nestedcombat-harmonization
Description
This software performs feature-level harmonization of radiomic data extracted from medical images. Its primary goal is to reduce variability introduced by differences in acquisition protocols, scanner types, and imaging centers. The tool is based on the ComBat methodology, which uses an empirical Bayes linear model to correct batch effects by estimating location and scale parameters. It offers two harmonization strategies: the ComBat method, which aligns radiomic features to a common scale based on either all centers (ComBat) or a user-defined batch grouping (M-ComBat), and the Nested ComBat method, which enables simultaneous correction for multiple batch effects in hierarchical settings. This tool is intended to improve the robustness and comparability of radiomic features in multi-center studies.
Usage
You can run the training or inference based on the configuration file provided in your input directory. The application accepts 2 positional parameters:
- The input path of a dataset directory (containing your data and config file).
- The output directory where the harmonization results will be saved.
Basic Command:
jobman submit -i nestedcombat-harmonization -- <INPUT_DIR> <OUTPUT_DIR>
Example for Training:
jobman submit -i nestedcombat-harmonization -- ~/datasets/87f3be56-4725-45c3-9baa-d338de530f73/train_data/ ~/persistent-home/results/nestedcombat_train/
Example for Testing/Inference:
jobman submit -i nestedcombat-harmonization -- ~/datasets/87f3be56-4725-45c3-9baa-d338de530f73/test_data/ ~/persistent-home/results/nestedcombat_test/
Note: The output directory path should be within the persistent-home (~/persistent-home/), which is shared between all desktops and jobs created by the user; otherwise, the results will be lost after the end of the job.
Authors
Benito Farina
BIT-UPM
Contact info
benito.farina@upm.es
mj.ledesma@upm.es
https://github.com/BenitoFar/NestedComBat/issues
URL
Public dockerfile repository:
https://harbor.eucaim.cancerimage.eu/harbor/projects/4/repositories/nestedcombat_harmonization/info-tab
License
https://github.com/EUCAIM/upv-node-workstation-images/blob/87a57cff6b79d8c85dc65a3f01a27e3f967d22a4/LICENSE
Visibility of application
Public
nestedcombat-harmonization
Description
This software performs feature-level harmonization of radiomic data extracted from medical images. Its primary goal is to reduce variability introduced by differences in acquisition protocols, scanner types, and imaging centers. The tool is based on the ComBat methodology, which uses an empirical Bayes linear model to correct batch effects by estimating location and scale parameters. It offers two harmonization strategies: the ComBat method, which aligns radiomic features to a common scale based on either all centers (ComBat) or a user-defined batch grouping (M-ComBat), and the Nested ComBat method, which enables simultaneous correction for multiple batch effects in hierarchical settings. This tool is intended to improve the robustness and comparability of radiomic features in multi-center studies.
Usage
You can run the training or inference based on the configuration file provided in your input directory. The application accepts 2 positional parameters:
Basic Command:
jobman submit -i nestedcombat-harmonization -- <INPUT_DIR> <OUTPUT_DIR>Example for Training:
Example for Testing/Inference:
Note: The output directory path should be within the persistent-home (
~/persistent-home/), which is shared between all desktops and jobs created by the user; otherwise, the results will be lost after the end of the job.Authors
Benito Farina
BIT-UPM
Contact info
benito.farina@upm.es
mj.ledesma@upm.es
https://github.com/BenitoFar/NestedComBat/issues
URL
Public dockerfile repository:
https://harbor.eucaim.cancerimage.eu/harbor/projects/4/repositories/nestedcombat_harmonization/info-tab
License
https://github.com/EUCAIM/upv-node-workstation-images/blob/87a57cff6b79d8c85dc65a3f01a27e3f967d22a4/LICENSE
Visibility of application
Public