Brain Tumor Segmentation

3D Unet Architecture

Dataset

BraTS2020 Dataset

https://www.kaggle.com/datasets/awsaf49/brats20-dataset-training-validation

Results :-

Version 3

DiceCoef	IOU	Recall	Precision
0.94	0.89	0.98	0.98

Version 2

DiceCoef	IOU	Recall	Precision
0.88	0.79	0.97	0.97

Version 1

DiceCoef	IOU	Recall	Precision
0.80	0.68	0.78	0.82

Lessons Learned

After analyzing various results from different version of the model, I should have used a Weighted Loss function as the dataset contain less number of samples from classes 2 and 3. This lead to high IOU score but the model preform worst for predicting classes 2 and 3

Reference

1. Ozg ̈un C ̧ i ̧cek, Ahmed Abdulkadir, Soeren S. Lienkamp, Thomas Brox, and Olaf Ronneberger (2016) 3D U-Net: Learning Dense Volumetric Segmentation from Sparse Annotation Google DeepMind, London, UK, Computer Science Department, University of Freiburg, Germany

2. 3D-UNet Medical Image Segmentation for TensorFlow NVIDIA