Skip to content

Smart India Hackathon 2019 project given by the Department of Atomic Energy

Notifications You must be signed in to change notification settings

ShubhayanS/Brain-MRI-Segmentation

 
 

Repository files navigation

Brain MRI Segmentation

The method we use comes from this paper: From neonatal to adult brain mr image segmentation in a few seconds using 3d-like fully convolutional network and transfer learning

Soft tissue segmentation.

This project was part of the Smart India Hackathon 2019 in which our team was the runner ups. The problem statement was Brain MRI Segmentation using Machine Learning given by Department of Atomic Energy, Government of India

DAE

Preprocessing

preprocess

architecture

VGG16

Why our model?

• Since we are using transfer learning, a novel approach in this field, so we do not need to train our model from scratch which makes it very fast in training in comparison to other models.

• Stacking 3 successive 2D slices allows us to make a RGB image, another novel idea.This representation enables us to incorporate some 3D information, while avoiding the expensive computational and memory requirements of fully 3D FCN.

• Using Transfer Learning we do not need many training images, so we could train our model very well only on a few training images.

• We are also using traditional data augmentation methods like rotating, cropping and flipping the images in training set for improving our model.

GUI and giving input

GUI

Output

gui

Tumour prediction

tumour

Other regions

Contributors

Vaibhav Shukla
Abhijeet Singh
Omkar Ajnadkar
Govind Singh Rajpurohit
Ratna Priya
Sanath Singavarapu

About

Smart India Hackathon 2019 project given by the Department of Atomic Energy

Resources

Stars

Watchers

Forks

Releases

No releases published

Packages

No packages published

Languages

  • Python 100.0%