Mapping the mouse brain with rs-fMRI: An optimized pipeline for functional network identification

• Published in: NeuroImage (Orlando, Fla.) Vol. 123; pp. 11 - 21
• Main Authors: Zerbi, Valerio, Grandjean, Joanes, Rudin, Markus, Wenderoth, Nicole
• Format: Journal Article
• Language: English
• Published: United States Elsevier Inc 01.12.2015; Elsevier Limited
• Subjects: Algorithms; Animals; Artifact detection; Brain - physiology; Brain Mapping - methods; Classification; Cluster Analysis; Conflicts of interest; Data cleaning; Datasets; Hierarchical clustering; Image Processing, Computer-Assisted; Independent component analysis; Magnetic Resonance Imaging - methods; Male; Methods; Mice; Mice, Inbred C57BL; Mouse; Neural Pathways - physiology; Noise; Physiology; Resting state fMRI; Signal Processing, Computer-Assisted; Resting state fMRI; Artifact detection; Mouse; Hierarchical clustering; Data cleaning; Independent component analysis
• ISSN: 1053-8119; 1095-9572
• DOI: 10.1016/j.neuroimage.2015.07.090

The use of resting state fMRI (rs-fMRI) in translational research is a powerful tool to assess brain connectivity and investigate neuropathology in mouse models. However, despite encouraging initial results, the characterization of consistent and robust resting state networks in mice remains a methodological challenge. One key reason is that the quality of the measured MR signal is degraded by the presence of structural noise from non-neural sources. Notably, in the current pipeline of the Human Connectome Project, a novel approach has been introduced to clean rs-fMRI data, which involves automatic artifact component classification and data cleaning (FIX). FIX does not require any external recordings of physiology or the segmentation of CSF and white matter. In this study, we evaluated the performance of FIX for analyzing mouse rs-fMRI data. Our results showed that FIX can be easily applied to mouse datasets and detects true signals with 100% accuracy and true noise components with very high accuracy (>98%), thus reducing both within- and between-subject variability of rs-fMRI connectivity measurements. Using this improved pre-processing pipeline, maps of 23 resting state circuits in mice were identified including two networks that displayed default mode network–like topography. Hierarchical clustering grouped these neural networks into meaningful larger functional circuits. These mouse resting state networks, which are publicly available, might serve as a reference for future work using mouse models of neurological disorders. •We applied an automatic ICA denoiser (FSL-FIX) on a mouse resting state fMRI dataset.•We examined test–retest consistency and compared FIX with other pre-processing strategies.•FIX showed very high accuracy in detecting true signals and true noise automatically.•FIX-cleaned data improved the strength and detectability of mouse neural networks.•Hierarchical clustering grouped the revealed networks into meaningful larger functional circuits.