Wireless wearable functional near-infrared spectroscopy (fNIRS) has attracted growing attention as a candidate for real-life brain monitoring systems. It is important to determine the onsets at which neuronal activation is evoked by cognitive status in real-time analysis. We propose a machine learning approach for the classification of cognitive event onsets (CogEOs) in hemodynamic signals during three cognitive tasks. The approach does not require a threshold to be set or additional measurement for the rest state. A support vector machine is trained by labeled features obtained from the mean amplitude of hemodynamic changes and then predicts the type of onset points. The problems caused by the imbalance between CogEOs and non-event onsets (NonEO) are solved by oversampling the feature samples labeled by cognitive events. By oversampling, the classification accuracy from an average of five classification scores reaches 74%, 77%, and 75% for the simple arithmetic, 1-back, and 2-back tasks. We achieve the best onset classification performance when the NonEOs are randomly distributed and when the subject is performing the 1-back task. Our study extends fNIRS to real-life applications by detecting the time point when brain activation starts among random observations using machine learning without additional triggers or threshold.

Objective: The purpose of this study is to describe the noise reduction in the hemodynamic responses, obtained by functional near-infrared spectroscopy (fNIRS), using the proposed extended Kalman filter (EKF) with a non-linear state-space model, aided by the short separation (SS) measurement. Methods: The authors used the simulated data by adding the synthetic hemodynamic response function (HRF) to the multi-distance four-channel fNIRS signals obtained during the resting state. EKF was used to estimate the non-linear state-space model designed based on the Balloon model. The SS channel was used as a regressor that is sensitive only to superficial noises. The whole segments were grouped by the existence of motion artifacts (MAs) to investigate the improvement by EKF compared to the linear Kalman filter (LKF) and adaptive filter (AF) in extracting neural-evoked hemodynamic. Results: Kalman-based approaches were better than AF in reducing noises. Using EKF, the averages of the decreased errors and increased correlation between the recovered and true HRF were 34% in oxy-hemoglobin and 62% in deoxy-hemoglobin concentrations in segments having MAs, compared with LKF. In the MA-free condition, EKF is more robust to the poor quality of signals in noise reduction than LKF. Conclusion: The proposed non-linear Kalman approach is better in noise reduction than AF and LKF especially in noisy deoxy-hemoglobin concentrations, and less affected by the conditions of measurements and contaminations by MAs. Significance: The proposed method can be used for reducing superficial noises and MAs from fNIRS signals as an upgraded alternative to existing AFs.

Functional connectivity (FC) is a potential candidate that can increase the performance of brain-computer interfaces (BCIs) in the elderly because of its compensatory role in neural circuits. However, it is difficult to decode FC by the current machine learning techniques because of a lack of physiological understanding. To investigate the suitability of FC in BCIs for the elderly, we propose the decoding of lower- and higher-order FC using a convolutional neural network (CNN) in six cognitive-motor tasks. The layer-wise relevance propagation (LRP) method describes how age-related changes in FCs impact BCI applications for the elderly compared to younger adults. A total of 17 young adults $\left(24.5\pm 2.7 \mathrm{years}\right)$ and 12 older $\left(72.5\pm 3.2 \mathrm{years}\right)$ adults were recruited to perform tasks related to hand-force control with or without mental calculation. The CNN yielded a six-class classification accuracy of 75.3% in the elderly, exceeding the 70.7% accuracy for the younger adults. In the elderly, the proposed method increased the classification accuracy by 88.3% compared to the filter-bank common spatial pattern. The LRP results revealed that both lower- and higher-order FCs were dominantly overactivated in the prefrontal lobe, depending on the task type. These findings suggest a promising application of multi-order FC with deep learning on BCI systems for the elderly.

Functional connectivity (FC) is a potential candidate that can increase the performance of brain-computer interfaces (BCIs) in the elderly because of its compensatory role in neural circuits. However, it is difficult to decode FC by current machine learning techniques because of a lack of its physiological understanding. To investigate the suitability of FC in BCI for the elderly, we propose the decoding of lower- and higher-order FCs using a convolutional neural network (CNN) in six cognitive-motor tasks. The layer-wise relevance propagation (LRP) method describes how age-related changes in FCs impact BCI applications for the elderly compared to younger adults. Seventeen younger (24.5±2.7 years) and twelve older (72.5±3.2 years) adults were recruited to perform tasks related to hand-force control with or without mental calculation. CNN yielded a six-class classification accuracy of 75.3% in the elderly, exceeding the 70.7% accuracy for the younger adults. In the elderly, the proposed method increases the classification accuracy by 88.3% compared to the filter-bank common spatial pattern (FBCSP). LRP results revealed that both lower- and higher-order FCs were dominantly overactivated in the prefrontal lobe depending on task type. These findings suggest a promising application of multi-order FC with deep learning on BCI systems for the elderly.