This repository implement ComBat and AutoComBat describe in Carré, et al. (2022).
If you face any problem, please feel free to open an issue.
Current harmonization/normalization methods such as ComBat use a Bayes parametric empirical framework to robustly adjust the data to site / scanner effects. This method requires a representative statistical sample and is therefore not suitable for radiomics machine learning models for clinical translation, where the emphasis is on evaluating individual scans from previously unseen scanners. In addition, it may not always be obvious to define a batch effect that would be linked to the site or scanners, as a change in a machine parameter may be more appropriate for another scanner type or site. AutoComBat has thus been implemented, and it allows to associate a sample to a given site / scanner by a clustering method. Thus, the site/scanner can be defined by dicom tags defining the machine (i.e. magnetic field, TI, TR …) or metrics of image quality.
This repository has been coded to be compatible with scikit-learn and thus facilitate machine learning projects.
ImageComBat is under development and allows to normalize the image directly (using Combat or AutoCombat) based on neuroHarmonize.
1. Create a conda environment (recommended)
ENVNAME="ComScan" conda create -n $ENVNAME python==3.7.7 -y conda activate $ENVNAME
pip install git+https://github.com/Alxaline/ComScan.git
git clone https://github.com/Alxaline/ComScan.git cd ComScan pip install -e .
https://comscan.readthedocs.io/en/latest/
If you find this repository useful for your research, please cite our work:
Carré, A., Battistella, E., Niyoteka, S. et al. AutoComBat: a generic method for harmonizing MRI-based radiomic features. Sci Rep 12, 12762 (2022). https://doi.org/10.1038/s41598-022-16609-1
BibTeX:
@article{carreAutoComBatGenericMethod2022, title = {AutoComBat: a generic method for harmonizing MRI-based radiomic features}, volume = {12}, issn = {2045-2322}, url = {https://www.nature.com/articles/s41598-022-16609-1}, doi = {10.1038/s41598-022-16609-1}, language = {en}, number = {1}, urldate = {2022-07-27}, journal = {Scientific Reports}, author = {Carré, Alexandre and Battistella, Enzo and Niyoteka, Stephane and Sun, Roger and Deutsch, Eric and Robert, Charlotte}, year = {2022}, keywords = {Cancer imaging, Computational science, Tumour biomarkers}, pages = {12762}}
Based on: ComBatHarmonization