This repository contains an end-to-end pipeline for 3D brain cell segmentation using multi-parametric MRI volumes. It is based on the well-known BraTS (Brain Tumor Segmentation) challenge data and implements a 3D U-Net to segment glioma sub-regions (edema, enhancing tumor, and necrotic core) from stacked MRI modalities. The included Jupyter notebook demonstrates the following steps:
-
3D U-Net Model: A convolutional neural network (CNN) architecture for semantic segmentation in 3D images.
-
BraTS Dataset: Multi-modal MRI data, including FLAIR, T1c, T2, and segmented labels.
-
Segmentation: Automatic segmentation of tumor sub-regions, including edema, necrotic core, and enhancing tumor.
-
Custom Loss Functions: Combination of Dice loss and categorical focal loss for improved training performance.
-
TensorFlow: A machine learning framework for building and training deep learning models.
-
Keras: High-level neural networks API, running on top of TensorFlow.
-
NumPy: For numerical operations, including data manipulation.
-
Scikit-learn: For preprocessing and metric calculations.
-
Matplotlib: For plotting and visualizing training results and outputs.
-
Nibabel: For reading and processing MRI data formats such as
.nii.gz. -
Segmentation Models 3D: For 3D segmentation loss functions and metrics.
- Python 3.7+
- Libraries: numpy, nibabel, tensorflow, keras, matplotlib, tifffile, scikit-learn, splitfolders
- Download BraTS 2021 dataset to
C:\Users\20101309\Desktop\ASNR-MICCAI-BraTS2023-GLI-Challenge-TrainingData\. - Preprocess data to 128x128x128 format in
E:\BRATS 2021\input_data_128\.
Run the notebook Cancer_cell_segmentation.ipynb in Jupyter.
- Loads and normalizes images/masks.
- Predicts on validation images (e.g., 402, 34, 247).
- Saves predictions to
C:\Users\20101309\Desktop\outputs\.
Example prediction plot:
Predictions for slices shown in notebook. Model evaluates on validation set.
MIT
Did this project in Spring 2024 for my CSE438 course with my teammates : Md Ashiq Ul Islam Sajid & Rumaysa Mumtahana