Support Vector Based Autoregressive Mixed Models of Longitudinal Brain Changes and Corresponding Genetics in Alzheimer’s Disease

Abstract

Longitudinal data used as repeat measures may capture the proportion of total variance due to genetic factors with greater sensitivity. However, for brain imaging in studies of older adults, there is a steady decline of brain tissue. It is important to establish such estimation methods using longitudinal data, while properly modeling within-subject variation and rate of tissue atrophy. However, to date, neuroimaging studies have been limited to using only two timepoints, and have not considered diagnostic-specific trends in clinically heterogeneous samples. Modeling temporal patterns of brain structure specific to neurodegenerative disease, while simultaneously assessing the contribution of genetic and environmental risk factors, is essential to understanding and predicting disease progression. We use data from the Alzheimer’s Disease Neuroimaging Initiative (ADNI) to model the genetic effects on brain cortical measurements from three repeated measures across two years. We refine our model for specific diagnostic groups, including cognitively normal elderly individuals, individuals with mild cognitive impairment and AD, and then distinguish between those who remain stable or convert to AD. We propose a support vector based, longitudinal autoregressive linear mixed model (ARLMM) for long-term repeated measurements, offering greater sensitivity than cross-sectional analyses in baseline scans alone.