Enhancing classification accuracy of HRF signals in fNIRS using semi-supervised learning and filtering

Area of Science:

• Neuroscience
• Biomedical Engineering
• Signal Processing

Background:

• Functional near-infrared spectroscopy (fNIRS) measures brain activity by detecting changes in blood oxygenation.
• Hemodynamic response function (HRF) signals, reflecting oxyhemoglobin (oxyHb) and deoxyhemoglobin (deoxyHb) variations, are crucial for interpreting fNIRS data.
• Accurate classification of HRF signals is essential for advancing cognitive neuroscience and brain-computer interfaces.

Purpose of the Study:

• To introduce a novel semi-supervised learning (SSL) approach combined with a filtering technique to enhance HRF signal classification accuracy.
• To preprocess and analyze HRF signals from the prefrontal cortex acquired via fNIRS.
• To validate the effectiveness of the proposed filtering method in improving classification performance.

Main Methods:

• Utilized a semi-supervised learning (SSL) framework for HRF signal classification.
• Implemented a novel filtering technique for preprocessing fNIRS-acquired HRF data.
• Collected HRF signals from the prefrontal cortex in response to odor stimuli and air state.
• Extracted features from filtered HRF signals for model training.

Main Results:

• The classification model demonstrated significantly improved accuracy when trained on filtered and feature-extracted HRF signals.
• The proposed filtering technique proved effective in enhancing the performance of the SSL algorithm.
• Comparative analysis confirmed the superiority of the algorithm post-filtering.

Conclusions:

• The combined SSL and filtering approach offers a robust method for improving HRF signal classification in fNIRS.
• This advancement has significant implications for functional brain imaging, cognitive studies, and understanding brain responses.
• The findings pave the way for more precise and reliable analysis of neuroimaging data.