Brain Wave Detect

Project Logo

Inspiration

During one of my visits to a hospital, I came across a lady who shared her experience with brain tumor. After a brief conversation, I realized how important it is to have an early diagnosis in most cases. Her near-death experience and the troubles her family had to face, both financially and mentally, provided me with an incentive to work on a project of this nature. I believe I can quote it as an inspiration.

What it does

As suggested by its name, our project classifies MRI images of brain to detect three most common kinds of brain tumor. It also informs the users briefly on those and ways to follow up in case of positive diagnosis. It is an assisting tool for medical personnels.

How we built it

We used the Brain Tumor Detection Classification dataset from Kaggle. There are 4370 training images and 874 validation images. We built a CNN architecture which consists of Convolution layers, Maxpooling, Batch normalization, dropout layers and Dense layers. There are 3 convolution layers, 3 Maxpooling and 3 batch normalizations. We have also performed data augmentation for improving generalization. After training and testing the model, we integrated our model with an interactive web interface built using Streamlit.

Challenges we ran into

Fortunately, we got relatively easy access to datasets, but the training procedure was rigorous. First, we worked on identifying the tumor in the scan. This was followed by training for classification of tumors into either of Glioma, Meningioma or Pituitary Adenoma. Both were carried on different training sets. This resulted in some conflicts initially but was resolved gradually with time. It was also our first time using Streamlit, which had limited resources to learn from. This also resulted in slight inconvenience while learning.

Accomplishments that we're proud of

First and foremost, our project is entirely complete. NO BUGS, NO ERROR. All the features we have included work completely fine. Second, all of this was completely new to us, and achieving this feat is in itself a victory for us.

What we learned

Streamlit
Machine Learning
Image Segmentation
Image Classification
Tumor Anatomy ( Slight introduction to radio-pathology)
Working collaboratively

What's next for Brain Wave Detect

The second phase for this project would be to introduce image segmentation. This would add features like Tumor Location, Size Estimation, Severity, Age of tumor, etc. This would provide it with higher usability and reliability. The third phase would include integration in the MRI scanner machine and Real time data processing to medical personnel and Hospital management.