Segmentation Transformer - Medical Image Segmentation with Neural Networks This is an early-stage research project focusing on segmentation of tissue in medical images using lightweight transformers. This project aims to provide accurate and reliable segmentation, aiding in the diagnosis and monitoring of breast cancer.
Segmentation masks for the DUKE dataset are saved at : data/Duke
Create a virtual environment and install the required packages: Segmentation Transformer - Medical Image Segmentation with Neural Networks This is an early-stage research project focusing on segmentation of tissue in medical images using lightweight transformers. This project aims to provide accurate and reliable segmentation, aiding in the diagnosis and monitoring of breast cancer.
Segmentation masks for the DUKE dataset are saved at : data/Duke
Create a virtual environment and install the required packages:
- Navigate to mammoview/data/dataset_breast.py and create a new Dataset class based on BaseDataset.
- Go to mammoview/data/datamodule_breast.py and create a new DataModule class based on BaseDataModule.
- Optional: Pretrain the model using scripts/main_pretrain.py
- Train the model using scripts/main_train.py
python -m venv venv
source venv/bin/activate
pip install -e .- For cross-fold prediction, use scripts/main_predict_kfold.py
- For ensemble prediction on an external dataset, use scripts/main_predict.py
Please note that this project is in its early research stage. The project is separate from any previous projects and does not bear any relation to them.
The comparison of transformer models for segmentation