brain-tumor-segmentation

A UNet model for brain tumor segmentation. Pytorch version. A web app to visualize learning curves of pretrained models and make prediction with pretrained models.

visit web page

data source

video introduction

Here is a video Introduction about my project

Unet

Training Model

To train the model on your own computer:

1. Download project
2. Install dependencies by runing pip install -r requirements.txt in terminal
3. Train the model by runninng train.py:

A quick view of training process:

Test result of pretrained Model

Model	IOU score	Dice score
dice_loss + cross_entropy_loss, l_rate=0.0001	0.7465402	0.8548798
dice_loss, l_rate=0.0001	0.7388876	0.8498393
dice_loss + cross_entropy_loss, l_rate=0.001	0.7465402	0.8548789
dice_loss + cross_entropy_loss, l_rate=0.01	0.7522497	0.8586102
dice_loss + cross_entropy_loss, l_rate=0.1	0.7797701	0.8762593

Make Prediction

To start the web app on your own computer:

1. Download project
2. install dependencies by runing pip install -r requirements.txt in terminal
3. open the page by running streamlit run home.py in terminal.

A quick view of prediction page: