Brain Tumor Segmentation with Attention U-Net

Overview

This repository contains a Convolutional Neural Network (CNN) model for brain tumor segmentation, utilizing the Attention U-Net architecture. The model has been trained and evaluated on the BraTS 2020 dataset, achieving an impressive accuracy of 98%.

Dataset

The BraTS 2020 dataset used for training and evaluation consists of high-resolution MRI scans with various modalities:
- T1-weighted
- T1-weighted post-contrast
- T2-weighted
- FLAIR

The dataset includes:
- Training Images: Annotated MRI scans of brain tumors.
- Validation Images: A separate set of MRI scans used for validation.
- Test Images: Used for evaluating the model's performance.

The dataset is publicly available and can be accessed [here](https://www.med.upenn.edu/sbia/brats2020/data.html).

Model

Attention U-Net

The model is based on the U-Net architecture, which is a popular deep learning model for biomedical image segmentation. The U-Net architecture features an encoder-decoder structure with skip connections that help preserve spatial information throughout the network. In this implementation, **Attention Gates** are integrated to enhance the standard U-Net model.

Features of Attention U-Net:

- Encoder-Decoder Architecture: Consists of a contracting path (encoder) that captures context and an expansive path (decoder) that enables precise localization. Skip connections between corresponding layers in the encoder and decoder preserve important features.
- Attention Gates: Enhance the model's ability to focus on relevant regions of the input images and suppress irrelevant information, leading to improved segmentation performance.
- Skip Connections: Direct connections between encoder and decoder layers help maintain spatial information and improve accuracy.

Model Performance

The Attention U-Net model has achieved a 98% accuracy on the BraTS 2020 dataset, demonstrating its effectiveness in accurately segmenting brain tumors.

Acknowledgments

- [BraTS 2020](https://www.med.upenn.edu/sbia/brats2020/data.html) for providing the dataset.
- [U-Net](https://arxiv.org/abs/1505.04597) and [Attention U-Net](https://arxiv.org/abs/1804.03999) for the base architecture and enhancements.