A UNet for the analysis of perfusion CT imaging in the setting of acute ischemic stroke.
Please cite as: Klug, J. et al. Bayesian Skip Net: Building on Prior Information for the Prediction and Segmentation of Stroke Lesions. in Brainlesion: Glioma, Multiple Sclerosis, Stroke and Traumatic Brain Injuries (eds. Crimi, A. & Bakas, S.) 168–180 (Springer International Publishing, 2021). doi:10.1007/978-3-030-72084-1_16.
Further work: BayesianSkipNet
pip install -r requirements.txt
- Environment must use python 3.7 (for torch and CUDA compatibility)
- The main file for training can be found under
train_segmentation.py. It takes a config file as argument, examples can be found in the./configfolder. - A visdom server can launched as well for visualisation:
python -m visdom.server
- This is a fork of ozan-oktay/Attention-Gated-Networks
- "Attention U-Net: Learning Where to Look for the Pancreas", MIDL'18, Amsterdam, original paper
- Clinical implications: https://jamanetwork.com/journals/jamanetworkopen/fullarticle/2762679
- Downscaling to a 2.5 dimensional unet (eg. https://github.com/xyzacademic/multipathbmp), "A multi-path 2.5 dimensional convolutional neural network system for segmenting stroke lesions in brain MRI images", Xue et al, paper
- Multi-scale attention network: https://github.com/sinAshish/Multi-Scale-Attention