Comparing fNIRS signal qualities between approaches with and without short channels

Abstract

Functional near-infrared spectroscopy (fNIRS) is a non-invasive technique used to measure changes in oxygenated (HbO) and deoxygenated (HbR) hemoglobin, related to neuronal activity. fNIRS signals are contaminated by the systemic responses in the extracerebral tissue (superficial layer) of the head, as fNIRS uses a back-reflection measurement. Using shorter channels that are only sensitive to responses in the extracerebral tissue but not in the deeper layers where target neuronal activity occurs has been a 'gold standard' to reduce the systemic responses in the fNIRS data from adults. When shorter channels are not available or feasible for implementation, an alternative, i.e., anti-correlation (Anti-Corr) method has been adopted. To date, there has not been a study that directly assesses the outcomes from the two approaches. In this study, we compared the Anti-Corr method with the 'gold standard' in reducing systemic responses to improve fNIRS neural signal qualities. We used eight short channels (8-mm) in a group of adults, and conducted a principal component analysis (PCA) to extract two components that contributed the most to responses in the 8 short channels, which were assumed to contain the global components in the extracerebral tissue. We then used a general linear model (GLM), with and without including event-related regressors, to regress out the 2 principal components from regular fNIRS channels (30 mm), i.e., two GLM-PCA methods. Our results found that, the two GLM-PCA methods showed similar performance, both GLM-PCA methods and the Anti-Corr method improved fNIRS signal qualities, and the two GLM-PCA methods had better performance than the Anti-Corr method.

Conflict of interest statement

Dr. Litovsky discloses that she is a consultant for Frequency Therapeutics; however, this does not alter our adherence to PLOS ONE policies on sharing data and materials. Other authors have no conflicts of interest to disclose.

Figures

Fig 1. Short-channel setting and fNIRS montage for data collection.

Panel (A) shows the light source and paired short-channel or regular detectors. Panel (B) shows a schematic diagram of light sources and detectors connections and light travelling through the brain tissue. Pairing a light source (in red) and a regular detector (in blue) at 30 mm distance provides a regular fNIRS channel, which measures response from the local region between them, including from the shallower extracerebral tissue and deeper cerebral tissue. Pairing a short channel detector (in green) at 8 mm distance from the light source provides a short channel, which measures signals only from the extracerebral tissue. Panel (C) shows the montage of optodes for fNIRS data collection. Red dots and blue dots are for light sources and regular detectors, respectively. The lines between them and labeled circles indicate the clusters of channels into regions of interest (ROIs). Green dots with green lines connected to the light sources (red) are for the short-channel detectors. Panel C was generated with software NIRSite and then adapted.

Fig 2. Diagram of fNIRS signal processing and systemic response reduction.

SCI is for scalp coupling index. ‘d’ is for the distance between source and detector in an fNIRS channel.

Fig 3. Scalp coupling index (SCI) in the short channels and the number of channels involved for data analysis.

Panel A plots the mean (circles) and standard deviation (bars) of SCI values across three sessions for the 8 short channels, i.e., S1—S8. The magenta horizontal dash line denotes the cut-off threshold (SCI = 0.15); channels that showed SCI less than the threshold were excluded from further analysis. Panel B shows the histogram of numbers of regular (right) and short channels (left) that were included in the GLM-PCA methods for each session of data collection among participants.

Fig 4. Short-channel responses.

Mean (solid lines) and standard error of means (SEM; shaded areas) responses are shown for ΔHbO in red and ΔHbR in blue. The two black vertical dash lines denote the onset and offset of stimulation, which had a duration of 13.6 s. The value 0 (horizontal dash lines) was the change in response relative to baseline activity, i.e., the average of responses during the 5 s prior to the time 0. S1 –S4 are for the 4 short channels on the left hemisphere, and S5 –S8 are for the short channels in the symmetric positions in the right hemisphere.

Fig 5. Block-average ΔHbO responses in 12 regions of interest (ROIs).

The block-average results, i.e., group means (solid lines) and standard error of means (SEM, shaded areas) of ΔHbO responses without further reducing the systemic responses (Original, red) and after applying the Anti-Corr (purple), the GLM-PCA (with stimulus-related regressors and 2 PCs, yellow), and the GLM-PCA2 (with the 2 PCs as regressors, brown) methods are plotted. The two vertical black dash lines plot the onset and offset of stimuli, with a duration of 13.6 s. Responses of zero are relative to the average of 5-s baseline.

Fig 6. Contrast-to-noise ratios (CNRs) from different methods.

Violin plots show the CNRs from HbO responses before (Original, red) and after applying Anti-Corr (purple) or GLM-PCA method (yellow), or GLM-PCA2 (brown) in individual participants in 12 regions of interest (ROIs).