Systematic signal differences across HCP cohorts—particularly weaker temporal signal-to-noise ratio in HCP-YA—propagate to downstream resting-state network measures and can qualitatively alter estimated lifespan trajectories, including partially inverting expected lifespan patterns, but harmonization approaches substantially lessen these artifactual differences.
Key Findings
Results
The HCP-YA cohort exhibits systematically weaker temporal signal-to-noise ratio (tSNR) relative to HCP-D and HCP-A cohorts.
tSNR differences across cohorts were attributed to changes in scanner hardware and acquisition sequences across study phases of the Human Connectome Project.
HCP spans three cohorts: development (HCP-D), young adulthood (HCP-YA), and aging (HCP-A), each with differing neuroimaging acquisition protocols.
The signal quality discrepancy in HCP-YA was characterized as systematic rather than random noise.
HCP datasets are widely used as high-quality reference datasets in tool validation, replication studies, and cross-cohort meta-analyses, amplifying the impact of these differences.
Results
Signal quality discrepancies in HCP-YA propagate to differences in overall resting-state functional correlations across cohorts.
The weaker tSNR in HCP-YA led to measurable differences in resting-state functional correlation magnitudes compared to HCP-D and HCP-A.
These differences were observed at both the whole-brain level and at the node level.
The propagation of signal differences to functional connectivity measures was described as systematic rather than random.
Results
Protocol-driven signal differences in HCP-YA alter whole-brain and node-level measures of resting-state network organization.
Affected network measures included system segregation, modularity, and participation coefficient.
These are established metrics of functional brain network organization commonly used in lifespan neuroscience research.
Differences were observed at both macro (whole-brain) and micro (node-level) scales of network analysis.
Results
Resting-state network measures derived from HCP-YA depart from expected lifespan trajectories in a manner consistent with protocol-driven artifacts.
The departure from expected trajectories was confirmed by examination of two additional lifespan datasets beyond the HCP cohorts.
The artifactual departure included partial inversion of expected lifespan patterns of functional network organization.
These distortions were described as 'qualitatively altering estimated lifespan trajectories' rather than merely adding noise.
Without appropriate harmonization, combining HCP cohorts can result in 'biologically misleading inferences about brain development and aging.'
Results
Harmonization approaches accounting for protocol and scanner-model differences substantially reduce artifactual differences in brain network measures.
Harmonization that accounted for both acquisition protocol and scanner-model differences was evaluated.
These approaches 'substantially lessen' the artifactual differences in brain network measures across cohorts.
The harmonization was found to restore network metrics closer to expected lifespan trajectory patterns.
The authors frame harmonization as a best practice for valid inference in multi-cohort lifespan neuroscience research.
Discussion
Small acquisition differences between neuroimaging cohorts can qualitatively bias resting-state-derived network metrics, including partially inverting expected lifespan patterns.
The paper describes the effect as not merely introducing noise but 'qualitatively altering estimated lifespan trajectories of functional network organization.'
Expected lifespan patterns were described as being 'partially inverted' by the artifactual signal differences.
The authors emphasize that even 'modest protocol differences between cohorts' can have 'outsized impacts on the field of neuroscience research' given HCP's widespread usage.
The findings advance 'best practices for valid inference in multi-cohort lifespan neuroscience research.'
What This Means
This research examined a widely used set of brain imaging datasets from the Human Connectome Project (HCP), which includes data from three groups of people spanning development, young adulthood, and aging. The researchers discovered that the young adult group (HCP-YA) had systematically lower image signal quality compared to the other two groups, not because of biological differences, but because of changes in the MRI scanner hardware and scanning procedures used when that data was collected. This difference in signal quality then rippled through all the brain network measurements derived from those scans, affecting measures of how well-organized and segregated brain networks are—metrics that scientists commonly use to track how the brain changes across the lifespan.
The practical consequence is significant: when scientists combine these three HCP datasets to study how the brain develops and ages, the lower signal quality in the young adult group creates a false dip or distortion in the lifespan curve that doesn't reflect true biology. In some cases, this artifact actually reversed the expected direction of brain changes across age. The researchers confirmed this was an artifact—rather than a real biological finding—by comparing against two other independent lifespan brain imaging datasets that showed the expected patterns. They also tested harmonization methods (statistical techniques to correct for known differences in scanning protocols), finding that these approaches substantially reduced the artificial distortions.
This research suggests that when neuroscientists combine brain imaging data collected with different equipment or protocols—even data from a highly respected and carefully collected resource like HCP—they need to carefully account for technical differences between datasets. Failing to do so can lead to incorrect conclusions about how the brain develops and ages. The study provides practical guidance on which harmonization strategies can help, and serves as a cautionary example for the broader field of lifespan neuroimaging research.
Chan M, Han L, Wig G. (2026). Systematic fMRI signal differences across cohorts alter lifespan trajectories of functional brain networks.. Imaging neuroscience (Cambridge, Mass.). https://doi.org/10.1162/IMAG.a.1338