Mounting evidence shows that many neuro-disorders can be understood as a dysfunction syndrome where the structural and functional connectivities of the large-scale network are progressively disrupted. In this regard, the subject-specific longitudinal pattern of network alterations is often more putative than other biomarkers discovered from cross-sectional network data. However, current computational approaches for brain network analysis are mainly designed for cross-sectional data, lacking the capability to accurately characterize the longitudinal brain network changes. To address this limitation, we propose to jointly estimate the brain networks at all time points by hierarchically regularizing temporal consistency in a local to a global manner, where we not only penalize the neurologically unreasonable longitudinal changes at each connection but also encourage maintaining the network geometry along time. By integrating this domain knowledge, we present a group-wise graph learning approach to construct longitudinal networks, which allows us to enhance the statistical power and replicability in network analyses. We present promising group comparison results in a longitudinal study of Alzheimer’s disease using structural neuroimaging data from the ADNI database, compared with the conventional cross-sectional network analysis approach.
CITATION STYLE
Turja, M. A., Zsembik, L. C. P., Wu, G., & Styner, M. (2019). Constructing Consistent Longitudinal Brain Networks by Group-Wise Graph Learning. In Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics) (Vol. 11766 LNCS, pp. 654–662). Springer Science and Business Media Deutschland GmbH. https://doi.org/10.1007/978-3-030-32248-9_73
Mendeley helps you to discover research relevant for your work.