Temporal hemodynamic classification of two hands tapping using functional near-infrared spectroscopy
Nguyen Thanh Hai1, Ngo Q Cuong, Truong Q Dang Khoa
1Biomedical Engineering Department, International University of Vietnam National Universities in Ho Chi Minh City Ho Chi Minh City, Vietnam.
Frontiers in Human Neuroscience
|September 14, 2013
Summary
This study introduces a novel algorithm using Near-Infrared Spectroscopy (NIRS) to recognize left or right hand tapping. The method effectively decodes hand movements by analyzing brain oxygenation changes, demonstrating high accuracy in experimental trials.
Area of Science:
- Cognitive Neuroscience
- Biomedical Engineering
- Signal Processing
Background:
- Non-invasive brain imaging technologies like NIRS, MRI, and EEG are crucial in neuroscience.
- Near-Infrared Spectroscopy (NIRS) offers a convenient method for experimental brain activity monitoring.
- Changes in oxygenated hemoglobin (oxy-Hb) concentration correlate with brain activity and can predict behavior.
Purpose of the Study:
- To develop and evaluate a recognition algorithm for distinguishing between left hand (LH) and right hand (RH) tapping using NIRS data.
- To explore the utility of oxy-Hb concentration changes as predictive features for hand-tapping recognition.
- To compare the performance of Support Vector Machines (SVM) and Artificial Neural Networks (ANNs) for this classification task.
Main Methods:
- Multi-channel NIRS data from hand tapping tasks were collected and pre-processed using a Savitzky-Golay filter to reduce noise.
- Polynomial regression (PR) was employed to extract signal characteristics, represented by polynomial coefficients related to oxy-Hb concentration.
- SVM and ANNs were utilized as machine learning models to classify LH versus RH tapping based on the extracted features.
Main Results:
- The Savitzky-Golay filter effectively smoothed noisy NIRS data, enabling better feature extraction.
- Polynomial regression coefficients, indicative of oxy-Hb changes, served as effective features for hand-tapping recognition.
- Both SVM and ANNs demonstrated the capability to accurately recognize hand-tapping side using the derived features, with experimental validation on multiple trials.
Conclusions:
- The proposed method effectively utilizes NIRS-based oxy-Hb concentration changes for accurate hand-tapping recognition.
- The combination of signal pre-processing, feature extraction via polynomial regression, and machine learning (SVM/ANNs) provides a robust approach.
- This study highlights the potential of NIRS for developing brain-computer interfaces and decoding motor intentions.
