YSPH Biostatistics Seminar: “Exploring Aging through Statistical Shape Analysis in Imaging Studies”

NOTE: BIS 526 students are required to attend in person. Others are invited to attend in person, but may also attend via zoom.

Speaker- Chao Huang, Ph.D.

Title- “Exploring Aging through Statistical Shape Analysis in Imaging Studies”

Abstract

Alterations in subcortical brain structures, such as changes in the shape of the hippocampus, are typically interconnected and associated with the natural process of brain aging. Detecting and comprehending the geometric distinctions within these diverse subcortical brain structures is crucial for monitoring brain aging. However, existing learning methods encounter several challenges, including: (i) the non-Euclidean representation of subcortical shapes; (ii) the complex spatial correlation structure in local geometry; (iii) subject-level imaging heterogeneity due to misalignment of shapes in imaging preprocessing steps; (iv) group-level imaging heterogeneity arising from distinct brain aging patterns; and (v) geometric variations associated with covariates of interest (e.g., gender and education length), which may be high-dimensional. To tackle these challenges, we propose a Mixture of Shape-on-scalar Factor Regression Model (MS-FARM). In each cluster, a geodesic regression structure, including covariates of interest and an alignment step, is established along with the Riemannian Gaussian distribution in the pre-shape space. Additionally, a latent factor model is constructed in the tangent space. A penalized likelihood approach is employed for variable selection in MS-FARM. Furthermore, to integrate multiple 2D/3D subcortical shapes and explore optimal Riemannian local embedding representations of shapes across different subgroups, we introduce a Geometric Deep Multi-Autoencoders (Geo-DMAE) framework. Specifically, in each geometric deep autoencoder, the prealigned shapes go through a rotation & embedding layer and a multilayer perceptron (MLP) based autoencoder. Subgroup patterns are learned with mixing proportions defined through a regression model (e.g., logistic) applied to the latent features integrated from multiple autoencoders. Finally, simulation studies and real data analysis based on multiple brain subcortical structures from the brain aging study are conducted to assess the finite sample performance of MS-FARM and Geo-DMAE.