The Effect of Overcoming the Digital Divide on Middle Frontal Gyrus Atrophy in Aging Adults: Large-Scale Retrospective Magnetic Resonance Imaging Cohort Study

Abstract

Background: The rapid integration of information technology into daily life has exacerbated the digital divide (DD), particularly among older adults, who often face barriers to technology adoption. Although prior research has linked technology use to cognitive benefits, the long-term neurostructural and cognitive consequences of the DD remain poorly understood.

Objective: The aim of this study is to use large-scale neuroimaging data to examine how the DD affects long-term brain structure and cognitive aging in older adults. It specifically investigates (1) structural and cognitive differences between older adults with and without DD engagement, (2) predictive relationships between group-distinctive brain regions and cognitive outcomes, and (3) longitudinal impacts of DD exposure on accelerated aging trajectories of neural substrates and cognitive functions.

Methods: The study included 1280 community-dwelling older adults (aged 65-90 y) who completed comprehensive cognitive assessments and structural magnetic resonance imaging scans at baseline. Longitudinal data were available for 689 participants (mean follow-up 3.2 y). Participants were classified into the DD (n=640) and overcoming DD (n=640) groups using rigorous propensity score matching to control for age, education, gender, and baseline health conditions. A computational framework using the searchlight technique and cross-validation classification model investigated group differences in structural features and cognitive representation. The aging rate of each voxel's structural feature was calculated to explore the long-term influence of the DD.

Results: The DD group showed significant deficits in executive function (t=4.75; P<.001; Cohen d=0.38) and processing speed (t=4.62; P<.001; Cohen d=0.37) compared to the overcoming DD group. Reduced gray matter volume in the DD group spanned the fusiform gyrus, hippocampus, parahippocampal gyrus, and superior temporal sulcus (false discovery rate-corrected P<.05). The computational framework identified the key structural substrates related to executive function and processing speed, excluding the ventro-orbitofrontal lobe (classification accuracy <0.6). Longitudinal findings highlighted the long-term impact of the DD. The DD group exhibited faster gray matter volume decline in the middle frontal gyrus (t=3.95 for the peak voxel in this cluster, false discovery rate-corrected P<.05), which mediated 17% of episodic memory decline (P=.02).

Conclusions: Older adults who overcome the DD demonstrate preserved gray matter structure and slower cognitive decline, particularly in frontotemporal regions critical for executive function. Our findings underscore that mobile digital interventions should be explored as potential cognitive decline prevention strategies.

Keywords: cognitive aging; digital divide; internet use; neural decline; neuroplasticity; sMRI; structural magnetic resonance imaging.

Figure 1.. An illustration of the statistical framework using the searchlight technique. (A) Searchlight analysis generated a neighborhood voxel matrix using a sliding spherical window. PCA was then applied to reduce data dimensions and extract meaningful features. (B) SVM was used for voxel-based group classification with 10-fold cross-validation. Distinct brain regions with high classification accuracy were identified in a 3D accuracy map using a specified threshold. (C) The features were added to a GLM to predict multidomain cognitive function scores. Correlations between predicted and observed scores were calculated to assess domain-specific representation. DD: digital divide; EF: executive function; GLM: general linear model; MRI: magnetic resonance imaging; ODD: overcoming the digital divide; PCA: principal component analysis; PS: processing speed; SVM: support vector machine; VFT: Verbal Fluency Test.

Figure 2.. The structural differences between the older adults overcoming the digital divide and those who failed to overcome the divide. Cluster 1 (peak label at temporal pole: t=4.41; P<.001) and cluster 2 (peak label at hippocampus: t=4.19; P<.001) included the fusiform gyrus, parahippocampal gyrus, hippocampus, and temporal pole. Cluster 3 (peak label at Rolandic operculum: t=4.03; P<.001) included the Rolandic operculum, superior temporal gyrus, supramarginal gyrus, and Heschl gyrus. Cluster 4 (peak label at frontal orbital cortex: t=4.14; P<.001) included the orbital part of the inferior frontal gyrus and part of the insula.

Figure 3.. Constraining the brain regions specific for classifying the overcoming the digital divide group and the digital divide group. (A) The constrained regions refer to the voxels with classified accuracy above the threshold. The excluded regions are the rest regions that displayed significant gray matter volume difference between the 2 groups; however, the classified accuracy was lower than the threshold. Classified accuracy refers to the accuracy of classifying 2 groups based on voxel-level characteristics through the searchlight statistical framework (see the Methods section). (B) The specific distribution and accuracy value of the constrained and excluded regions. The upper map depicts the accuracy distribution of constrained regions, indicating that the average classification accuracy from high to low is the fusiform gyrus, Heschl, Rolandic operculum, parahippocampal gyrus, and hippocampus. The lower map shows the accuracy distribution of the excluded regions, mainly including the inferior frontal regions and orbital frontal regions.

Figure 4.. Cognitive representation of the constrained regions of the digital divide. (A) The distribution of the constrained regions predicting EF and PS. (B) The predictive accuracy(r) of the specific brain region distribution of EF and PS. Combining (A) and (B), we found that the brain areas in the constrained region representing executive function and processing speed are mainly the hippocampus, parahippocampal gyrus, temporal pole, and fusiform gyrus. (C) EF regions compared to PS regions: the result indicated the regions specific to EF, including the Rolandic operculum, Heschl gyrus, and superior temporal gyrus, which were decoded as tactile and auditory-related cortices. EF: executive function; PS: processing speed.

Figure 5.. The difference in the decline rate of brain structure between the DD and ODD groups. (A-B) The GMV decline rate in the MFG was significantly lower in the ODD group than in the DD group. (C) The decline rate in the MFG significantly correlated with individuals’ memory performance (R=0.17; P=.02) and showed no significant association with scores in other cognitive domains. DD: digital divide; GMV: gray matter volume; MFG: middle frontal gyrus; ODD: overcoming the digital divide.