TopoVST: Towards Topology-fidelitous Vessel Skeleton Tracking

This work is the code base for paper TopoVST: Towards Topology-fidelitous Vessel Skeleton Tracking(Under Review).

By Yaoyu Liu, Minghui Zhang, Junjun He, Yun Gu

Institute of Medical Robotics, Shanghai Jiao Tong University, Shanghai, China.

Shanghai AI Lab, Shanghai, China.

A brief overview of our work. We utilize multi-scale GNNs for vessel skeleton direction and radius estimation.

Training and testing

Before training TopoVST, you need to generate training samples. Run:

python src/scripts/data/generate_samples.py

NOTE that you need to specify path to your own dataset.

Training and testing are driven by CSV config files. Each row is one run_id that pins dataset, tracker, hyper-parameters, and checkpoint paths:

python src/scripts/train/run_train.py --config-csv src/scripts/configs/train_configs.csv --run-id my_run --device cuda:0
python src/scripts/test/run_test.py  --config-csv src/scripts/configs/test_configs.csv  --run-id my_run --device cuda:0

Example CSV configs are coming soon.