A0733
Title: Spatial semiparametric model applied to neuroscience
Authors: Jinxuan Bai - University of Southampton (United Kingdom) [presenting]
Ian Galea - University of Southampton (United Kingdom)
Aravinthan Varatharaj - University of Southampton (United Kingdom)
Chao Zheng - University of Southampton (United Kingdom)
Zudi Lu - City University of Hong Kong (China)
Abstract: A spatial semiparametric model is presented for analyzing spatial heterogeneity in blood-brain barrier permeability. Traditional permeability measurement methods struggle to differentiate between actual permeability changes and the influences of microvascular surface area and solute movement from adjacent voxels. The developed spatial semiparametric model integrates fixed effects with a nonparametric spatial trend component while accounting for spatial autocorrelation in the data. 3D brain imaging data analysis presents enormous computational challenges, with a typical dataset of 40 patients generating more than 5 million voxels. Conventional estimation methods are computationally infeasible at this scale. To address this issue, a profile likelihood method is employed for parameter estimation, and computational strategies are developed based on the biological properties of BBB permeability. Efficient determinant calculation methods are implemented for large-scale sparse matrices, utilizing multi-level regularization techniques and massive parallel computing to reduce computation time from months to hours. The analysis applied to actual DCE-MRI datasets reveals regional differences in BBB permeability among healthy individuals and successfully differentiates permeability patterns between gray and white matter.
