A Cross-Sectional Investigation of Cognition and Ultrasound-Based Vascular Strain Indices

Abstract

Objective

We examine the relationship between variability in the plaque strain distribution estimated using ultrasound with multiple cognitive domains including executive, language, visuospatial reasoning, and memory function.

Method

Asymptomatic (n = 42) and symptomatic (n = 34) patients with significant (>60%) carotid artery stenosis were studied for plaque instability using ultrasound strain imaging and multiple cognitive domains including executive, language, visuospatial reasoning, and memory function. Correlation and ROC analyses were performed between ultrasound strain indices and cognitive function. Strain indices and cognition scores were also compared between symptomatic and asymptomatic patients to determine whether there are significant group differences.

Results

Association of high-strain distributions with dysexecutive function was observed in both asymptomatic and symptomatic patients. For memory, visuospatial, and language functions, the correlations between strain and cognition were weaker for the asymptomatic compared to symptomatic group.

Conclusions

Both asymptomatic and symptomatic patients demonstrate a relationship between vessel strain indices and executive function indicating that silent strokes and micro-emboli could initially contribute to a decline in executive function, whereas strokes and transient ischemic attacks may cause the further decline in other cognitive functions.

Introduction

Stroke is a major cause of worldwide mortality and disability. About 87% of all strokes are estimated to be ischemic (Benjamin et al., 2018). It is estimated that up to five silent strokes occur for every clinically recognized stroke (Dempsey, Vemuganti, Varghese, & Hermann, 2010). Although silent strokes do not typically present with clinical symptoms associated with stroke, they may cause cognitive impairment.

Ultrasound strain imaging (Ophir, Cespedes, Ponnekanti, Yazdi, & Li, 1991; Varghese, 2009) is an imaging modality that can non-invasively estimate tissue deformation and provides images of the strain distribution in tissue. Strain in its simplest form represents essentially the change in length divided by the original length of the tissue segment. In carotid ultrasound imaging, this change is referred to as tissue deformation caused by changes in pressure within the vessel lumen that occur throughout systole and diastole in the cardiac cycle. Typically, carotid strain imaging is performed in a two-dimensional imaging plane. Strain is computed along both the direction of the beam propagation (axial strain) and perpendicular to the ultrasound beam (lateral strain). Shear strains, which occur due to shearing forces on tissues, can also be calculated based on the axial and lateral deformations computed. Higher strains (estimated with carotid ultrasound strain imaging) indicate the presence of larger deformations in plaque and suggest that this plaque is softer, undergoes larger deformations when compared to the artery and surrounding tissue, is subject to fatigue and can thereby become unstable and more prone to rupture.

Carotid plaque may release micro-emboli which enter the brain vasculature and, while not presenting with clinical symptoms, may result in cognitive deficits. This phenomenon is referred to as a “silent stroke,” and the associated cognitive deficit is the silent stroke burden. Previous studies have shown that dysexecutive function in patients prior to carotid endarterectomy (CEA) was correlated with high-strain distributions in carotid plaques measured using ultrasound strain imaging (Dempsey et al., 2018; Meshram et al., 2017; Varghese, 2009; Wang, Jackson, et al., 2016). Carotid plaques incurring large deformations over a cardiac cycle can become unstable over time and become more likely to release micro-emboli, resulting in silent strokes and related cognitive deficits. Importantly, the association of high strain with dysexecutive function was observed in both asymptomatic and symptomatic patients (Meshram et al., 2017; Wang, Jackson, et al., 2016). Wang, Jackson, et al. (2016) identified a single region of interest (ROI) with highest strain values and demonstrated a negative correlation with executive function. Subsequently, Meshram et al. (2017) utilized multiple ROIs in plaque with high-strain values and also demonstrated a negative correlation between high strain and executive function, in addition to strain indices derived from these multiple ROIs.

Another non-invasive ultrasound measure that has been used to assess risk for cardiovascular disease (CVD) and risk for future CVD events is measurement of the carotid intima-media thickness (IMT). Prior to the development of carotid stenosis, carotid IMT has been shown to identify risk for future cerebrovascular and cardiovascular events (Bots, Hoes, Koudstaal, Hofman, & Grobbee, 1997). Moreover, IMT has also been shown to be associated with cognitive deficits (Haley et al., 2007). Haley et al. (2007) demonstrated that although IMT was negatively correlated with attention-executive-psychomotor function, it was not significantly related to language, memory and visual–spatial function. Relationships between arterial stiffness, white matter hyperintensity and cognitive impairment have also been an intensive area of study. Singer, Trollor, Baune, Sachdev, and Smith (2014) and van Sloten et al. (2015) provided a systematic review examining these relationships. Arterial stiffness in these studies was typically measured using carotid-femoral pulse wave velocity (cfPWV) or brachial-ankle PWV or pulse pressure. These studies demonstrated that increased arterial stiffness was associated with an increase in white matter hyperintensity (WMH) volume and decreased cognition. However, the strength of the association between arterial stiffness and cognition was generally weaker than that of arterial stiffness with WMH (van Sloten et al., 2015). The authors (van Sloten et al., 2015) postulated that factors other than microvascular diseases could play a role in causing cognitive impairment and thus this may be why arterial stiffness has a weaker association with cognition. However, global cognitive measures such as Mini-Mental State Examination (MMSE) provided inconsistent results (Singer et al., 2014). In the investigation to be reported here, we examined patients with advanced atherosclerosis awaiting CEA using an ultrasound strain imaging algorithm (McCormick et al., 2012) to directly estimate the stiffness and stability of carotid plaque.

Patients reporting and treated for stroke-like or transient ischemic attack (TIA) symptoms are hypothesized to be affected by both the stroke and the silent stroke burden and are classified as symptomatic patients. Asymptomatic patients on the other hand could potentially be affected by only the silent stroke burden. Jackson et al. (2015) demonstrated that among a group of patients scheduled for CEA, both symptomatic and asymptomatic patient groups exhibited cognitive deficits compared to normal age-matched controls. Jackson et al. (2015) examined cognitive function in symptomatic and asymptomatic patients including executive function, memory, language, and motor skills. It was observed that both groups presented with deficits in executive and memory function, but only the symptomatic group exhibited deficits in language and motor skills.

In this investigation, we examine the relationship between ultrasound-based strain distribution (McCormick et al., 2012; Shi & Varghese, 2007) and variability in plaque with multiple cognitive domains including executive, language, visuospatial reasoning, and memory function. We hypothesize that the association of carotid plaque deformation using strain may vary based on the symptomatic status of the patients. Ultrasound-based strain indices (Meshram et al., 2017; Wang, Jackson, et al., 2016) in carotid artery plaque estimate the stiffness of carotid plaque non-invasively. This is performed using the natural cardiac pulsation of the carotid artery. Strain for an entire cardiac cycle is accumulated because this accumulation can estimate the amount of instability evident in carotid plaque and hence predict the probability of rupture and future ischemic events.

Methods

Participants

Seventy-six patients underwent CEA for clinically significant (>60%) carotid stenosis. Patients were enrolled in the study in the period from September 2011 to February 2017. Inclusion criteria were as follows: age 18 years or older; native English speaker; symptomatic carotid stenosis meeting North American Symptomatic Carotid Endarterectomy Trial (NASCET) criteria or asymptomatic stenosis meeting Asymptomatic Carotid Atherosclerosis Study (ACAS) criteria; and scheduled to undergo CEA under general anesthesia. Patients were excluded if they presented with a previous history of endovascular or open carotid surgery, and/or cervical radiation, or were considered unsuitable for CEA or lacked informed consent capacity. The study was approved by the Health Sciences IRB and all patients provided informed consent. Of the 76 patients, 42 had a previous ipsilateral stroke or TIA as determined by patient history, chart review, or clinical imaging results and were therefore considered symptomatic (M age = 68.43; SD = 10.62; range = 43–85), whereas 34 were identified as asymptomatic (M age = 71.88, SD = 7.04; range = 54–84). Among the 42 symptomatic patients, 22 (52.4%) suffered a stroke whereas 20 (47.6%) experienced a TIA. Stroke and TIA patients were combined in order to provide sufficient power for analysis. The occurrence of either strokes or TIAs made the patients clinically symptomatic based on the current clinical definition. Demographic and clinical information are provided in Table 1. The groups did not differ significantly in age, sex, body mass index (BMI), level of education, or degree of stenosis. Statistical power was adequate. For moderate effect, at an alpha of 0.05, a correlation of r = 0.33 with 76 subjects, we have a power exceeding 0.83 to detect significant correlation of this magnitude.

Neuropsychological Assessment

All patients were assessed 30 days or less before CEA during their clinic visit for mandatory pre-surgical assessment. The research protocol was performed in accordance with the Health Sciences Institutional Review Boards. After providing informed written consent, each participant was administered the 60-min neuropsychological test protocol according to the National Institute of Neurological Disorders (NINDS) and Canadian Stroke Network (CSN) guidelines (Hachinski et al., 2006). This protocol was selected specifically for stroke patients and assesses several important functional domains with tests of executive function/activation, verbal and nonverbal memory, language, and visuospatial skills. For current study, additional IQ subtests (WAIS-IV Information, Digit Span, and Block Design) were added. For a full list of tests administered, see Table 2.

Strain Estimation

A Siemens ACUSON S2000 system was used to acquire ultrasound Radio-frequency (RF) data sampled at 40 MHz using an 18L6 transducer (Siemens Ultrasound, Mountain View, CA, USA) operated at a center frequency of 11.4 MHz. Longitudinal RF acquisitions were performed at three locations, namely the common carotid artery (CCA), internal carotid artery (ICA), and the bifurcation region over two cardiac cycles on both the left and right carotid arteries. Plaque regions were segmented using the Medical Imaging Interaction Toolkit (MITK) software on ultrasound B-mode images with the thickest plaque dimensions. A digitized loop of RF data was first acquired. Following this localized inter frame displacements between subsequent RF frames were estimated and summed appropriately over a cardiac cycle. A gradient of the accumulated displacement was then performed to calculate the strain tensor (McCormick et al., 2012) over several time instants over a cardiac cycle.

Strain indices (Meshram et al., 2017; Wang, Jackson, et al., 2016) also defined in Table 3, namely the Maximum Accumulated Strain Index (MASI) and peak-to-trough values were then computed over these accumulated strain maps and utilized as non-invasive biomarkers of the vascular health of the patients (Meshram et al., 2017). First, a region of interest with high-strain value was identified. MASI is the maximum strain value observed in the region of interest over the entire cardiac cycle. The peak-to-trough (PTT) value is the difference between the maximum and the minimum strain value observed in the region of interest. A high PTT value indicates that plaque within the ROI incurred significant deformation over the cardiac cycle whereas a low value indicates the opposite. PTT was first used by Wang et al. (2014) and relationships similar to MASI were obtained. This observation was also confirmed by Meshram et al. (2017) and Wang, Jackson, et al. (2016) and, hence, reliable estimates have been obtained using PTT. Note that both the plaque region and the adventitial region was included in the computation of the strain indices used as vascular biomarkers (Wang, Mitchell, et al., 2016). The larger of the two strain indices computed for the left or right carotid arteries was used in this study.

Cognitive Data Analysis and Statistics

All test scores were demographically corrected based on published norms were used. The Revised Comprehensive Norms for an Expanded Halstead–Reitan Battery (Heaton, Miller, Taylor, & Grant, 2004) were used for tests of lexical fluency and Trail Making Test A and B. To normalize distributions and allow scores to be pooled within domain, z scores for each test were computed. The BNT-SF did not have enough variance (M = 14.39, SD = 1.08) and, hence, was not included in the study. Within each domain (see Table 2), z scores were averaged to obtain a single composite score per domain. Thus, the average of the different executive tests z scores would form the executive composite score, and similarly for the visuospatial, language, and memory domains. Pearson’s correlation coefficients were computed to assess the relationship between cognition and strain measures. In order to examine the strength of relationships between the cognitive domains, a linear regression of Axial MASI strain versus the four cognitive domains was performed.

ROC Analysis

A median split was done for each domain to classify patients into high and low cognition groups and the MASI strain index was then used to classify groups. Maximum likelihood-based estimation was used to calculate the parameters of a binormal distribution used for the ROC plots using ROC-kit software (Version 0.9.1 beta, Metz ROC Software at the University of Chicago). The 95% confidence interval at the operating point in the ROC plot was also traced.

Comparison of Symptomatic and Asymptomatic Patients

The significance of the difference in strain indices and cognitive function across symptomatic and asymptomatic was examined. The Kolmogorov–Smirnov (KS) test was used to determine if the variables followed a normal distribution and two-sample Student's t-test was used for normally distributed variables whereas the Wilcoxon rank sum test was used for non-normally distributed variables.

Results

Correlation and Linear Model

The means and standard deviations for individual executive, memory, visuospatial, and language tests are presented in Tables 4–7, respectively. Further analysis was performed on the composite scores rather than individual test scores. Relationships between cognitive domain performance and plaque strain are shown in Table 8. For symptomatic patients, significant inverse correlations were observed between cognition and most strain indices except for language function (executive function: range = −0.52 to −0.30; memory: −0.44 to −0.32; visuospatial: −0.50 to −0.31; language = −0.40 to −0.13), respectively. For asymptomatic patients, inverse correlations was observed between two of the strain indices, namely the maximum axial (r = −0.36) and maximum shear (r = −0.39) and visuospatial ability. Strong inverse correlations were observed with executive function and all strain indices (range: −0.54 to −0.39). However, memory and language functions in asymptomatic patients were not significantly correlated with plaque strain.

Fig. 1 presents the relationship between axial MASI strain and composite (a) executive, (b) memory, (c) visuospatial, and (d) language scores. The linear fit for the symptomatic group is lower than for the asymptomatic group for all cognitive domains, demonstrating lower scores for the symptomatic group. The slope of the fit is similar for executive and visuospatial function between the two groups, whereas the symptomatic group has a higher slope for memory and language function.

ROC Analysis

ROC plots with MASI as a classifier for the different domains revealed a similar trend. As shown in Fig. 2, for asymptomatic patients (executive function Area Under the Curve (AUC) range = 0.81–0.68; memory: 0.58–0.44; visuospatial: 0.61–0.54; language: 0.58–0.57) only executive function classifier was better than randomly selecting the patients to classify into high and low cognition groups (AUC = 0.5). Fig. 3 demonstrates that classification was of similar order for all domains for symptomatic patients across cognitive domains.

Comparison of Symptomatic and Asymptomatic Patients

Table 9 presents the effect of symptomatic status on cognitive domains and MASI strain indices. Language function trends to significance (p = .07). Symptomatic patients presented with lower language domain scores. Executive, visuospatial, and memory function and all strain indices had no significant trends that differed between symptomatic and asymptomatic patient groups.

Discussion

Table 10 summarizes the findings of the study. Overall, as observed in Table 10, for symptomatic patients, all cognition domains had significant negative correlation to strain indices. For asymptomatic patients, only executive function had a significant negative relationship to all the strain indices.

Fig. 1 also demonstrates that the relationship between executive function and visuospatial function to axial MASI strain is similar for symptomatic and asymptomatic patients although the asymptomatic group has comparatively lower cognition scores. Memory function demonstrated a stronger decrease with strain for symptomatic when compared to asymptomatic in Fig. 1. Fig. 1 also shows that the relationship between language function and strain is weak for both groups but much weaker for asymptomatic patients with a slope of −0.0029 whereas the symptomatic group had a slope of −0.0085. Fig. 1 also demonstrates that both patient groups have a relationship of high-strain to low cognition across all cognition domains and this relationship was linear.

Executive function demonstrated a significant negative relationship with strain for both symptomatic and asymptomatic patients (Table 8) revealing that the presence of soft plaque appears to be associated with executive dysfunction, less so to the other cognitive abnormalities. The ROC plots in Figs 2 and 3 also demonstrate this where executive function presents with AUC values of 0.81, and 0.78 for asymptomatic and symptomatic patients, respectively, with axial MASI. Why executive function is more sensitive than other cognitive domains to vessel strain regardless of the symptomatic/asymptomatic status of the patients remains to be determined. Possibilities include but are not limited to differential outcome measure sensitivity across cognitive domains versus the diffuse nature of executive control.

For memory, visuospatial, and language abilities, the correlations between strain and cognition were weaker for the asymptomatic compared to symptomatic group (Table 8). ROC plots also demonstrate this relationship with highest ROC values of 0.76, 0.79, and 0.74 for memory, visuospatial, and language function, respectively, for symptomatic patients. Similar values for asymptomatic patients were 0.58, 0.61, and 0.58, indicating that these cognitive abilities may be influenced more by stroke and TIA but less so by silent strokes.

Patients with symptomatic status tended to exhibit greater impairment in language function (Table 9, p value .07) suggesting that language abilities may be more resistant to occurrences of silent stroke and micro-emboli, tending to be more directly affected by stroke or TIA. Table 9 also reveals that symptomatic status did not mean patients had more severe plaques because the strain indices obtained were not significantly different. We examined the number of symptomatic and asymptomatic patients with lower (−1SD) performance across the cognitive domains (executive, memory, visuospatial, language). For the symptomatic group there were 10 (23.8%), 6 (14.28%), 7 (16.67%), and 7 (16.67%) impaired patients per domain compared to 2 (5.88%), 4 (11.76%), 3 (8.82%), and 0 (0%) for asymptomatic patients. Thus, there were more impaired patients in the symptomatic group across all cognition domains, especially in executive function.

The trend for asymptomatic patients to have higher language and memory scores (Fig. 1) compared with symptomatic patients may explain why the correlations and AUC values were significant for all cognitive domains in symptomatic and only for executive function in asymptomatic patients. The relatively weaker relationship between strain indices and cognition measures for language, visuospatial, and memory function for asymptomatic patients when compared with the stronger relationship (higher correlation and AUC values) for symptomatic patients for the same parameters indicate possibly different etiologies for these different patient groups. On the other hand, both asymptomatic and symptomatic patients demonstrate a stronger relationship between the strain indices and the executive function may indicate that silent strokes and micro-emboli could initially cause a decline in executive function whereas strokes and TIAs may cause the further decline in other cognitive functions. This additional decline is observed in symptomatic patients. These relationships should alert clinicians to potential cognitive and functional needs for these patients, especially in asymptomatic patients where an unproblematic clinical status has been classically assumed. Concern and vigilance should be increased regarding the possibility of dysexecutive function and its day-to-day complications.

We had limited access to the patients’ general medical history of comorbid conditions and related treatments. In the future, we plan to extract a more complete medical and medication dataset to explore potential relationships between disease-related factors and carotid plaque strain. The administered tests of executive function were varied in nature with some reliant on working memory/mental manipulation but others more reliant on processing speed. Despite arguably different emphases, the tests were moderately positively correlated (ranging from 0.63 to 0.41). Future research should pursue a more thorough evaluation of discrete aspects of executive function (e.g., working memory, processing speed, problem solving) in order to determine whether the carotid plaque strain/dysexecutive relationship is generalized in nature or limited to discrete areas of function.

Some limitations of the study include the use of radio-frequency ultrasound data; as clinical scanners may not provide this information without a research agreement. However, most commercial ultrasound systems currently provide strain imaging as a clinical application on the system utilizing radio-frequency signals. Because we are using standard cognitive tests for stroke patients and strain imaging is available clinically, this work can be reproduced by other groups. Another limitation is that cognitive deficits in some of these patients may also arise from emboli due to cardiac sources.

Funding

Supported by the National Institutes of Health [R01 NS064034 to R. J. D. and 2R01 CA112192 to T. V.].

Conflict of interest

Carol Mitchell reports authorship contract with Davies Publishing, Inc, authorship with future royalties for textbook chapters with Elsevier and Wolters-Kluwer and contracted research grants from WL Gore to UW Madison.

Acknowledgements

We gratefully acknowledge the support of NVIDIA Corporation with the donation of the Tesla K40 GPU used for this research. Support for this research was also provided by the Office of the Vice Chancellor for Research and Graduate Education at the University of Wisconsin—Madison with funding from the Wisconsin Alumni Research Foundation.

References

Author notes