Morphometric features enhance phenotype discrimination in frontotemporal lobar degeneration

Abstract Frontotemporal lobar degeneration is associated with diverse clinical phenotypes underlain by multiple disease pathologies and genetic mutations for which traditional structural magnetic resonance imaging (MRI) analyses lack discriminatory sensitivity and specificity. Here, we use a data-driven multivariate method to extract a concise set of MRI-derived shape morphometric features and cross-sectionally examine the discriminatory capability of their unique combinations in three frontotemporal lobar degeneration clinical phenotypes. Patients with sporadic or familial frontotemporal lobar degeneration clinical syndromes across two cohorts (i.e. behavioral variant (n = 173), non-fluent variant primary progressive aphasia (n = 63), semantic variant primary progressive aphasia (n = 41)) and 158 controls were assessed. Cortical morphometry measures of cortical thickness, surface curvature, and metric distortion were extracted, contrasted with controls using linear models, and additionally entered into a sparse partial least squares discriminatory analysis (sPLS-DA) designed to model multimodal signatures unique to each phenotype. Discriminatory power of partial least squares-derived features was tested on independent, age-matched test data. We found that each cortical morphometric feature significantly differed between clinical syndromes in dissociable spatial patterns. On independent data, the combination of cortical thickness and surface curvature best discriminated between behavioural variant and non-fluent variant primary progressive aphasia patients from controls. For semantic variant primary progressive aphasia, any model including cortical thickness maximized model performance. The sparse partial least squares approach indicated distinctive brain regions contribute to discrimination for each shape feature, suggesting each feature may reflect unique aspects of neurodegeneration across groups. This method could prove invaluable in future studies for early detection of frontotemporal lobar degeneration phenotypes.