Neuroimaging Biomarkers for Anxiety Using fNIRS and Machine Learning

Sibo Huang's dataset contains resting-state functional near-infrared spectroscopy (fNIRS) data from ninety participants, including 50 anxiety patients and 40 healthy controls. The data was used to train a Gradient Boosting Decision Tree (GBDT) model for classification, achieving an AUC of 0.900. The dataset was last updated on figshare in April 2026.

Use Cases

• Train machine learning classifiers for anxiety diagnosis based on hemodynamic patterns.
• Identify cortical region biomarkers for anxiety using oxyhemoglobin fluctuation features.
• Compare model performance (e.g., GBDT vs. RF) for neuroimaging-based classification tasks.
• Perform interpretability analysis (e.g., SHAP) on neuroimaging features to understand predictive regions.

Strengths

• Includes data from 90 participants (50 anxiety patients, 40 controls), providing a balanced sample.
• Model performance metrics are reported (AUC: 0.900, sensitivity: 0.921, specificity: 0.709).
• Specific predictive cortical regions (e.g., prefrontal, sensorimotor-visual areas) are identified.

Limitations

• Column-level documentation is absent; field semantics must be inferred after download.
• Row count is unknown, which may limit suitability assessment.
• The dataset is small (79.6 KB), indicating limited scope.

Provenance

Source

figshare

Collection Method

Data collected from participants via demographic questionnaires, GAD-7 assessments, and 5-minute resting-state fNIRS recordings.

Freshness

Last updated 2026-04-10 06:03:27