A Between-Subject fNIRS-BCI Study on Detecting Self-Regulated Intention During Walking

Area of Science:

• Neuroscience
• Biomedical Engineering
• Rehabilitation Technology

Background:

• Most brain-computer interface (BCI) studies focus on resting-state motion intention detection.
• Dynamic regulation of motion states, crucial for real-life activities, is understudied in BCI research.
• Existing within-subject BCI methods require extensive, time-consuming training for new users.

Purpose of the Study:

• To develop an adaptable method for discriminating dynamic gait-adjustment intentions in real-time.
• To address the limitations of current BCI approaches by enabling cross-subject applicability.
• To explore the feasibility of using functional near-infrared spectroscopy (fNIRS) for decoding complex motor intentions.

Main Methods:

• Collected cerebral hemoglobin signals from 30 subjects using fNIRS technology.
• Applied mathematical morphology filtering and Entropy Weight Method (EWM) for signal preprocessing.
• Utilized Gradient Boosting Decision Tree (GBDT) for intention onset detection and a 2-layer Genetic Algorithm-Support Vector Machine (GA-SVM) for classifying four types of gait adjustments.

Main Results:

• GBDT achieved an average Area Under Curve (AUC) of 0.894 for detecting intention onset.
• The 2-layer GA-SVM model significantly improved classification accuracy from 70.6% to 84.4% (p=0.005).
• Pseudo-online testing demonstrated high performance with AUC=0.883, True Positive Rate=97.5%, False Positive Rate=0.11%, and Detection Latency=-0.52 ± 2.57 seconds.

Conclusions:

• Decoding dynamic gait-adjustment intentions from motion states using fNIRS is feasible across different subjects.
• The proposed method demonstrates strong adaptability and potential for practical BCI applications.
• This research paves the way for advanced fNIRS-based BCIs to control walking-assistive devices effectively.