Group independent component analysis (group-ICA) is a powerful data-driven method used for the resting-state fMRI (rfMRI) data analysis. This method is typically used to extract the resting state networks (RSNs). However, this approach does not allow the investigation of the within-network FC changes that occur within the networks.

This study combines the ICA and a local clustering method in order to find common spatial patterns and split the anatomically distinct areas belonging to a specific network. In particular, we focused on the default mode network (DMN) as it is hypothesized to play an important role in behavior and cognition¹ and to physiologically change its functional behavior across the life span. This joint approach gave us the chance to obtain more detailed information about within-network FC changes among the DMN sub-regions.

The study was conducted on 172 healthy right-handed volunteers (6-79y, 110 females) divided into 15-year age groups: G1 (N=19, 6-15y, 9 females), G2 (N=51, 16-30y, 34 females), G3 (N=35, 31-45y, 23 females), G4 (N=33, 46-60y, 22 females) and G5 (N=34, over 60y, 22 females). rfMRI images (TR/TE=2500/30 ms; resolution=3.1x3.1x2.5 mm³; 39 axial slices; 160 volumes) were acquired using a 1.5T MRI scanner. High-resolution T1-weighted scans were also collected. After standard preprocessing with FSL², data were coregistered to MNI space using the Advanced Normalization Tools (ANTs)³. Common spatial patterns were extracted from the whole dataset using the group-ICA. The FSL tool cluster (http://fsl.fmrib.ox.ac.uk/fsl/fslwiki/Cluster) was then used to split the RSNs into clusters (CLs). Subjects-specific spatial maps and time series were recovered from the group CLs⁴. Second level analyses were focused on the DMN CLs: medial pre-frontal cortex (mPFC), posterior cingulate cortex (PCC), left and right parietal lobes (LPar and RPar). Statistical analyses were performed using SPSS 21 (Statistical Package for the Social Sciences) to examine age-related FC differences between the DMN CLs in the five groups (one-way univariate ANOVA with Bonferroni post-hoc tests).The ANOVA showed statistically significant age-related FC changes for the pairs mPFC-PCC, mPFC-LPar and mPFC-Rpar (p ≤ 0.01). Post-hoc analyses highlighted that the FC between mPFC-PCC significantly increases from G1 to G2 and decreases from G2 to G5. FC between mPFC-LPar revealed to be lower in G1 compared to all the other groups (G2, G3, G4 and G5). FC between mPFC-RPar was lower in G1 compared to G2. All these significant results (p ≤ 0.05) were Bonferroni-corrected for multiple comparisons.Our findings are in line with previous studies^5,6,7,8 and strongly support the hypothesis that the long-range FC between anterior and posterior regions of the DMN increases from the childhood (G1) to the young adulthood (G2) and slowly decreases with aging. We used a robust and powerful method to identify common spatial patterns in a heterogeneous group of subjects, and combined it with an anatomical parceling of the network of interest, in order to study its within-network interactions. This approach enabled us to study functional changes among areas belonging to the DMN. This work studied the neuro-physiological cerebral evolution and could be performed on all the RSNs, leading to a better understanding of the physio-pathological cognitive functionality.The joint use of ICA and anatomical clustering enabled to evaluate the functional changes across circuits involved in the DMN and provided evidence that this network can be studied through the interplay of its elements to follow age-related brain functionality.