Brain microbleeds (BMBs), often found in stroke patients, are widely accepted as a manifestation of cerebral small vessel disease¹. BMBs are related to white matter changes and reflect diffuse pathology². They are also a major risk factor for vascular dementia³. A previous stroke study⁴ indicated that BMBs might play an important role in post-stroke cognitive impairment⁵. Because the mechanism by which BMBs affects cognition remains elusive, it is imperative that we understand their effect on the structural brain network of stroke patients at a global level.

Participants 48 stroke patients (63 ± 14 years old) were recruited.

Image acquisition DWIs were acquired using single-shot EPI with b values of 1000 and 2000 s/mm2 along 32 gradient directions using a 3T scanner (Achieva TX scanner, Philips Healthcare).

Post-processing 3D-MPRAGE images were segmented into 90 brain regions according to AAL atlas. Whole-brain tractography was obtained using Diffusion Toolkit (trackvis.org/dtk/). Fiber tracts traversing two regions were counted, resulting in a connectivity matrix. Network metrics, such as clustering coefficient, characteristic path length, normalized clustering coefficient (γ), normalized characteristic path length (λ), global and local efficiency were obtained using the Brain Connectivity Toolbox⁶.

BMBs scoring SWI images were examined by a radiologist to score BMBs using the Microbleed Anatomical Rating Scale⁷ (MARS). 12 patients had MARS = 1, and 3 had MARS = 2.

Statistical Analysis Repeated measures ANOVA (age and gender as covariates) followed by post-hoc independent samples t-tests were performed. Pearson correlations were performed to determine the association between network metrics and MARS.

Global network Larger clustering coefficient and local efficiency were observed for patients with MARS = 1 (both p < 0.001) and MARS = 2 (p = 0.007 and 0.017, respectively) as compared to those without BMBs. Clustering coefficient (r = 0.69, p < 0.001) and nodal efficiency (r = 0.73, p < 0.001) were correlated with MARS.

Nodal characteristics The connections amongst the regions identified as hubs in normal subjects (data from another study) and stroke patients without BMB (red spheres in the top row of Figure 1) are more clustered in patients with BMBs. Larger local clustering coefficient were found in the superior frontal cortex (MARS = 1: p = 0.006; MARS = 2: p = 0.016), hippocampus (MARS = 1: p = 0.009; MARS = 2: p = 0.036), precuneus (MARS = 1: p = 0.039; MARS = 2: p = 0.025), putamen (MARS = 1: p < 0.001; MARS = 2: p < 0.001) and globus pallidus (MARS = 1: p = 0.001; MARS = 2: p = 0.001). Higher nodal efficiency was observed in insula (MARS = 2: p = 0.034), hippocampus (MARS = 2: p = 0.019), and precuneus (MARS = 1: p = 0.003; MARS = 2: p = 0.004).

The balance between segregation and integration, as achieved by the coordination of activities amongst various neural components, is fundamental in maintaining normal brain functioning⁸, and the disruption thereof could lead to dysfunction of cognition and behavior⁹. The fact that we observed significant correlation between segregation measurements and the MARS of stroke patients suggests that the presence of BMBs may cause structural network to be more functionally segregated and have stronger local connectivity (in other words, loss of long-range connections). We also observed increased local clustering coefficient and nodal efficiency of some brain hubs in patients with BMBs, further supporting the notion that the connections amongst the hubs that share similar function are strengthened. Similar alteration in network organization was also observed in various neurological and psychiatric disorders^10-12.

Another noteworthy finding is that some of the brain hubs of patients without BMBs were replaced by regions such as the calcarine sulcus and lingual gyrus for patients with BMBs, suggesting an enhancement in specialized functions such as visual and language function due to the presence of BMBs. Similar enhancement in specialized functions is also found in patients with autism spectrum disorders¹³ whom had relatively good performance of language function. In other words, the balance between segregation and integration for stroke patients with BMBs is clearly disrupted, thus causing cognitive impairment (although we don’t have any cognitive measures to corroborate our findings). Indeed, when brain could not fulfill cognitive demand, it would develop into a more clustered and less costly configuration¹⁴.

Our findings suggested that the presence of BMBs disrupts the balance between the ability to process integrated and segregated information for stroke patients, and that brain network analysis is a sensitive method to assess cognitive impairment caused by cerebral small vessel disease.