BACKGROUND AND AIM: Over the last few years, significant developments have been taking place in high-dimensional neuroimaging data analysis. METHODS: To determine which neuroimaging features may mediate effects of neighborhood socioeconomic status (SES) on cognitive assessments, we analyzed data from the Adolescent Brain Cognitive Development (ABCD) study and conducted mediation analysis with the Causal Inference Test (CIT). To facilitate this analysis, we extended the data reduction method, Partition, for application to high-dimensional neuroimaging data. Unlike the previously proposed version that directs reduction based on attribute correlation only, we adopted a strategy for the extension that combines spatial distance with attribute correlation. RESULTS:We performed computer simulation studies to contrast the new version with the original approach and found that incorporating spatial information substantially increased data reduction for the same level of information capture and improved computational efficiency. Neuroimaging data were reduced via the modified Partition approach, yielding 88% reduction for 80% information capture. A sulcal depth feature in the superior frontal language area was identified as consistent with causal mediation (CIT P = 0.04). For this feature, there was strong evidence of marginal associations for i) neighborhood vs cognitive, ii) neighborhood vs sulcal depth, and iii) sulcal depth vs cognitive (P = 6.91e-11, 6.11e-06, and 9.63e-04, respectively). CONCLUSIONS:These results suggest that higher socioeconomic status of a neighborhood may lead to less sulcal depth in the superior frontal language area, resulting in improved cognitive performance. KEYWORDS: Big data, Causal inference, Socio-economic factors