Brain Tumor Segmentation Using Deep Learning

This project implements a deep learning-based approach for brain tumor segmentation in MRI images using a U-Net architecture with a pre-trained ResNet50 encoder. The model aims to accurately identify and isolate brain tumors, assisting in medical image analysis and potentially aiding radiologists in diagnosis.

---

Table of Contents

• Project Overview
• Dataset
• Data Exploration
• Model Architecture
• Preprocessing & Augmentation
• Training
• Evaluation & Visualization
• Usage
• References

---

Project Overview

The goal of this project is to build a semantic segmentation model to detect brain tumors in MRI images. Semantic segmentation involves classifying each pixel of an image into predefined categories—in this case, tumor vs. non-tumor regions.

Key features:

• Uses deep learning for accurate segmentation.
• Employs transfer learning with a pre-trained ResNet50 as the encoder.
• Handles class imbalance with a custom Dice coefficient loss function.
• Provides visualizations for qualitative evaluation.

---

Dataset

The project uses the kaggle_3m dataset, which contains MRI images and corresponding segmentation masks.

• Data File: data_mask.csv

• Columns:

  • patient_id: Unique identifier for each patient
  • image_path: Path to the MRI image
  • mask_path: Path to the corresponding tumor mask
  • mask: Binary label (0 = no tumor, 1 = tumor)

• Number of entries: 3,929

  • Images without tumors: 2,556
  • Images with tumors: 1,373

Note: The dataset is imbalanced, so specialized loss functions like Dice coefficient are used to mitigate this during training.

---

Data Exploration

Initial exploration of the dataset includes:

• Checking image and mask distribution
• Visualizing sample MRI images and masks
• Identifying data imbalance and class distribution

# Example: Visualizing sample MRI images and masks
import matplotlib.pyplot as plt
import cv2

image = cv2.imread(image_path)
mask = cv2.imread(mask_path, cv2.IMREAD_GRAYSCALE)
plt.subplot(1, 2, 1)
plt.imshow(image)
plt.title("MRI Image")
plt.subplot(1, 2, 2)
plt.imshow(mask, cmap='gray')
plt.title("Mask")
plt.show()

---

Model Architecture

The model uses a U-Net convolutional neural network, which is widely adopted for biomedical image segmentation tasks.

Key points:

• Encoder: Pre-trained ResNet50
• Input size: (256, 256, 3)
• Total layers: 175
• Total parameters: ~23 million
• Purpose of U-Net: Combines encoding (feature extraction) with decoding (upsampling) to produce precise pixel-wise segmentation maps.

---

Preprocessing & Augmentation

Images are preprocessed before training:

• Rescaling: Pixel values are normalized
• Data augmentation: Using ImageDataGenerator to apply transformations like rotation, flipping, and zoom to increase dataset diversity.

from tensorflow.keras.preprocessing.image import ImageDataGenerator

datagen = ImageDataGenerator(
    rescale=1./255,
    rotation_range=10,
    width_shift_range=0.1,
    height_shift_range=0.1,
    horizontal_flip=True
)

---

Training

Configuration:

• Loss function: dice_coef_loss – effective for imbalanced datasets and measures overlap between predicted and ground truth masks.

• Optimizer: Adam

• Metrics:

  • Dice Coefficient (dice_coef)
  • Jaccard Index (jaccard_coef)

Training steps:

1. Load images and masks
2. Split data into training and validation sets
3. Fit the model with data generators
4. Monitor metrics for performance evaluation

---

Evaluation & Visualization

To qualitatively assess model performance, the notebook visualizes results by overlaying predicted masks on MRI images.

Visualization steps:

1. Original MRI image
2. Original (ground truth) mask
3. MRI with mask overlaid in red
4. Predicted AI mask
5. MRI with predicted mask overlaid in green

plt.imshow(image)
plt.imshow(pred_mask, alpha=0.5, cmap='Greens')  # overlay predicted mask
plt.title("Predicted Mask Overlay")
plt.show()

This helps in assessing whether the model correctly identifies tumor regions.

---

Usage

1. Clone this repository:

git clone https://github.com/your-username/brain-tumor-segmentation.git
cd brain-tumor-segmentation

2. Install dependencies:

pip install -r requirements.txt

3. Run the Jupyter Notebook:

jupyter notebook Untitled2.ipynb

4. Modify paths in data_mask.csv if needed and start training or evaluating the model.

---

References

• U-Net: Convolutional Networks for Biomedical Image Segmentation
• Kaggle Brain MRI Images for Brain Tumor Detection
• TensorFlow/Keras Documentation

---

This README provides a comprehensive overview of the project, making it clear for collaborators, recruiters, or anyone exploring your GitHub repository.

---