Alterations in Causal Functional Brain Networks in Alzheimer's Disease: A resting-state fMRI study

Abstract

BackgroundAlterations in functional connectivity (FC) of the brain is known to predate the onset of clinical symptoms of Alzheimers disease (AD) by several decades. Identifying the altered functional brain networks in AD can help in its prognosis and diagnosis. ObjectiveFC analysis is predominantly correlational. However, correlation does not necessarily imply causation. This study aims to infer causal functional connectivity (CFC) from functional magnetic resonance imaging (fMRI) data and obtain the sub-networks of CFC that are altered in AD compared to cognitively normal (CN) subjects. MethodsWe used the recently developed Time-aware PC algorithm to infer CFC between brain regions. The CFC outcome was compared with correlation-based functional connectivity obtained by sparse partial correlation. Then, Network-based Statistics (NBS) was used to obtain CFC sub-networks that altered in AD subjects compared to healthy controls while correcting for multiple comparisons at 5% level of significance. ResultsOur findings identified causal brain networks involving the inferior frontal gyrus, superior temporal gyrus (temporal pole), middle temporal gyrus (temporal pole), and different lobes of the cerebellum to be significantly reduced in strength in AD compared to CN group (p-value = 0.0299; NBS corrected). In the sample dataset that has been analysed, no brain networks were found to exhibit significant increase in strength in AD compared to CN group at 5% level of significance with NBS correction. ConclusionsOur findings provide insights into disruptions in causal brain networks in AD. The corresponding brain regions are in agreement with published medical literature on brain regions impacted by AD. Our work establishes a methodology for finding causal brain networks that are affected by AD using TPC algorithm to compute subject-specific CFC and then using NBS for finding CFC subnetworks that alter between AD and CN groups. Larger datasets are expected to identify further subnetworks affected by AD.