trains U-Net (or similar)
• Generates segmented tumor regions
• Visualizes predictions vs. ground truth

Results

• Classification
  • Achieved high accuracy in distinguishing tumor types.
  • Confusion matrices and classification reports available in the notebook.
• Segmentation
  • Successfully segmented tumor regions with high Dice Similarity Coefficient (DSC).
  • Visual comparisons between predicted masks and ground truth provided.

Future Improvements

• Integrate classification and segmentation into a single pipeline.
• Deploy as a web application (e.g., Streamlit, Flask, or FastAPI).
• Incorporate explainable AI (Grad-CAM, SHAP) for model interpretability.
• Use larger datasets and transfer learning for improved generalization.

Requirements

• Python ≥ 3.8
• TensorFlow or PyTorch
• OpenCV
• NumPy, Pandas, Matplotlib, Seaborn
• scikit-learn
• Jupyter Notebook

Applications

• Assisting radiologists in early tumor detection.
• Supporting treatment planning with precise tumor localization.
• Enabling research in medical imaging and deep learning.

License