Multivariate testing and effect size measures for batch effect evaluation in radiomic features.

Area of Science:

• Medical Imaging Analysis
• Biostatistics
• Radiomics

Background:

• Radiomics analysis holds promise for precision medicine, but image acquisition variability introduces batch effects that compromise reproducibility.
• Current methods for evaluating batch effects in radiomics are inconsistent, hindering reliable downstream predictive analyses.

Purpose of the Study:

• To introduce and evaluate PERMANOVA and RESI as robust statistical tools for quantifying batch effects in radiomics data.
• To compare the performance of PERMANOVA and RESI against standard univariate statistical testing for batch effect assessment.

Main Methods:

• Utilized the multivariate statistical test PERMANOVA and the Robust Effect Size Index (RESI).
• Evaluated methods using simulated radiomics features and real radiomics features from full-field digital mammography (FFDM) data.
• Compared PERMANOVA's power against univariate statistical tests and RESI's interpretability at large sample sizes.

Main Results:

• PERMANOVA demonstrated superior statistical power compared to standard univariate tests for detecting batch effects.
• RESI effectively quantified the effect size of site-specific variations, even with very large datasets.
• Both methods proved valuable in characterizing batch effects in radiomics features.

Conclusions:

• PERMANOVA and RESI offer more powerful and interpretable approaches for detecting and quantifying batch effects in radiomics studies.
• These methods can improve the reproducibility and reliability of radiomics analyses for precision medicine.
• Enhanced batch effect assessment is crucial for advancing the clinical application of radiomics.