A0750
Title: Biclustering multivariate longitudinal data in sport-related concussion recovery
Authors: Jaroslaw Harezlak - Indiana University School of Public Health-Bloomington (United States) [presenting]
Luo Xiao - North Carolina State University (United States)
Yu-Chien Wu - Indiana University (United States)
Qiuting Wen - Indiana University (United States)
Caleb Weaver - Apple (United States)
Abstract: Biclustering involves the simultaneous clustering of both samples and features within a dataset, enabling the identification of subsets of samples that exhibit similar behavior across specific subsets of features. Motivated by a longitudinal diffusion tensor imaging (DTI) study of sport-related concussion (SRC), the problem of biclustering multivariate longitudinal data is addressed. In this context, subjects and features are grouped based on shared longitudinal patterns rather than absolute magnitudes. A penalized regression-based method is proposed that leverages heterogeneity in these temporal patterns across subjects and features. The effectiveness of the approach is demonstrated through a simulation study and an application to the motivating DTI dataset. The analysis uncovers distinct subgroups of SRC cases characterized by heterogeneous patterns of white matter abnormalities.
