PennSIVE Virtual Seminar Series

A structured multivariate approach for removal of latent batch effectsAbstract: Neuroimaging data collected across different scanners or sites are often heterogeneous in which a good harmonization is necessary to achieve high statistical power. While a few relevant approaches have been proposed in neuroimaging literature, most use a univariate approach and rely on assumptions on covariates. These may suffer from the loss of statistical power by not capturing latent patterns specific to sites or scanners. In this talk, we introduce a new multivariate harmonization method based on a penalized objective function with structured nuclear norm penalization. The proposed method decomposes a data matrix into (i) a low-rank structure shared across all scanners/sites, (ii) a low-rank structure capturing scanner-/site-specific variations, and (iii) a full-rank noise matrix, and then corrects for both (latent) additive and multiplicative batch effects. Using simulation studies, we show that the proposed method is superior to existing methods. We then apply our method to multi-modal neuroimaging data from the Social Processes Initiative in Neurobiology of the Schizophrenia (SPINS) study to highlight our findings.Bio: Jun Young Park is an Assistant Professor appointed by the Department of Statistical Sciences and the Department of Psychology. He received his Ph.D. in Biostatistics at the University of Minnesota -Twin Cities in 2020 and B.A. in Mathematics/Statistics at Carleton College in 2012. Jun Young Park is an applied statistician primarily working on the field of neuroimaging, and hopes to provide new insights to better understand the human brain. While understanding the mechanism of the human brain and its relationship with cognition and neurodegenerative diseases is an active area of research, the neuroimaging data is considered "big" and complex to statisticians. As a statistician, he tries to develop statistical models and inference procedures to account for several layers of correlation structures, including spatial, temporal, and functional dependencies