Cortical hemodynamic response during cognitive Stroop test in acute stroke patients assessed by fNIRS

2.1 Design

This was a sub-study of a prospective observational study conducted at the Stroke Unit, Department of Neurology, Rigshospitalet, which is described elsewhere (clinicaltrials.gov: NCT02111408). Subjects in this sub-study were enrolled from April 2016 to September 2016 after written informed consent. The study was approved by The Regional Ethics Committee in The Capital Region of Denmark on September 29, 2015 (H-2-2013-091), the Danish Data-Protecting Agency (GLO-2013-18; IT suite nr. 02385) and in accordance with the Declaration of Helsinki of 1964 and its later amendments. We report in compliance with STROBE guidelines for observational studies (von Elm et al., 2007).

2.2 Participants

Patients were enrolled during weekdays after being diagnosed clinically with stroke following computed tomography (CT) or magnetic resonance imaging (MRI) to exclude patients with intracranial hemorrhage. Patients were excluded if they did not have the ability to consent, had other significant brain disease, short remaining life-expectancy (months), could not perform the SCWT due to colorblindness or other deficits. Patients were only included if the complete NIRS examination and all other procedures in main study could be performed within 7 days of their index stroke.

A group of healthy sex- and age-matched control subjects (HC) were recruited and examined with NIRS after the same protocol as stroke patients. Only diseases relevant for cerebrovascular disease were viewed as exclusion criteria (diagnosed or initiated treatment of atherosclerotic disease, hypertension, hypercholesterolemia, diabetes, etc.), while diseases in non-relevant organs were allowed (allergies, skin disease, osteoporosis, etc.).

2.3 Examinations

Patients received standard-of-care stroke treatment and examinations during admission including patient history (current and former diagnosed or treated diseases, medications, physical activity, consumption of alcohol and tobacco), neurological examination, blood pressure, body mass index (BMI), CT or MRI of the brain, chest x-ray, ECG, carotid ultrasound and/or CT-angiography, routine blood samples, and at least 48-hour cardiac telemetry. Selected patients also underwent echocardiography and extended blood examination when deemed appropriate.

Outside of the standard examinations, patients underwent the following examinations and classifications:

Index stroke classified according to the Causative Classification of Stroke system (CCS) (Ay et al., 2007).

OPEN IN VIEWER

Fig. 1

The Stroop Color and Word Test applied with screengrabs showing incorrect (top row) and correct stimuli (bottom row).

Participants were placed comfortably and upright positioned in a chair in a silent room with dimmed lights. A 15.4-inch laptop computer placed approximately 70 cm from the participants eyes was then presented showing only a white crosshair on black background for a resting period of five min. Before performing the Stroop task, participants received instructions and did a trial block of each degree of difficulty.

The SWCT was constructed in a block design with three degrees of difficulty (Fig. 1): Neutral, congruent (text and descriptive color-word matching) and incongruent (text and descriptive color-word not matching). Each block was composed by 6 consecutive stimuli of the same degree of difficulty. Stimuli was constituted by two words displayed on separate rows on the black background screen. The top row was written in colored letters whereas the bottom row was written in white letters. The bottom row either correctly or incorrectly denoted the color of the letters in the top row, which participants were instructed to assess by clicking either the left mouse button in the case of correct denotation or right mouse button in case of incorrect denotation (Fig. 1). Participants were asked to answer as quickly and accurately as possible. The maximum response time of five seconds was increased from original test (Schytz et al., 2010) to fit the cognitive abilities of the current cohort. Distribution of correct and incorrect stimuli was 50% of each across the entire test. The top row was presented 100 milliseconds before the bottom row to maintain focus on the colored letters (MacLeod, 1991). In neutral blocks the top row letters were XXXX in contrast to the other blocks that either congruently or incongruently wrote a color-word (GREEN, YELLOW, BLUE, or RED) with the colored letters (Fig. 1).

The inter-block period was 15 seconds minus the response time from the last stimulus from the former block creating varying inter-stimulus periods as recommended (Herold, Wiegel, Scholkmann, & Müller, 2018). During the inter-block period the laptop again presented the crosshair. The order of blocks was the same for every participant with 5 blocks for each degree of difficulty.

2.5 NIRS examination

Relative changes in Oxy-Hb and Deoxy-Hb were measured with using continuous wave NIRS (CW6, TechEn Inc., Milford, Massachusetts, USA) at a sampling rate of 200 Hz. Infrared light sources of two wavelengths (690 nm and 830 nm) were placed laterally and medially on each side of the forehead with detectors distanced 3.5 cm in between the light sources together comprising a rhomb over each prefrontal cortex while carefully avoiding the midline sinus. Sensitivity mapping of the probe configuration on the average head size in the current cohort (Fig. 2) was generated in AtlasViewer (Aasted et al., 2015). Channels were named accordingly (superomedial, superolateral, inferomedial and inferolateral PFC). To enable short-separation regression by filtering out extracerebral signals and thus increasing brain sensitivity, four detector optodes were placed in close proximity to the light sources (eight mm). Both light sources and detectors were arranged in an elastic band tightened firmly around the participants head to avoid movement of optodes and ensure the best possible signal.

OPEN IN VIEWER

Fig. 2

Sensitivity map of participant with head size average to all participants.

Data was preprocessed using the HOMER3 toolbox (Huppert, Diamond, Franceschini, & Boas, 2009) in MATLAB R2018b (Mathworks, Massachusetts, USA). Channels were pruned automatically with a signal-to-noise threshold of five (enPruneChannels function) and manually thorough visual quality control. Raw intensity signals were downsampled to 20 Hz before conversion to optical density (hmrIntensity2OD function). Motion artefacts (MA) was then detected channel by channel with a standard deviation threshold of 10 (within tMotion = 0.5 seconds and marking the neighboring period as MA, tMask = 1 seconds), which fitted most accurately to investigators visual assessment of MA across all subjects. Both Spline interpolation (p = 0.99), targeted principal component analysis (tPCA, nSV = 0.97 and maximum iterations of 5) and wavelet transformation (iqr = 1.5) was employed separately and compared. No smoothing filters were applied. Correlation-Based Signal Improvement-method was disregarded beforehand as it relies on the assumption of a constant relationship between Oxy-Hb and Deoxy-Hb, which is not always met under cerebral pathology as in the current study population (Obrig & Steinbrink, 2011). Hybrid methods of more than one MA correction were deemed too invasive. After correction MA detection was processed once again to assess efficiency of correction methods and to reject any stimuli in the vicinity of the remaining MA (–2 to 15 seconds). MA correction with tPCA resulted in the fewest stimuli rejections and was chosen going forward. All subjects had at least 3 blocks of each stimulus. Targeted PCA performs a principal component analysis in the areas defined as MA and thus ignores the most extreme signal outliers. Optical density was then converted to concentrations of oxygenated (Oxy-Hb) and deoxygenated (Deoxy-Hb) hemoglobin with partial pathlength factor of 1 to indicate the uncertainty of travelled light distance and instead report hemoglobin changes in units of micromolar x mm. Lastly, the general linear model (GLM) was calculated using the hmrDeconvHRF_DriftSS function with a time frame of –2 to 20 seconds around stimulus start. The function applies short-separation regression using the channel with the greatest correlation and a 3rd order polynomial drift correction. The temporal basis function was a consecutive sequence of Gaussian functions defined by a width of 1.5 seconds and temporal spacing of 1.5 seconds to give the GLM temporal flexibility without giving outlying responses excessive influence. The least squares method was used to solve the GLM (Barker, Aarabi, & Huppert, 2013).

Hemodynamic response function (HRF) was then exported to MATLAB and analyzed individually channel by channel. Peak was defined as maximum increase or decrease from baseline with the corresponding time from stimulus onset defining time-to-peak and the average slope was calculated as ratio between peak and time-to-peak.

To examine the hemodynamic Stroop effect, the difference between the parameters of hemodynamic response (peak, average slope) from incongruent and congruent stimuli (I-C) was analyzed. Lateralization was assessed by the difference in peak during incongruent stimuli between contralateral channels.

2.7 Statistical analysis

Power calculations based on results from the most comparable study (M. L. Schroeter et al., 2005) revealed that a sample size of 40 participants would be sufficient to show group differences in almost every PFC channel.

Categorical data are reported as numbers and percentages. Continuous data of non-normal distribution are reported as median and interquartile range (IQR) and as mean and standard deviation (SD) in case of normal distribution. Group differences were tested with Fisher’s exact test or Chi-square test for categorical data and Welch unpaired t-test for normally distributed continuous data. In cases of ordinal or non-normal continuous data the Wilcoxon-Mann-Whitney test or Kruskal-Wallis test was applied depending on the number of independent variables. Associations were assessed by Spearman’s rank rho correlation when data was non-normally distributed.

All statistical analyses were performed using RStudio version 1.4.1106 (Boston, MA, USA). Overall statistical significance level of 0.05 was chosen with Bonferroni correction for multiple comparisons by the number of channels for fNIRS data.

3.1 Baseline characteristics

The baseline characteristics of both stroke patients and HC disclosed no significant group differences concerning age and sex (Table 1). Patients showed higher BMI, less physical activity and more smoking than the HC group.

The stroke patients’ disease characteristics concerning medical history and index stroke are presented in Table 2. Patients experienced mild to moderate stroke symptoms according to NIHSS and mRS at enrollment and most patients had hemispherical infarction in either MCA or PCA territory with a slight shift towards left hemispherical infarction (57.1%). All patients were examined median 4 days, IQR 3–5 days after their index stroke. CCS etiology classification showed most patients experienced their stroke due to small-artery occlusion. Stroke patients were dichotomized to almost equal groups with SVD scores of 0 (47.4%) and 2–4 (42.1%). Only 4 patients (19.0%) were treated with rTPA. Incidence of cognitive impairment (Table 3) ranged from 9.5% to 23.8% depending on the test, while accumulated impairment across cognitive tests was higher (38.1%).

Stroke patients had longer response time and higher error percentage in all stimuli conditions except during neutral stimuli (Table 4).

The SCWT was successful in showing the Stroop effect with greater response time and error percentage in the incongruent tasks. While the congruent tasks were significantly harder than the neutral tasks for all subjects regarding response time, the error percentage did not increase.

Performance in Stroop task with stroke patients dichotomized to a SVD score of ≥2 or 0 are displayed in Table 5. Patients with higher SVD burden performed equally to stroke patients without small vessel disease.

Peak values of all channels and chromophores of stroke patients and HC are shown in Table 6. Stroke patient trended towards activation in right inferomedial PFC during congruent stimuli as opposed to HC, but no channel exhibited the classic activational pattern on the group level. However, superolateral PFC channels in both hemispheres show significant increase in Deoxy-Hb especially for congruent and incongruent stimuli in healthy subjects. Stroke patients showed a corresponding response, which did not reach significance level after Bonferroni correction. Healthy subjects exhibited a similar pattern in superomedial channels. Mean Oxy-Hb peaks in the corresponding channels were negative values though not significantly differing from zero.

The average slope values of stroke patients and HC (Table 7) showed that stroke patients exhibited classic activation pattern in the right inferomedial PFC channels during congruent stimuli, while no other channel in any groups or conditions did. The pattern from Table 6 was otherwise largely repeated.

Spearman’s rank correlations showed no associations between hemoglobin peaks and response time during congruent and incongruent stimuli (Table 8) outside of a positive association in left inferolateral PFC between Deoxy-Hb and response time during congruent stimuli, but only in stroke patients.

The increase in Deoxy-Hb peak values was significantly less in stroke patients than HC especially in right superomedial PFC during neutral and congruent stimuli and in right inferomedial PFC during neutral stimuli, while the same tendency did not reach significance level after Bonferroni correction in right superolateral and left superomedial PFC during neutral stimuli.

Average Deoxy-Hb slope were lower for stroke patients than HC in right superolateral channels during neutral stimuli, but not significantly so in more difficult conditions. No other group differences reached significance level, but Deoxy-Hb slope in right superomedial PFC showed the same pattern across all conditions.

3.5 Group comparisons of hemodynamic Stroop effect

Further group comparisons of hemodynamic Stroop effect were limited to channels exhibiting significant changes in activational analysis (excluding inferolateral channels). Group comparisons of differences between peak and average slope corresponding to the Stroop effect (response during congruent stimuli subtracted from response during incongruent stimuli) are presented in Tables 9–11. No hemodynamic Stroop effect and no group differences was shown in any channel.

Comparing Deoxy-Hb incongruent to congruent peak difference in superolateral and superomedial PFC between stroke patients with SVD score of ≥2 to HC and stroke patients with SVD score of 0, indicated that accounting for SVD score could be desirable in future studies.

We found no lateralization in any channels of any group analysis during incongruent stimuli. Analysis did not show any group differences between stroke patients and HC nor when accounting for the affected hemisphere (Table 12).

4.1 Stroop task performance

Although the SCWT is very commonly applied in stroke populations, there are significant differences in the execution of the test. Many SCWTs are performed with a certain time limit and the result being the number of correct answers. This limits our understanding of behavior and physiology between each test condition and as such the Stroop effect.

The SCWT applied in this study showed the Stroop effect as both response time and error percentage showed increments with greater difficulty, though response time was slightly more reliable in line with a previous study (Shao et al., 2020).

Stroke patients in the subacute phase performed the SCWT significantly worse than HC with respect to both response time and error percentage, despite suffering from stroke with heterogenous anatomical lesions and relatively low NIHSS. Response time and error percentage was slightly greater in this study compared to the findings in chronic stroke patients who had more time to recover (5–20 weeks) (Morein-Zamir, Henik, Balas, & Soroker, 2005). While the incidence of CI (assessed by other cognitive tests) was comparable to other acute stroke studies (Blackburn et al., 2013; Jaillard et al., 2009), no cut-off for CI have been established for the applied SCWT.

To explore the importance of small vessel disease in our stroke population, we dichotomized patients according to SVD score quite similar to Schroeter et al. who found significantly increased response time and error percentage in cerebral microangiopathy patients with moderate to severe microangiopathy scored by lacunar infarctions and periventricular white matter lesions (M. L. Schroeter et al., 2007). Our study could not replicate this finding as stroke patients with moderate to severe SVD score (2–4) showed equal response time and error percentage compared to stroke patients with SVD score of 0. A drawback of the SVD scoring system in the current setting is the omission of acute lacunary infarctions in calculation of the SVD score, despite being a clear imaging sign of small vessel disease, which could possibly explain the incoherence between the two studies.

4.2 Activational analysis of fNIRS response

Stroke patients had significant activation of their right inferomedial PFC during congruent stimuli, but not during neutral and incongruent stimuli although their mean average response of Oxy-Hb was positive and Deoxy-Hb close to zero, which could be consistent with this finding. However, neither HC nor left inferomedial PFC showed a similar response pattern which could indicate a type 1 error despite Bonferroni correction.

Absence of classic activation in the PFC is contradictory to previous studies of both healthy young (Ehlis et al., 2005; Lague-Beauvais et al., 2013; Leon-Carrion et al., 2008; Matthias L. Schroeter et al., 2003) and elderly people (Lague-Beauvais et al., 2013; Matthias L. Schroeter et al., 2003) as well as in patients with migraine (Schytz et al., 2010), cerebral microangiopathy (M. L. Schroeter et al., 2007), depression (Ikeda et al., 2013) and traumatic brain injury (Plenger et al., 2016). However, fNIRS studies of both subjects with obsessive compulsive disorder and attention deficit hyperactivity disorder did not find any PFC activation (Okada, Ota, Iida, Kishimoto, & Kishimoto, 2013; Ueda et al., 2018).

Possible reasons for type 2 errors in fNIRS studies include inadequate stimuli, adaptation and fatigue during prolonged testing, poorly positioned optodes, untimely examination window, excessive MA correction and contamination by changes in systemic and extracerebral perfusion (Tachtsidis & Scholkmann, 2016). While the applied SCWT in our study is quite similar or even identical to other studies (Lague-Beauvais et al., 2013; M. L. Schroeter et al., 2007; Matthias L. Schroeter et al., 2003; Schytz et al., 2010), we believe the inter-block period was insufficient as the HRF for some subjects did not to return baseline within 20 seconds. This could potentially diminish the hemodynamic response in the subsequent stimuli block. Furthermore, most fNIRS studies observe functional hemodynamic responses within 5–10 seconds from stimuli onset (Paola Pinti et al., 2020) including Stroop studies (Ikeda et al., 2013; Leon-Carrion et al., 2008; M. L. Schroeter et al., 2007; Matthias L. Schroeter et al., 2003; Schytz et al., 2010), but others have observed activation within a prolonged observation window (Ehlis et al., 2005; Jahani, Hemmati, Rahimpour, & Setarehdan, 2015).

The configuration of optodes applied in this study was constructed to examine most of the superficial PFC, while deeper cortical tissue cannot be examined with fNIRS. Both DLPFC and ACC were areas of interest and examined with superolateral and superomedial channels according to sensitivity mapping of average optode positioning (Fig. 1). However, not all subjects matched the average subject, and thus we suspect optodes were not optimally positioned in some cases to possibly account for some of the large variations observed. Future studies could benefit from digitized optode positioning to account for individual anatomical differences.

In contrast to most other fNIRS studies examining the SCWT, we did apply short-separation regression. This was performed to improve signal quality and brain sensitivity, but also to minimize the possibility of sympathetic nervous activation increasing extracerebral blood flow and mimicking a functional cerebral activation (Tachtsidis & Scholkmann, 2016). However, running the GLM without short-separation regression did not alter our findings (data not shown).

In the processing of data, we examined several MA correction methods and eventually applied a relatively conservative method and then removed all other stimuli blocks with MA to avoid excessive MA correction. Results did not change significantly when other MA correction methods were applied (data not shown).

The inverse activational response observed in superolateral and superomedial PFC was quite consistent especially in healthy subjects. Inverse activational patterns are often attributed to pathologic conditions in the brain with disruption of the neurovascular coupling (Lindauer et al., 2010). Thus, blood flow remains constant, and the increased neuronal metabolism leads to a reduction in Oxy-Hb and an increase in Deoxy-Hb. While this could be true for stroke patients, it should not be the case in our group of HC. Inverse activational pattern have also been observed in fNIRS studies of some subject performing motor tasks (Sato et al., 2005) as well as in motor imagery studies in which inhibition of motion have been proposed to account for the inverse pattern in frontal and prefrontal cortex (Holper, Shalóm, Wolf, & Sigman, 2011). While subjects in the current study only performed slight motor tasks in clicking the mouse, a significant proportion found the response actions (left mouse click for correct stimuli and right mouse click for incorrect stimuli) counterintuitive and conceivably had to employ some level of motor inhibition.

Cases of slight optode misplacement (i.e. to activated parts of DLPFC and ACC) could also lead to such an inverse activational pattern due to a steal phenomenon, as have been observed in multichannel fNIRS setups in healthy subjects (Amiri et al., 2014; P. Pinti, Siddiqui, Levy, Jones, & Tachtsidis, 2021) and in cerebrovascular patients (Akiyama et al., 2005; Murata, Sakatani, Katayama, & Fukaya, 2002). Nearby activation adjacent to the examined cortical area can cause regional increases in the activated area and reductions to blood flow in the examined area and thus diminished Oxy-Hb and greater Deoxy-Hb concentration. A denser and more widespread optode configuration would be preferable in future studies to assess this possibility.

Partial volume effect (i.e., when sampling from activated and non-activated tissues) is another possible explanation for the observed response (Boas et al., 2001; Kleinschmidt et al., 1996; Strangman, Culver, Thompson, & Boas, 2002), but that is impossible to assess in one-modality studies. Systemic perfusion noise has also been attributed to inverse fNIRS responses (Caldwell et al., 2016), which is unlikely in our study due to short-separation regression.

As no definitive physiological explanation of the inverted hemodynamic response have been identified, we conclude that fNIRS have significant path ahead before implementation into a clinical neurorehabilitation setting is feasible. The need for simultaneous and multi-modal examinations (i.e., together with MRI or PET) to investigate inverted hemodynamics is apparent.

Stroke patients exhibited associations between Deoxy-Hb peaks and response time in left inferolateral PFC, but not in right inferolateral PFC nor in any other channels including all HC channels, which could indicate a type 1 error. The lack of associations between hemoglobin peaks and response time conceivably speaks to the heterogenicity in the stroke patient’s disease characteristics, shortcomings in optode positioning and examination protocol as covered above and the complex nature of regional blood flow. Such factors could have created too much noise for any associations to be detected. Instead, this finding emphasizes the need for using any individual as its own control condition whenever it is possible as accounting for the functional response during congruent stimuli when analyzing the functional response during incongruent stimuli.

4.3 Group differences

Regardless of the physiological explanation for the inverse activational patterns observed, HC had significant higher Deoxy-Hb increments than stroke patients in superomedial PFC across conditions and to a lesser extent in superolateral and inferomedial PFC. These findings indicate possible differences in cortical hemodynamics or neuronal activation due to ischemic stroke, though further interpretation of the pathophysiology is incomprehensible.

While group differences were observed in raw hemodynamic response, there was no coherence between hemodynamic response and test performance during incongruent stimuli when accounting for subjects’ response during congruent stimuli. Although rational from a test performance logic, the congruent and incongruent stimuli in the SCWT are quite different in nature and could very well activate neural pathways differently, thus not proving logical in neural activational nor hemodynamic patterns. To our knowledge, there are no comparable studies, but one study of subacute TIA patients showed the same increment in prefrontal perfusion as in HC after physical activity during a one-block SCWT (36 consecutive stimuli). The hemodynamic change was not related to the improvement in test performance in accordance with our findings (Faulkner et al., 2017).

We dichotomized stroke patients to either no small vessel disease or moderate to severe small vessel disease, which indicated higher Deoxy-Hb in superomedial and superolateral PFC during Stroop test in patients with small vessel disease, although these findings were non-significant after Bonferroni correction. Nonetheless, we believe examining and accounting for small vessel disease is crucial in stroke patients when performing hemodynamic or cognitive investigations as most stroke patients have some degree of small vessel disease regardless of stroke etiology (Simonsen et al., 2022; Staals et al., 2014) that can affect blood flow regulation (Kim et al., 2021; Liu et al., 2022) and cognitive performance (Rost et al., 2022). Stroke patients conform an inherently heterogenous population with different lesions and symptoms, cognitive performance, medical history, stroke etiology and anatomy, medication, stroke-related complications, etc. It is beyond the scope of this study to do further investigations across such heterogeneities.

Although activational response seemed more pronounced in the right hemisphere, lateralization analysis showed no significant findings in either group or between groups, even when accounting for infarction hemisphere.

4.4 Strengths and limitations

The NIRS examinations in this study were of high quality with short-separation regression to filter out extracerebral contamination, but with no digitization of optode placement creating uncertainty of the examined area in the individual subject and with insufficient inter-stimulus resting periods in some subjects.

The enrollment criteria for stroke subjects were designed to reflect the common heterogeneity among stroke patients in the everyday clinical setting. Patients were for instance not excluded if the index stroke was a recurrent stroke, nor if their stroke lesion was not within the examined cortical area. While this choice favors the generalizability, it also generates the possibility of multiple cofounders for which it is not possible to account for in such as limited sized population. A very large-scale study would be required to incorporate precise anatomical lesion as well as stroke etiology, domains of executive dysfunction and other important variables into the statistical model beyond the scope and nature of this exploratory study. The stroke patients in our study had to be excluded if their symptoms were too severe for them to perform the Stroop test resulting in a population with only mild to moderate stroke patients. While we recruited HC with sex and age matched to stroke patients, we did not control for educational level which could be a possible confounder.

The large variations in hemodynamic response were greater than anticipated. Thus, the sample size in this study conceivably led to inadequate statistical power as several findings did not survive Bonferroni correction. While it is interesting to examine the entire PFC rather than certain regions of interest, this tendency is also aggravated in multi-channel fNIRS setups. However, increasing the sample size and statistical power would probably not have changed the main conclusions from this study.

Implementation of fNIRS as a biomarker during stroke rehabilitation would require a better understanding of inverted functional responses. Future studies should focus on this as well as examining functional hemodynamic responses during other domains of cognitive and executive function perhaps leaning on lessons learned in the current study.