Cognitive Training Improves Task Performance and Modulates Brain Activation in Healthy Populations

A systematic review and meta-analysis of 67 neuroimaging studies across 2,410 healthy individuals indicates that cognitive training yields domain-general cognitive gains while triggering bidirectional functional brain adaptations. Published in Behavioral and Brain Functions, the investigation synthesized coordinate-based neuroimaging data and behavioral performance outcomes across diverse training domains, paradigms, and age groups spanning childhood to late adulthood. The research team showed that structured cognitive practice produces a statistically significant, moderate enhancement in task performance, characterized by an overall effect size of Hedges’ g = 0.546 (95% confidence interval = 0.414 to 0.678; t = 8.149; p < 0.0001). Concurrently, whole-brain neuroimaging meta-analysis revealed two convergent clusters of task-related activation changes across healthy participants: increased neural activity in the right superior parietal lobule alongside decreased activation within the bilateral anterior cingulate cortex. These reciprocal adaptations appeared consistently across varied intervention formats, identifying candidate domain-general neural signatures of training-induced neuroplasticity in healthy populations.

The motivation for this comprehensive synthesis stems from an urgent public health need to prioritize upstream, preventive strategies that support cognitive health across the human lifespan. Cognitive functioning forms the bedrock of everyday independence, overall quality of life, and psychological well-being. Concurrently, cognitive decline and dementia represent an escalating international challenge, with approximately 57 million people currently living with dementia worldwide—a figure projected to reach 153 million by 2050, with Alzheimer’s disease accounting for roughly 115 million cases. While several disease-modifying pharmacological therapies have demonstrated preliminary efficacy in slowing early progression, their clinical benefits remain modest and safety concerns persist. Within this framework, cognitive training has emerged as a scalable, non-pharmacological behavioral approach to enhance cognitive performance. However, existing trials have differed substantially in training parameters, cohort demographics, and assessment tasks, leaving unclear whether training produces a domain-general effect and which task-related neural changes most reliably track observed behavioral improvements.

Although prior systematic reviews evaluated cognitive training outcomes, important methodological limitations constrained earlier literature syntheses. Most coordinate-based meta-analyses using activation likelihood estimation evaluated task-related activation increases and decreases in separate models, preventing the detection of concurrent, bidirectional neural adaptations within a unified framework. Furthermore, previous studies did not systematically evaluate heterogeneity across participant demographics, training parameters, and assessment tasks, which hindered mechanistic interpretation and generalizability. Crucially, brain-behavior coupling had rarely been quantified directly, leaving it uncertain whether observed activation changes consistently covaried with behavioral gains. To resolve these gaps, the investigators prospectively registered their review protocol in PROSPERO under identifier CRD42024599205 and followed the Preferred Reporting Items for Systematic Reviews and Meta-Analyses guidelines. The team implemented advanced multivariate modeling and seed-based mapping techniques capable of integrating bidirectional activation patterns and directly testing brain-behavior associations across diverse healthy cohorts.

To identify eligible literature, the authors conducted comprehensive searches across five major databases—PubMed, Web of Science, APA PsycINFO, MEDLINE, and Embase—from database inception through January 2026 without publication year restrictions. The search strategy incorporated medical subject headings and controlled terms related to functional neuroimaging and cognitive training. Initial searches retrieved 6,164 records; after removing 3,293 duplicates using EndNote 21, 2,871 unique citations remained. Two authors independently screened titles and abstracts, excluding 2,784 records and advancing 87 articles to full-text assessment, with disagreements resolved through consensus or consultation with a senior author. Eligible studies were required to report stereotactic coordinates in Montreal Neurological Institute or Talairach space for task-related activation changes following structured cognitive practice, perform whole-brain analyses, use pre-post or controlled designs, and recruit cognitively normal participants without dementia, mild cognitive impairment, or neurological and psychiatric disorders. Ultimately, 67 studies met all neuroimaging criteria, with 60 contributing behavioral outcome data.

Across the 67 neuroimaging studies included in the meta-analysis, the combined sample comprised 2,410 healthy participants with an average age of 33.62 years, spanning an age range from 7.8 to 72.1 years. Cognitive training programs across the included trials showed substantial procedural variation. On average, the interventions encompassed 19.68 training sessions (ranging from 1 to 360 sessions), representing a mean total training duration of 902.03 minutes (with a range spanning 3 to 21,600 minutes). Training programs were delivered at an average frequency of 4.39 sessions per week (ranging from 1 to 8 sessions weekly), with individual sessions averaging 42.33 minutes in duration (ranging from 3 to 240 minutes). To ensure methodological transparency, the authors assessed reporting quality across the included studies using a 17-item checklist evaluating participant characteristics, study design, imaging acquisition, data preprocessing, statistical analysis, and conclusion reporting standards.

To synthesize cognitive improvements without small-sample bias or outcome dependency issues, the investigators applied a three-level hierarchical multivariate meta-analysis in RStudio using the metafor and clubSandwich packages. Random effects were specified across individual outcomes (Level 1), participant samples (Level 2), and primary studies (Level 3), with within-study correlation among cognitive outcomes fixed at 0.5 according to Cochrane guidelines. Standardized mean differences were derived as Hedges’ g from pre-post comparisons or change scores standardized by pooled within-group standard deviations, with reaction times and error rates reverse-coded so positive values denoted performance gains. For neuroimaging synthesis, the authors utilized the anisotropic effect size algorithm in seed-based d mapping (AES-SDM, version 5.15). Unlike conventional activation likelihood estimation, AES-SDM incorporates increases and decreases into a single map and permits null-finding inclusion. Preprocessing used a 20-millimeter full-width at half-maximum Gaussian kernel with 500 permutations, thresholding clusters at uncorrected p < 0.0001, peak SDM-Z ≥ 1, and extent ≥ 10 voxels.

The meta-analytic results confirmed significant cognitive performance enhancements alongside localized brain activation changes. The multivariate analysis of 60 studies demonstrated that cognitive training significantly improved task performance (Hedges’ g = 0.546, 95% confidence interval = 0.414 to 0.678; t = 8.149; p < 0.0001), representing a moderate behavioral effect across domains. In parallel, the SDM meta-analysis of 67 studies identified two significant clusters of task-related functional activation changes following training. These clusters exhibited increased activation in the right superior parietal lobule and significantly reduced activation within the bilateral anterior cingulate cortex. Robustness analyses supported these findings. Jackknife sensitivity analyses demonstrated high replicability across study combinations for both regions. Furthermore, multivariate behavioral models remained consistent under alternative within-study correlation assumptions of r = 0.3, 0.7, and 1.0, and repeating the SDM analysis after excluding studies with fewer than 10 participants confirmed the stability of both neuroimaging clusters.

The investigation further established that cognitive performance improvements were directly coupled with regional functional brain adaptations. Regression analyses showed that cognitive gains were significantly associated with training-induced activation increases in the right superior parietal lobule (r = 0.233, p = 0.004) and activation decreases in the bilateral anterior cingulate cortex (r = -0.385, p < 0.001). Among tested moderators, only intervention completion rate significantly moderated activation changes in both the right superior parietal lobule (r = 0.214, p = 0.027) and bilateral anterior cingulate cortex (r = -0.255, p = 0.009), an effect that remained robust after excluding high-leverage studies. In contrast, total training duration and participant sex showed no significant moderating effects on neural changes or cognitive gains. Similarly, comparative analyses found no significant differences across categorical moderators, including training domain, training paradigm, assessment task type, delivery mode, difficulty regulation, age category, study design, or control condition type.

The authors interpret these bidirectional functional adaptations as evidence of coordinated neural resource optimization supporting cognitive performance. The right superior parietal lobule serves as a key hub within the frontoparietal attention network, supporting attentional control and cognitive resource allocation. Increased right superior parietal lobule activation following training may therefore reflect enhanced recruitment of task-relevant attentional networks during cognitive execution. Conversely, the bilateral anterior cingulate cortex participates in task-irrelevant monitoring or introspective processing. Attenuated bilateral anterior cingulate cortex activation following cognitive training may indicate the suppression of task-irrelevant monitoring or introspective processing, thereby freeing resources for task execution. Together, these complementary shifts suggest that cognitive training improves performance by strengthening task-relevant frontoparietal activity while suppressing extraneous monitoring, fostering more efficient cross-network neural coordination during cognitive tasks.

Despite these compelling findings, the authors highlighted several methodological limitations. The meta-analysis relied on coordinate-based study-level summary data rather than individual-level imaging scans, which constrained statistical power for detecting subtle moderator effects across spatially diffuse activations. To balance specificity and sensitivity without overly conservative corrections, the authors applied an uncorrected voxel-wise threshold of p < 0.0001 combined with whole-brain randomization-based inference, meaning subtle but potentially meaningful moderator effects may have gone undetected. Furthermore, an uneven distribution of published literature across specific cognitive subdomains precluded formal tests of domain-specific adaptations. Finally, Egger’s regression tests indicated significant publication asymmetry for bilateral anterior cingulate cortex findings (β = 0.85, t = 4.20, p < 0.001), despite no significant asymmetry for the right superior parietal lobule (β = 0.06, t = 0.39, p = 0.698) or behavioral performance (β = 0.76, t = 1.64, p = 0.102). Future research should prioritize larger cohorts, individual-level imaging datasets, and non-linear modeling to refine these neuroplasticity insights.

Subject of Research: Neuroimaging meta-analysis of cognitive training effects on brain activation and performance in healthy populations

Article Title: Domain-general behavioral gains and neural correlates of cognitive training in healthy populations: a neuroimaging meta-analysis

Article References: Li, G., Liu, C., Liu, Y., & Zhang, Y. (2026). Domain-general behavioral gains and neural correlates of cognitive training in healthy populations: a neuroimaging meta-analysis. Behavioral and Brain Functions, 22(1), Article 29. https://doi.org/10.1186/s12993-026-00338-w

Image Credits: AI Generated

DOI: 10.1186/s12993-026-00338-w

Keywords: cognitive training, neuroimaging meta-analysis, brain activation, right superior parietal lobule, bilateral anterior cingulate cortex, Hedges g, seed-based d mapping, multivariate meta-analysis, neuroplasticity, frontoparietal attention network, brain-behavior coupling, training completion rate, healthy populations, task performance, attentional control, jackknife sensitivity analysis, PRISMA framework