Riemannian geometry boosts functional near-infrared spectroscopy-based brain-state classification accuracy.

Area of Science:

• Neuroscience
• Biomedical Engineering
• Signal Processing

Background:

• Functional near-infrared spectroscopy (fNIRS) is a popular, non-invasive brain imaging technique due to its portability and movement robustness.
• However, fNIRS has limitations in spatial resolution, coverage, and penetration depth compared to fMRI.
• Current fNIRS brain-state classification methods lag behind those using fMRI due to fewer methodological advancements.

Purpose of the Study:

• To develop and evaluate a novel classification approach for fNIRS data using Riemannian geometry.
• To leverage temporal and spatial channel relationships and the dual nature of hemoglobin signals in fNIRS.
• To enhance the accuracy of brain-state classification from fNIRS signals.

Main Methods:

• A classification approach based on Riemannian geometry was applied to kernel matrices derived from fNIRS data.
• Compared different kernel matrix estimators and classifiers (Riemannian Support Vector Classifier, Tangent Space Logistic Regression).
• Benchmarked against traditional feature extraction methods in eight-choice and two-choice brain-state classification tasks.

Main Results:

• The Riemannian geometry approach achieved 65% mean accuracy in eight-choice classification, outperforming traditional methods (42%).
• Achieved 96% average accuracy in two-choice classification across all task combinations, significantly better than traditional models (78%).

Conclusions:

• The proposed Riemannian geometry-based classification is a powerful and viable method for fNIRS data.
• This approach substantially increases accuracy for both binary and multi-class classification of brain activation patterns.
• This work represents a significant advancement in fNIRS data analysis and brain-state classification.