First, the project uses a neural network model to classify the brain MRI images into these four categories :
- Glioma
- Pituitary
- Meningioma
- No Tumor
Then I started working on the No Tumor dataset to see whether certain rare tumor cases were erringly classified as having no tumor. I used a Generative Adversarial Network (GAN) model to augment more images to mitigate bias as much as possible and then checked those images for certain characteristics that might signal the presence of a rare case of the tumor. In the end, a binary classification model (here CNN) was trained with the help of a labeled dataset ( No Tumor / Rare Case ) and then tested on an unseen chunk of the same dataset.
Shield:
This work is licensed under a Creative Commons Attribution-NonCommercial-ShareAlike 4.0 International License.
