Load-dependent relationships between frontal fNIRS activity and performance: A data-driven PLS approach

Abstract

Neuroimaging research frequently demonstrates load-dependent activation in prefrontal and parietal cortex during working memory tasks such as the N-back. Most of this work has been conducted in fMRI, but functional near-infrared spectroscopy (fNIRS) is gaining traction as a less invasive and more flexible alternative to measuring cortical hemodynamics. Few fNIRS studies, however, have examined how working memory load-dependent changes in brain hemodynamics relate to performance. The current study employs a newly developed and robust statistical analysis of task-based fNIRS data in a large sample, and demonstrates the utility of data-driven, multivariate analyses to link brain activation and behavior in this modality. Seventy participants completed a standard N-back task with three N-back levels (N = 1, 2, 3) while fNIRS data were collected from frontal and parietal cortex. Overall, participants showed reliably greater fronto-parietal activation for the 2-back versus the 1-back task, suggesting fronto-parietal fNIRS measurements are sensitive to differences in cognitive load. The results for 3-back were much less consistent, potentially due to poor behavioral performance in the 3-back task. To address this, a multivariate analysis (behavioral partial least squares, PLS) was conducted to examine the interaction between fNIRS activation and performance at each N-back level. Results of the PLS analysis demonstrated differences in the relationship between accuracy and change in the deoxyhemoglobin fNIRS signal as a function of N-back level in eight mid-frontal channels. Specifically, greater reductions in deoxyhemoglobin (i.e., more activation) were positively related to performance on the 3-back task, unrelated to accuracy in the 2-back task, and negatively associated with accuracy in the 1-back task. This pattern of results suggests that the metabolic demands correlated with neural activity required for high levels of accuracy vary as a consequence of task difficulty/cognitive load, whereby more automaticity during the 1-back task (less mid-frontal activity) predicted superior performance on this relatively easy task, and successful engagement of this mid-frontal region was required for high accuracy on a more difficult and cognitively demanding 3-back task. In summary, we show that fNIRS activity can track working memory load and can uncover significant associations between brain activity and performance, thus opening the door for this modality to be used in more wide-spread applications.

Keywords: Cognitive load; N-back task; Neural efficiency; Partial least squares; Working memory; fNIRS.

Fig. 1.

N-back Task. Example of 1-back task (Top) and 3-back (Bottom). 2-back task not shown.

Fig. 2.

fNIRS Montage in international 10–10 coordinate space. Montage with 16 × 16 frontal source-detector pairs and 4 × 4 right parietal source-detector pairs. Sources are indicated in red, detectors are indicated in gray, and channels are indicated by purple lines. Cz highlighted in green.

Fig. 3.

Gross ROI depth maps with superimposed montage. fNIRS montage (registered to Colin27 atlas) and depth map for 6 ROIs taken from the talairach daemon parcellation: Left and Right Inferior Frontal Gyrus, Medial Superior Frontal Gyrus, Medial Orbitofrontal Cortex, Right Superior Parietal Gyrus, Right Inferior Parietal Gyrus.

Fig. 4.

Boxplots of Average Accuracy & RT by N-back Level for all participants.

Fig. 5.

Group Average HRF plots by N-back level. Group-level results as an HRF time series, averaged across 5 general ROIs with 8–10 channels each. Color-coded montage in lower right indicates the channels included in each ROI.

Fig. 6.

N-back level contrasts for HbO (left) and HbR (right). Only significant channels (q < 0.05) are shown. Channels are displayed on top of 10–20 coordinates and grayscale depth maps for left and right Inferior Frontal Gyri, medial Superior Frontal Gyri, and right Superior and Inferior Parietal Gyri. For HbO contrasts, positive t-values (red) correspond to relatively larger activity for the first term in the contrast, and negative t-values (blue) correspond to larger activity for the second term. The opposite pattern applies to HbR contrasts.

Fig. 7.

LV 1 demonstrated an N-back load-dependent relationship between changes in deoxyhemoglobin concentrations (HbR) and performance. (A) The left panel shows correlation between accuracy and HbR concentration change separately by N-back level. Error bars are 95% confidence intervals around the mean correlation. The right panel shows channels (labeled by number), which had bootstrap ratios (BSR) > |2|. (B) Scatterplots showing the correlation between HbR (β for task-evoked change from baseline) and performance (accuracy) at 4 channels with BSRs > 3, separated by N-back level.