This is the repository for OPNested, OPNested+GMM, and OPNested-ComBat from the manuscript entitled "Improved generalized ComBat methods for harmonization of radiomic features" by Horng et al., published at Nature Scientific Reports. These methods are updates to the algorithms previously published at generalized-combat. All functions needed to implement the methods are in the OPNestedComBat.py file.

Enables harmonization by multiple imaging parameters. See file comments for more details.

Uses Gaussian Mixture Modeling (GMM) to identify scan groupings associated with hidden covariates to better address bimodal feature distributions. OPNested+GMM treats the GMM grouping as a batch variable for harmonization.

Uses Gaussian Mixture Modeling (GMM) to identify scan groupings associated with hidden covariates to better address bimodal feature distributions. OPNested-GMM treats the GMM grouping as a clinical covariate for protection during harmonization.

To illustrate how to use the function file, an example with CAPTK features extracted from the publicly available NSCLC-Radiogenomics dataset has been added.

These methods are still a work in progress! Keep an eye out for future updates to the methods.