Tumor Detection on Brain MRI using Detectron2

This project demonstrates the application of computer vision in healthcare by detecting tumors in brain MRI images. The project utilizes the Detectron2 framework, focusing on the COCO-Detection/retinanet baseline model, to achieve accurate tumor detection. Additionally, the project is presented as a web application using Streamlit for user-friendly interaction.

Table of Contents

• Detectron2 Overview
• COCO-Detection/retinanet Baseline
• Web Application
• Dataset
• Project Structure
• Installation
• Web Application

Detectron2 Overview

Detectron2 is a cutting-edge object detection framework developed by Facebook AI Research (FAIR). It simplifies the process of building, training, and deploying object detection models. By providing modular components and pre-implemented state-of-the-art algorithms, Detectron2 enables efficient experimentation and development in the field of computer vision.

Retinanet Baseline

The COCO-Detection/retinanet baseline is a specific object detection model architecture available within Detectron2. RetinaNet is a one-stage object detection model known for its efficient speed and competitive accuracy. It's especially suited for detecting objects of varying scales and sizes, making it a suitable choice for medical image analysis like tumor detection in brain MRI images. The COCO-Detection baseline further fine-tunes the RetinaNet architecture on the COCO dataset, which is a widely-used benchmark for object detection tasks.

Web Application

The user-friendly web application powered by Streamlit simplifies the utilization of the tumor detection model. Users can effortlessly upload their brain MRI images and instantly visualize the detected tumor regions.

Dataset

The dataset used in this project is sourced from Roboflow Universe. This dataset consists of axial brain MRI images annotated with tumor regions.

Project Structure

• train: This directory contains files for training the model. The trained model weights are saved in the /train/output/directory, eliminating the need for retraining. The project's training progress, such as loss and validation plots, can be found here as well. The selected model weights for deployment were taken from epoch number 3000 based on the provided training and validation plots.

Installation

To set up and run the Tumor Detection project locally, follow these steps:

Clone the repository:

git clone https://github.com/sevdaimany/Tumor-Detection-Brain-MRI
cd Tumor-Detection-Brain-MRI

Create a virtual environment (recommended):

python -m venv venv
source venv/bin/activate  # On Windows: venv\Scripts\activate

Install the required dependencies:

pip install -r requirements.txt

Download the dataset from the provided link and place the images in the appropriate data/train/imgs, data/train/anns, data/val/imgs and data/val/anns folders.

Download the trained model dataset from here and put it in train/output folder.

Web Application

The web application provides a user-friendly interface for tumor detection:

Launch the Streamlit app:

streamlit run main.py

Open your web browser and navigate to the provided URL.

Upload an MRI image using the app's interface.

View the uploaded image with highlighted tumor regions.