Copula linked parallel ICA jointly estimates linked structural and functional MRI brain networks

Oktay Agcaoglu, Deniz Alaçam, Tülay Adali, Vince D. Calhoun, Rogers F. Silva, Sergey M. Plis, Biozid Bostami

Abstract: Different brain imaging methods provide valuable insights, and their combination enhances understanding of the brain. Existing fusion approaches typically use precomputed functional magnetic resonance imaging (fMRI) features, such as amplitude of low frequency fluctuations, regional homogeneity, or functional network connectivity while linking fMRI and structural MRI (sMRI). The fusion step typically ignores the detailed temporal information available in the complete 4D fMRI. Motivated by prior work showing covarying sMRI networks resemble resting fMRI networks, we introduce a new technique called copula linked parallel ICA (CLiP-ICA). This innovative method simultaneously estimates independent sources and an unmixing matrix for each modality while also linking spatial sources through a copula model. We tested the effectiveness of CLiP-ICA in both a simulation and a real-data using fMRI and sMRI data from an Alzheimer study. Results showed significant linkage in several domains including cerebellum, sensorimotor and default mode. In sum, we provide an approach to simultaneously estimate and link independent components of fMRI and sMRI while preserving temporal information.