End-to-end Segmentation of Brain Tumor and Deployment

In this work, I develop a segmentation network using U-Net to segment the tumor from the input image. This pipeline is hosted on Heroku with continuous integration.

I use the dataset from Kaggle: https://www.kaggle.com/mateuszbuda/lgg-mri-segmentation

I have written the UNET code with inspirations from Aladdin's youtube video. (https://www.youtube.com/watch?v=IHq1t7NxS8k)

I trained this network on my Laptop for 10 epochs using AdamW Optimizer with learning rate of 1e-4.

Results

As we can see from the above results, within just 10 epochs, the model starts to generalize well.