High-Density Functional Near-Infrared Spectroscopy and Machine Learning for Visual Perception Quantification.

Area of Science:

• Neuroscience
• Biomedical Engineering
• Human-Computer Interaction

Background:

• Wearable sensors generate crucial data for monitoring metrics.
• Functional near-infrared spectroscopy (fNIRS) enables non-intrusive monitoring of human visual perception.
• Quantifying visual perception via fNIRS has potential applications in engineering.

Purpose of the Study:

• To design experimental procedures to induce and quantify visual perception changes.
• To correlate visual task performance with hemodynamic responses measured by high-density fNIRS (HfNIRS).
• To explore the application of machine learning in analyzing HfNIRS data for visual perception assessment.

Main Methods:

• Utilized HfNIRS to record total hemoglobin (Hbt), hemoglobin (Hb), and oxygenated hemoglobin (HbO2) during simulated driving tasks with visual variations.
• Employed channel dimensionality reduction (mutual information), feature extraction (statistical measures), and K-Nearest Neighbors (KNN) for task classification.
• Correlated visual task scores with Hbt, Hb, and HbO2 fluctuations.

Main Results:

• HfNIRS successfully recorded hemodynamic changes corresponding to visual perception alterations.
• The KNN algorithm achieved high classification accuracy (96.3% ± 1.99%) for different visual tasks.
• Higher visual task scores correlated with more significant fluctuations in Hbt, Hb, and HbO2.

Conclusions:

• Changes in visual perception demonstrably trigger alterations in Hbt, Hb, and HbO2.
• HfNIRS, combined with machine learning, offers an effective method for quantifying visual perception.
• Further research is needed to refine the mathematical relationship between HfNIRS signals and visual perception for quantitative analysis.