Effective artifact removal in resting state fMRI data improves detection of DMN functional connectivity alteration in Alzheimer's disease

Artifact removal from resting state fMRI data is an essential step for a better identification of the resting state networks and the evaluation of their functional connectivity (FC), especially in pathological conditions. There is growing interest in the development of cleaning procedures, especially those not requiring external recordings (data-driven), which are able to remove multiple sources of artifacts. It is important that only inter-subject variability due to the artifacts is removed, preserving the between-subject variability of interest—crucial in clinical applications using clinical scanners to discriminate different pathologies and monitor their staging. In Alzheimer's disease (AD) patients, decreased FC is usually observed in the posterior cingulate cortex within the default mode network (DMN), and this is becoming a possible biomarker for AD. The aim of this study was to compare four different data-driven cleaning procedures (regression of motion parameters; regression of motion parameters, mean white matter and cerebrospinal fluid signal; FMRIB's ICA-based Xnoiseifier—FIX—cleanup with soft and aggressive options) on data acquired at 1.5 T. The approaches were compared using data from 20 elderly healthy subjects and 21 AD patients in a mild stage, in terms of their impact on within-group consistency in FC and ability to detect the typical FC alteration of the DMN in AD patients. Despite an increased within-group consistency across subjects after applying any of the cleaning approaches, only after cleaning with FIX the expected DMN FC alteration in AD was detectable. Our study validates the efficacy of artifact removal even in a relatively small clinical population, and supports the importance of cleaning fMRI data for sensitive detection of FC alterations in a clinical environment.