Brain tumors are among the deadliest diseases worldwide, with gliomas being particularly prevalent and challenging to diagnose. Traditionally, physicians and radiologists rely on MRI and CT scans to identify and assess these tumors. However, this diagnostic process is not only time-consuming but also susceptible to human error, which can delay crucial treatment decisions.
To enhance diagnostic accuracy and efficiency in clinical settings, deep learning techniques are increasingly being integrated into medical imaging. Over the years, deep learning has demonstrated exceptional performance in analyzing complex medical images, providing reliable support to healthcare professionals. By leveraging large datasets, deep learning models can be trained to recognize patterns and anomalies in brain scans with reasonable accuracy, facilitating early detection and treatment of brain tumors.
This repository utilizes the BraTS 2021 and BraTS 2023 datasets to develop and evaluate both new and existing state-of-the-art algorithms for brain tumor segmentation. To facilitate research, we have made the code for training, evaluation, data loading, preprocessing, and model development open source. Researchers can use this template to build their models, enhancing accuracy and explainability.
All BraTS23 mpMRI scans are available as NIfTI files and include T2 Fluid Attenuated Inversion Recovery (FLAIR), native (T1), T2-weighted (T2), and post-contrast T1-weighted (T1Gd) images. These scans were acquired using different clinical protocols and various scanners from multiple institutions.
Annotations consist of GD-enhancing tumor (ET — label 3), peritumoral edematous/invaded tissue (ED — label 2), and necrotic tumor core (NCR — label 1). More details are available here. These subregions can be clustered into three more segmentation-friendly regions which are used to evaluate the segmentation performance, including enhanced tumor (ET), tumor core (TC) (joining ET and NCR), and whole tumor (WT) (joining ED to TC).
The dataset contains 1,251 patient cases labeled by expert radiologists. However, cases in the validation and test sets are not annotated. Therefore, the actual training set is divided to training, validation, and test sets. The training set contains 833, validation and test sets contains 209 patient records each for model evaluation. [9]
└── dataset
└── brats2023
├── train
│ ├── BraTS-GLI-00000-000
│ │ ├── BraTS-GLI-00000-000-seg.nii.gz
│ │ ├── BraTS-GLI-00000-000-t1c.nii.gz
│ │ ├── BraTS-GLI-00000-000-t1n.nii.gz
│ │ ├── BraTS-GLI-00000-000-t2f.nii.gz
│ │ └── BraTS-GLI-00000-000-t2w.nii.gz
│ └── ...
├── val
│ ├── BraTS-GLI-00006-000
│ │ ├── BraTS-GLI-00006-000-seg.nii.gz
│ │ ├── BraTS-GLI-00006-000-t1c.nii.gz
│ │ ├── BraTS-GLI-00006-000-t1n.nii.gz
│ │ ├── BraTS-GLI-00006-000-t2f.nii.gz
│ │ └── BraTS-GLI-00006-000-t2w.nii.gz
│ └── ...
└── test
├── BraTS-GLI-00009-000
│ ├── BraTS-GLI-00009-000-seg.nii.gz
│ ├── BraTS-GLI-00009-000-t1c.nii.gz
│ ├── BraTS-GLI-00009-000-t1n.nii.gz
│ ├── BraTS-GLI-00009-000-t2f.nii.gz
│ └── BraTS-GLI-00009-000-t2w.nii.gz
└── ...
image from Baid et al.
Anconda environment recommended.
git clone https://github.com/faizan1234567/Brain-Tumors-Segmentation
cd Brain-Tumors-Segmentation
create a virtual environment in Anaconda and activate it.
conda create -n brats_segmentation python=3.9.0 -y
conda activate brats_segmentation
Now install all the dependencies
pip install --upgrade pip
pip install -r requirements.txt
To train on BraTS 2023 or BraTS 2021 as both datasets are same except the naming convention is different, run the training command below:
python train.py -h
python train.py dataset.dataset_folder=<path to dataset> training.max_epochs=100 training.batch_size=1 training.val_every=1 training.learning_rate=1e-4 model.architecture=nn_former
To test the model on the tess set use the following command:
python test.py -h
python test.py test.weights=<path> dataset.dataset_folder=<path> test.batch=1 model.architecture=nn_former
To visualize, use:
python show.py -h
python show.py --type "get-gif"
Each model has been trained for 300 epochs on NVIDIA RTX-4070 GPU by exactly following details from their papers.
| Model Name | Mean Dice Score | Mean Hausdorff Distance | Mean Sensitivity | Mean Specificity |
|---|---|---|---|---|
| SegResNet | 0.8896 | 8.6502 | 0.9117 | 0.9932 |
| UNet | 0.5441 | 39.0898 | 0.7377 | 0.9911 |
| V-Net | 0.842 | 10.8914 | 0.8278 | 0.9927 |
| ResUNet++ | 0.7839 | 22.2491 | 0.784 | 0.9908 |
| AttentionUNet | 0.7982 | 20.0479 | 0.857 | 0.9915 |
| UNETR | 0.8705 | 9.9235 | 0.8902 | 0.9930 |
| SwinUNETR | 0.886 | 9.0157 | 0.9034 | 0.9940 |
| nnFormer | 0.8117 | 10.0703 | 0.8528 | 0.9923 |
| 3DUXNET | 0.8741 | 14.264 | 0.9244 | 0.9945 |
- MLOps tools integration such as Weight & Baises
- Multi-GPU training
- New Data augmentation options
- Added state-of-the-art options
- test code updated
- visualization code to be updated
If you find this project useful in your research, please consider cite and star the repository:
@misc{brats23-tumor-segmentation,
title={Multi-modal BraTS 2023 brain tumor segmentation},
author={Muhammad Faizan},
howpublished = {\url{https://github.com/faizan1234567/Brats-20-Tumors-segmentation}},
year={2023}
}[1]. MICCAI BRATS. (n.d.). MICCAI brats - the multimodal brain tumor segmentation challenge. MICCAI BRATS - The Multimodal Brain Tumor Segmentation Challenge. http://braintumorsegmentation.org/
[2]. monai. (n.d.). Home. MONAI. https://monai.io/
[4]. Chadha, A. (n.d.). Amanchadha/Coursera-ai-for-Medicine-specialization: Programming assignments, labs and quizzes from all courses in the Coursera AI for Medicine Specialization offered by deeplearning.ai. GitHub. https://github.com/amanchadha/coursera-ai-for-medicine-specialization
[5]. Zhou, C., Ding, C., Wang, X., Lu, Z., & Tao, D. (2020). One-pass multi-task networks with cross-task guided attention for brain tumor segmentation. IEEE Transactions on Image Processing, 29, 4516-4529.
[6]. Liu, Z., Lin, Y., Cao, Y., Hu, H., Wei, Y., Zhang, Z., ... & Guo, B. (2021). Swin transformer: Hierarchical vision transformer using shifted windows. In Proceedings of the IEEE/CVF international conference on computer vision (pp. 10012-10022).
[7]. He, K., Zhang, X., Ren, S., & Sun, J. (2016). Deep residual learning for image recognition. In Proceedings of the IEEE conference on computer vision and pattern recognition (pp. 770-778).
[8]. Ronneberger, O., Fischer, P., & Brox, T. (2015). U-net: Convolutional networks for biomedical image segmentation. In Medical image computing and computer-assisted intervention–MICCAI 2015: 18th international conference, Munich, Germany, October 5-9, 2015, proceedings, part III 18 (pp. 234-241). Springer International Publishing.
[9]. Lin, J., Lin, J., Lu, C., Chen, H., Lin, H., Zhao, B., ... & Han, C. (2023). CKD-TransBTS: clinical knowledge-driven hybrid transformer with modality-correlated cross-attention for brain tumor segmentation. IEEE transactions on medical imaging, 42(8), 2451-2461.
[10]. Myronenko, A. (2019). 3D MRI brain tumor segmentation using autoencoder regularization. In Brainlesion: Glioma, Multiple Sclerosis, Stroke and Traumatic Brain Injuries: 4th International Workshop, BrainLes 2018, Held in Conjunction with MICCAI 2018, Granada, Spain, September 16, 2018, Revised Selected Papers, Part II 4 (pp. 311-320). Springer International Publishing.
[11]. Ranzini, M., Fidon, L., Ourselin, S., Modat, M., & Vercauteren, T. (2021). MONAIfbs: MONAI-based fetal brain MRI deep learning segmentation. arXiv preprint arXiv:2103.13314.
[12]. Zhou, Z., Rahman Siddiquee, M. M., Tajbakhsh, N., & Liang, J. (2018). Unet++: A nested u-net architecture for medical image segmentation. In Deep Learning in Medical Image Analysis and Multimodal Learning for Clinical Decision Support: 4th International Workshop, DLMIA 2018, and 8th International Workshop, ML-CDS 2018, Held in Conjunction with MICCAI 2018, Granada, Spain, September 20, 2018, Proceedings 4 (pp. 3-11). Springer International Publishing.
[13].Hatamizadeh, A., Nath, V., Tang, Y., Yang, D., Roth, H. R., & Xu, D. (2021, September). Swin unetr: Swin transformers for semantic segmentation of brain tumors in mri images. In International MICCAI brainlesion workshop (pp. 272-284). Cham: Springer International Publishing.
[14]. Milletari, F., Navab, N., & Ahmadi, S. A. (2016, October). V-net: Fully convolutional neural networks for volumetric medical image segmentation. In 2016 fourth international conference on 3D vision (3DV) (pp. 565-571). Ieee