Two-Stage Ensemble Scar Segmentation for the Left Atrium proposed for the LAScarQS 2022: Left Atrial and Scar Quantification & Segmentation Challenge in conjunction with STACOM and MICCAI 2022 (Sep 18th, 2022, Singapore).
TESSLA Paper Preprint: DOI: 10.1007/978-3-031-31778-1_10
- Task001_Blood -> Segmentation of left atrium blood pool from LGE-MRI
- Task002_Scar -> Segmentation of left atrium scars from LGE-MRI
- Task003_Scar -> Segmentation of left atrium scars from IIR wall
Docker image implementation supports both GPU and CPU modes. We recommend running Docker image with GPU as it takes approximately 6 min per subject on our machine (NVIDIA GeForce RTX 3060 and 16GB RAM) versus 2h per subject in CPU mode (12th Gen Intel(R) Core(TM) i7-12700KF 3.61 GHz).
To run GPU version of the code have NVIDIA Container Toolkit installed to the host machine and add '--gpus all' flag to docker run command, here is an example command:
docker run --gpus all -v <our_test_directory>:/input:ro -v :/output -it ai4af/tessla:latest
Our docker image does not require running additional commands.
Navigate to TESSLA root directory and run the following commands:
- Create a Docker image:
make create_image - Run the Docker image:
make run_image
The predicted left atrium bloodpool and scar segmentations are saved in Docker container /output folder, so the folder should be copied to the host machine at the end of the process.