Medizoid

Predicted_Segmented-Image_epoch_50
Predicted_Segmented-Image_epoch_100
T2_Predicted_Epoch_30
T2_Predicted_Image_Epoch_10

Inspiration

We came across that one of the latest problems in the Medical Sector is related to MRIs and CTs. MRIs are very expensive and require high amounts of energy. Also, because of their powerful Magnetic Power, patients with any iron part implemented are at risk. The same goes with CT scans since their use of radioactivity is quite harmful to weak, young, and old people. We are trying to solve some of these issues with the use of Deep Learning.

What it does

One of the main aspects of our project is MRI Image Synthesis. In this, we convert T1-Weighted images into T2-Weighted images. Along with this, our other model helps by generating Segmented Brain MRI images to show the distribution of white matter.

How we built it

So we have created two models, one for Image synthesis and the other does Segmentation. Both works on generating images. We have implemented VAEs to achieve the purpose.

Challenges we ran into

Learning and understanding the terminologies involved with MRIs really took a while like Preprocessing MRI images requires quite a different methodologies than in general. We were learning PyTorch along with the implementation, and although it's quite pythonic in nature still took some time to wrap our minds around.

What we learned

We learned about issues faced and how to use deep learning to solve them. One of the interesting things we learned is the terminologies and concepts related to MRI image synthesis. Along with that we also learned PyTorch which was one of our main objectives.

What's next for Medizoid

We will be implementing Conditional Generative Adversarial networks (cGANs) to improve the accuracy of our model. Along with that, we will be creating a web application so that it could be of some help to society.