Real-Time Reduction of Task-Related Scalp-Hemodynamics Artifact in Functional Near-Infrared Spectroscopy with Sliding-Window Analysis

Abstract

Functional near-infrared spectroscopy (fNIRS) is an effective non-invasive neuroimaging technique for measuring hemoglobin concentration in the cerebral cortex. Owing to the nature of fNIRS measurement principles, measured signals can be contaminated with task-related scalp blood flow (SBF), which is distributed over the whole head and masks true brain activity. Aiming for fNIRS-based real-time application, we proposed a real-time task-related SBF artifact reduction method. Using a principal component analysis, we estimated a global temporal pattern of SBF from few short-channels, then we applied a general linear model for removing it from long-channels that were possibly contaminated by SBF. Sliding-window analysis was applied for both signal steps for real-time processing. To assess the performance, a semi-real simulation was executed with measured short-channel signals in a motor-task experiment. Compared with conventional techniques with no elements of SBF, the proposed method showed significantly higher estimation performance for true brain activation under a task-related SBF artifact environment.

Short-channel regression in functional near-infrared spectroscopy is more effective when considering heterogeneous scalp hemodynamics

“…With respect to the incongruent assumptions on the spatial distribution of scalp hemodynamics (heterogeneous 9,17,28,[31][32][33][34][35][36][37][38] versus homogeneous 22,26,27,29,30 ) reported in literature for fNIRS experiments, we found evidence for a location-specific behavior of the scalp hemodynamics. This supports the hypothesis of heterogeneity.…”

“…70 Alternatively, the phase shifts of the MW signals could be calculated for a baseline measurement 71 when applied with a sliding window approach. 30 Real-time algorithms may further improve regression performance because of the nonstationary and nonlinear nature of SS signals. 18,72 The hardware employed provided SS channels next to the sources but not next to the detectors.…”

“…As a consequence, it is often assumed that global noise is distributed homogeneously over the entire scalp layer and thus that a single global signal can be used for the short-channel regression. [26][27][28][29][30] In contrast, it was repeatedly shown that scalp hemodynamics follow spatially heterogeneous patterns, 9,17,[31][32][33][34][35][36][37][38] i.e., different short-channel signals are measured at different locations on the scalp. An additional complexity is observed when considering that the physiological artifacts also show frequency-dependent spatial variations, 36 with a noteworthy effect observed for the MW-frequency band (i.e., average time lags up to 2 s between ipsilateral head regions 36 ).…”

fNIRS-Based Upper Limb Motion Intention Recognition Using an Artificial Neural Network for Transhumeral Amputees

“…Furthermore, remaining muscle sites for the prosthetic control are not physiologically identified to the distal arm functions [ 12 ]. In the past few years, the brain-machine interface (BMI) has appeared as a potential alternative that can offer an incredible opportunity to amputated individuals by empowering them to play out their daily routine [ 13 , 14 ]. It evades the muscles intentions.…”

“…Henceforth, sensing the light pass through the brain tissue employing noninvasive signal acquisition is performed using an optical source/detector pair, which is placed on the scalp. A relative change in the concentration of HbO and HbR indicates neuronal activation relevant to the executed motion [ 13 ]. The attained brain responses relative to distinct motion may comprise of noises that pollute these recorded signals.…”