Brain-MRI-Segmentation

Problem

The task is classify the image i.e. have tumor or not.Then if there is a tumor, segment that part of tumor

Approach

For classification used the transfer learning model and for segmentation used the ResUNet Architecture

Results

Used callbacks to avoid overfitting.For classification task training acuracy achieved is 96.92% and the testing accuracy achieved is 93.44% and defined the custom loss function i.e. Tversky Loss and achieved 94.59% training tversky accuracy and 86.26% testing tversky accuracy.

Classification Scores

Segmentation Scores

Flask Demo

Made a flask app for testing the model in google colab that can detect the tumor and then stores the data to a csv file.Below is the demo of the web app

Predictions

Here are the below predictions for 3 images