Resting‑State fNIRS Complexity Graphs Identify Prefrontal Network Biomarkers at the MCI Stage

A new Resting‑State fNIRS Complexity Graphs study reports a prefrontal resting-state framework that builds interpretable, subject-level graphs from coordinated fluctuations in nonlinear signal complexity, presenting the approach as a network-biomarker strategy for Alzheimer’s disease stage assessment.

In the authors’ report, statistically supported prefrontal network biomarkers distinguished mild cognitive impairment (MCI) from healthy aging, with a group difference reported as p = 0.001. The narrative further places these network-level biomarkers as most consistently expressed at the MCI stage rather than in later-stage disease.

The investigators describe representing resting-state prefrontal fNIRS signals as individual-subject graphs, with edges capturing coordinated fluctuations of nonlinear signal complexity across channels. As reported, edge weights came from sliding-window computations intended to quantify co-fluctuations in signal-complexity measures, translating time-varying complexity behavior into a network representation. In this description, the central modeling object is an individualized graph meant to encode coordinated complexity fluctuations across the prefrontal montage, rather than a single channel-wise feature or a static connectivity matrix. The authors present this per-subject network representation as the substrate for subsequent learning and biomarker extraction.

For analysis, the study reports applying graph neural networks (GNNs) to detect disease-stage-related network patterns from these subject graphs, and pairing model outputs with edge-level interpretability or importance measures to localize influential connections. To address reproducibility, the investigators describe fold-wise stability assessment across cross-validation, along with procedures to aggregate repeatedly selected connections into a summary representation reflecting consistent edge selection across folds. These steps are presented as linking classification-oriented modeling with network feature attribution, so results can be described in terms of stable, interpretable edges rather than only a stage label.

In the reported analyses, the authors state that the complexity–fluctuation graph representation outperformed conventional amplitude-based functional connectivity, framing the difference as tied to coordinated complexity dynamics rather than amplitude coupling alone. Alongside this comparison, the investigators describe a consensus network analysis (consensus network) within their reproducibility evaluation that highlighted a subset of prefrontal connections repeatedly selected across folds, and they report that attention- or importance-pattern maps spatially corresponded with statistically derived biomarkers.

When contrasting stages, the authors also report that Alzheimer’s disease–stage patterns were more heterogeneous and less consistently expressed than those observed at the MCI stage, and they emphasize a focus on coordinated complexity dynamics over prioritizing classification accuracy, supporting the potential for longitudinal monitoring and clinically applicable fNIRS-based assessment of neurodegenerative disease.

Taken together, the article links the reported MCI-stage consistency and AD-stage heterogeneity to a staging narrative centered on a transitional, more reproducibly expressed network alteration pattern.

Key Takeaways:

• The study describes subject-level prefrontal graphs built from sliding-window co-fluctuations of nonlinear signal complexity, and reports that these network biomarkers separated MCI from healthy aging.
• As reported, fold-wise stability assessments and edge-level importance/consensus summaries were used to support reproducible and interpretable identification of influential prefrontal connections.
• The authors describe MCI-associated patterns as more consistently expressed than AD-stage patterns and frame coordinated complexity dynamics as potentially relevant to longitudinal monitoring, without presenting the approach as clinically ready.