Multi-Task Active Learning with Uncertainty Weighted Loss for Coronary Calcium Scoring

This repository is the official implementation of Multi-Task Active Learning with Uncertainty Weighted Loss for Coronary Calcium Scoring.

Requirements

To install requirements:

pip install -r requirements.txt

Prediction

To predict the model in the paper, copy all non-contrast enhanced cardiac CT (.mhd file format) into the data folder. To run the script, you can run the cacs.sh script or run the python script in the terminal directly.

python cacs_predict.py -m <path_to_pretrained_model> -d <path_to_data_folder> -p <path_to_prediction_folder> -gpu cuda

Example prediction

Pre-trained Models

You can find the pretrained model in the model folder.

Training

Since the training set and corresponding annottaions are not public, a training script is not provided.

Contributing

Bernhard Föllmer
Charité - Universitätsmedizin Berlin
Klinik für Radiologie
Campus Charité Mitte (CCM)
Charitéplatz 1
10117 Berlin
E-Mail: bernhard.foellmer@charite.de
Tel: +49 30 450 527365
http://www.charite.de