Meeting Banner
Abstract #2835

ComBat Empirical Bayes Model Supersedes Naive Methods for Statistical Harmonization of Multi-Site 1H MR Spectroscopy

Lasya P Sreepada1,2, Sam H Jiang1, Huijun Vicky Liao1, Katherine M Breedlove1, Eduardo Coello1, and Alexander P Lin1
1Radiology, Center for Clinical Spectroscopy, Brigham and Women's Hospital and Harvard Medical School, Boston, MA, United States, 2Radiology, Center for Biomedical Image Computing and Analytics, Perelman School of Medicine, University of Pennsylvania, Philadelphia, PA, United States

As medical imaging enters the information era, there is a rapidly increasing need for big data analytics. Robust pooling and harmonization of multi-site data across diverse cohorts is critical. We compare the performances of current basic methods with ComBat, an Empirical Bayes method that removes batch effects, in harmonizing 1H Brain MR Spectroscopy (MRS) of healthy controls from 4 sites. Basic harmonization did not bring metabolite means closer together and increased variance, while ComBat successfully removed significant site effects as determined by ANOVA and Levene's tests. These results may improve reproducibility and generalizability of MRS studies, especially in clinical space.

This abstract and the presentation materials are available to members only; a login is required.

Join Here