Answers are grounded in this paper and cite their sources.
Comparison of ComBat and GLM regression for data harmonization in rhabdomyosarcoma, revealing distinct survival correlations.
Motivation: Rhabdomyosarcoma is a rare pediatric tumor without known biomarkers for early therapy adjustment. Their investigation requires multi-center patient cohorts and, consequently, data harmonization. Goal(s): Comparison of two harmonization approaches ("ComBat", "generalized linear models (GLM) regression") for evaluating potential differences in performance and outcome. Approach: Tumorous and healthy tissues were segmented. Healthy and tumorous ADC features were harmonized with ComBat, the regression based GLM - on healthy features and applied to tumors. Tumor histology, body area, center information were provided as covariates. Results: In head-and-neck tumors, a stronger correlation between ADC change and survival was observed after harmonization, but not in other tumor locations. Impact: The results suggest that harmonization should be applied to separate groups based on tumor location, thus, allowing the algorithm to focus on inter-center variability. The need for more extensive delineations of healthy tissues is suggested for more robust results.
No takes yet. Share an insight, caveat, or question.
Zhylka et al. (2025) studied this question.
Synapse has enriched 10 related papers. Consider them for comparative context: