Brain MRI Tumour Segmentation

Course Project - Medical Image Computing

Instructor - Prof. Suyash Awate, Dept of Computer Science, IIT Bombay

- Trained a U-Net architecture for segmentation of the tumor region using the soft dice loss in MRI slices and used a dataset consisting of around 4K MRI slices from 110 patients from The Cancer Imaging Archive
- Achieved a mean IOU of 0.77 on the validation set between the predicted and the true segmentation masks of tumor positive MRI slices and an accuracy of 96.8% on the binary classification of presence of tumor

[Presentation] [code]