Brain MRI Tumor Segmentation

This project implements a deep learning pipeline for detecting and segmenting brain tumors in MRI images.

Overview

The system uses a two-stage approach:

1. A classification model to detect the presence of tumors
2. A segmentation model to localize and outline tumors when detected

Key components:

• Data preprocessing and visualization
• ResNet50-based classification model
• ResUNet segmentation model
• Custom loss functions (Focal Tversky loss)
• Evaluation metrics and result visualization

Dataset

The project uses the LGG MRI Segmentation dataset, containing brain MRI images and corresponding tumor mask annotations.

Models

Classification Model

• Based on ResNet50 architecture
• Detects presence/absence of tumors

Segmentation Model

• Custom ResUNet architecture
• Generates pixel-wise tumor masks

Usage

The main notebook BrainMRIseg.ipynb contains the full pipeline:

1. Data loading and preprocessing
2. Model training (classification and segmentation)
3. Evaluation and visualization of results

Results

• Classification accuracy: 93.75%
• Segmentation Tversky score: 85.10%

Combining classification and segmentation can potentially reduce false positives and provide more detailed information about tumor presence and location.

Sample segmentation results are visualized in the notebook.

Requirements

• TensorFlow 2.x
• Keras
• OpenCV
• scikit-image
• pandas
• matplotlib

Future Work

• Experiment with other architectures
• Implement additional data augmentation
• Explore multi-class segmentation for different tumor types