Interpretable Network-Level Biomarker Discovery for Alzheimer's Stage Assessment Using Resting-State fNIRS Complexity...

Area of Science:

• Neuroscience
• Biomedical Engineering
• Data Science

Background:

• Resting-state functional near-infrared spectroscopy (fNIRS) is valuable for Alzheimer's disease (AD) assessment.
• Existing fNIRS methods often lack network-level insights and reproducibility.
• Coordinated network dynamics are crucial for understanding neurodegenerative processes.

Purpose of the Study:

• To develop a reproducible and interpretable graph-based framework for resting-state fNIRS.
• To enable network-level biomarker discovery for Alzheimer's disease (AD) and mild cognitive impairment (MCI).
• To move beyond static, channel-wise analyses towards dynamic network assessments.

Main Methods:

• Represented resting-state prefrontal fNIRS signals as subject-level graphs.
• Utilized sliding-window analysis to capture nonlinear signal complexity fluctuations for edge computation.
• Employed graph neural networks (GNNs) for network pattern identification and interpretability analysis.

Main Results:

• The complexity-fluctuation graph approach surpassed conventional amplitude-based connectivity.
• Identified statistically significant prefrontal network biomarkers distinguishing MCI from healthy aging (p=0.001).
• Observed more heterogeneous network patterns in AD, with MCI showing more consistent alterations.

Conclusions:

• Established a reproducible and interpretable framework for fNIRS analysis focusing on complexity dynamics.
• Network alterations are most consistently detected at the MCI stage, indicating its significance.
• The framework shows potential for longitudinal monitoring and clinical assessment of neurodegenerative diseases.