Keywords: Varying coefficient model, smoothing splines, high dimensionality, heterogeneous data, longitudinal analysis, Alzheimer's disease, magnetic resonance imaging
Recent evidence shows that structural magnetic resonance imaging (MRI) is an effective tool for Alzheimer's disease (AD) diagnosis. While traditional MRI-based diagnosis uses images acquired at a single time point, a longitudinal study is more sensitive and accurate in detecting early pathological changes of the AD. Two main difficulties arise in longitudinal MRI-based diagnosis: (1) the inconsistent longitudinal scans with the massive dataset (i.e., different scanning time points and different total number of scans); (2) the heterogeneous progressions of high-dimensional regions of interest in MRI. In this work, we propose a novel smoothing spline based feature selection method for AD diagnosis using longitudinal structural MR images, which is robust to inconsistencies among longitudinal scans and is able to adapt to heterogeneous progressions. Specifically, first, the linear model with varying coefficients is built for modeling heterogeneous high-dimensional regions of interest in MRI; and second, the coefficients for regions of interest are estimated based on the smoothing splines and a novel efficient convex penalty, which enable to simultaneously achieve estimation and feature selection for regions of interest with the guarantee of the false discovery rate (FDR). Using these selected longitudinal features, a linear support vector machine (SVM) is adapted to classify AD subjects or mild cognitive impairment (MCI) subjects from healthy controls (HCs). We perform experiments on the Alzheimer’s Disease Neuroimaging Initiative (ADNI) database demonstrate the superior performance and efficiency of the proposed method, with classification accuracies of 91.15% for AD versus HC and 81.13% for MCI versus HC, respectively, which outperform the state-of-art results in longitudinal studies on the same dataset (88.30% for AD versus HC and 79.02% for MCI versus HC).