Related Experiment Video
Updated: Jun 26, 2026
10:17
Guidelines and Experience Using Imaging Biomarker Explorer (IBEX) for Radiomics
Published on: January 8, 2018
RiTex: Harmonization of Radiomic Features Based on Riemannian Geometry
Darya A Voitenko1,2, Anton V Vladzymyrskyy1, Olga V Omelyanskaya1
1State Budget-Funded Health Care Institution of the City of Moscow "Research and Practical Clinical Center for Diagnostics and Telemedicine Technologies of the Moscow Health Care Department", 127051 Moscow, Russia.
Journal of Imaging
|June 25, 2026
Summary
RiTex, a novel Riemannian texture harmonization framework, effectively reduces batch effects in radiomics data. This method significantly improves model generalizability across multicentre studies without compromising biological data integrity.
Area of Science:
- Medical Imaging
- Radiomics
- Biostatistics
Background:
- Batch effects from hardware and protocol variations challenge radiomics model generalizability.
- Existing harmonization methods like ComBat and GANs struggle with high-dimensional radiomic data.
Purpose of the Study:
- Introduce RiTex, a new framework for Riemannian texture harmonization in radiomics.
- Evaluate RiTex's effectiveness in reducing batch effects and improving multicentre model generalizability.
Main Methods:
- RiTex solves a generalized eigenvalue problem using class-aware biological scatter and Ledoit-Wolf-regularized covariances.
- The SPD-manifold Fréchet mean is employed for principled averaging.
- Evaluated on the radMLBench (50 datasets) and a head-and-neck benchmark (4 centers, 380 patients).
Main Results:
- RiTex reduced batch auto-detection AUC in 96% of radMLBench datasets (mean reduction ΔBatch = -0.365).
- No significant degradation in biological AUC was observed (mean ΔBio = +0.018).
- On the H&N benchmark, RiTex reduced Batch AUC from 0.74 to 0.59, outperforming ComBat and CovBat (≈0.98).
Conclusions:
- RiTex offers a robust solution for harmonizing radiomic data, enhancing multicentre study generalizability.
- The generalized eigenvalue decomposition (GEVD) and Ledoit-Wolf shrinkage are key performance drivers.
- RiTex effectively mitigates batch effects while preserving biological information.