Brain tumor auto-segmentation using Magnetic Resonance Imaging (MRI), This project demonstrates data preprocessing, 3D sub-volume extraction, MRI standardization and deep learning–based segmentation with Dice coefficient and soft Dice loss. Includes evaluation metrics (sensitivity, specificity) for tumor detection performance.
python · deep-learning · tensorflow · keras · medical-imaging · mri · brain-tumor · segmentation · computer-vision · healthcare-ai
Deep learning pipeline for brain tumor auto-segmentation using MRI scans — includes preprocessing, Dice loss, and evaluation metrics.
Source: Stanford AI for Healthcare
The project demonstrates how to:
- Preprocess MRI scans
- Extract 3D sub-volumes from brain scans
- Standardize MRI image data
- Implement dice coefficient and soft dice loss for segmentation evaluation
- Train a neural network for brain tumor segmentation
- Compute sensitivity and specificity metrics for performance evaluation
- Brain_Tumor_Segmentation.py: Main Python assignment file with completed functions for sub-volume extraction, standardization, dice coefficient, loss function, and evaluation metrics.
- README.md: This file with information about the repository.
get_sub_volume: Extracts sub-volumes from MRI scans for training.standardize: Normalizes each channel and z-slice of MRI images.single_class_dice_coefficient: Dice metric for a single class.dice_coefficient: Mean dice coefficient across classes.soft_dice_loss: Loss function for segmentation tasks.compute_class_sens_spec: Sensitivity and specificity computation for segmentation results.
Run the Python file in Jupyter Notebook or Python environment to preprocess MRI data and evaluate segmentation models.
This project is inspired by deep learning applications in medical imaging, particularly brain tumor segmentation tasks in MRI data.