Pulsatility Index in the Basal Ganglia Arteries Increases with Age in Elderly with and without Cerebral Small Vessel Disease

Participants

In the present study, we included 48 older adults (mean age, 71 [SD, 8.3] years; 40% women) recruited between December 2016 and July 2018.

Twenty of the participants were selected from a longitudinal 3T MR imaging study on the pathophysiology of CSVD, conducted by the University Clinic of Magdeburg and the German Center for Neurodegenerative Disease. The inclusion criterion for the 3T study was the presence of hemorrhagic CSVD markers, ie, MBs or cSS on MR imaging iron-sensitive sequences (gradient recalled-echo T2*-weighted or susceptibility-weighted imaging) on a prior clinical MR imaging performed for a diagnostic work-up (eg, due to headache, epileptic seizures, gait disturbances, cognitive impairment, and TIA). MBs (classified through their localization as strictly lobar, deep [BG, thalamus], or mixed²⁸), cSS, and asymptomatic intracerebral hemorrhage (ICH) (hypointense lesions of  >1 cm, deep or lobar) were confirmed on 3T iron-sensitive sequences. 3T MR imaging was further rated for the existence and severity of CSVD downstream pathologies in the region of the BG, including WMH following the peripheral outline of the BG (peri-BG WMH, existent/not existent, which were identified on a FLAIR sequence); deep lacunes (number; defined in T1-weighted sequence as fluid-filled, 3- to 15-mm large cavities in the BG²⁹); and BG EPVS (severity: 0 = no EPVS; 1 = 1–10 EPVS; 2 = 11–20 EPVS; 3 = 21–40 EPVS; 4 = >40 EPVS; counted on a T2-weighted sequence).³⁰ A specialist neurologist (S.S., with 10 years’ experience) performed the rating.

CAA was diagnosed by applying the modified Boston criteria, which define probable CAA by the presence of at least 2 lobar hemorrhages (ICH and/or MBs) or 1 lobar hemorrhage and at least focal cSS, which had an alternative cause, in patients 55 years of age or older.³¹ For a diagnosis of possible CAA, the presence of 1 lobar hemorrhage is sufficient.³¹ Patients with HA were identified through the presence of strictly deep MBs and/or ICHs or mixed deep and lobar MBs and/or ICHs.³²

The remaining 28 participants were community-dwelling controls recruited from an existing pool of cognitively healthy volunteers at the German Center for Neurodegenerative Disease in Magdeburg without hemorrhagic CSVD markers on 3T MR imaging. Multiple WMH subcortical spots and grade I EPVS in the BG or centrum semiovale were allowed in all controls, because they represent a common finding in aging (Table 1).

Participants with large-artery stenosis of the circle of Willis, genetic neurologic disease, and a history of psychiatric disease, alcohol or drug abuse, and cerebrovascular malformations were excluded from the study. No patient had a cardiac ejection fraction of <45%, whereas these data were not available for controls. Contraindications for scanning at 7T, according to the recommendations of the German Ultrahigh-Field Imaging Network, were considered and represented a further exclusion criterion from our study. Each participant was subject to a neurologic examination to detect undiagnosed neurologic conditions and further underwent an extensive neuropsychological test battery (Online Supplemental Data) to exclude severe dementia and depression, which can be a cause of pseudodementia. Cardiovascular risk factors were additionally recorded for all participants (Table 2). Arterial hypertension was identified when blood pressure exceeded 130/80 mm Hg. Diabetes mellitus was diagnosed as a fasting plasma glucose level of >7.0 mmol/L or >11.1 mmol/L 2 hours after a glucose tolerance test. Hyperlipidemia was defined as abnormal blood levels of low-density lipoprotein cholesterol (>2.6 mmol/L) and/or triglycerides (>1.7 mmol/L).

All participants provided written informed consent according to the Declaration of Helsinki and were compensated for travel costs. The study was approved by the University Clinic of Magdeburg Ethics Committee (93/17; 28/16).

7T MR Imaging

Scanning of all participants was performed on a whole-body 7T scanner (Siemens; “classic” 7T), equipped with a quadrature transmit and 32-channel receive head coil (Nova Medical). First, a T1-weighted sequence with 3D-MPRAGE (voxel size = 1 mm³ isotropic) was acquired for anatomic reference, to plan the acquisition of the 2D PC sequence and to individualize the BG as an ROI during imaging analysis. The parameters were the following: TE = 2.89 ms, TR = 2250 ms, flip angle = 5°, TI = 1050 ms, receiver bandwidth = 130 Hz/pixel, echo spacing = 8.3 ms, 3D matrix dimensions = 256 × 256 × 176. A generalized autocalibrating partially parallel acquisition was performed with an acceleration factor of 2 and 32 reference lines.

We implemented the previously described^21,22 2D PC sequence on our scanner, using the multisite protocol as defined in the European Ultrahigh-Field Imaging Network for Neurodegenerative Diseases.³³ The sequence was acquired in a single section at the level of the BG and was planned manually on sagittal images of the T1-weighted sequence so that it crossed the anterior commissure and was tangential to the genu and splenium of the corpus callosum (Fig 1A). Thus, we ensured that the single section was approximately perpendicular to the perforating arteries. The reproducibility of this method has been tested in previous studies.²¹ Sequence parameters were the following: voxel size = 0.3 × 0.3 × 2 mm,³ TE = 16.6 ms, TR = 27.6 ms, 2 segments (phase-encoding steps per cardiac cycle), flip angle = 45°, receiver bandwidth = 59 Hz/pixel. Encoding velocity was 20 cm/s in the direction perpendicular to the section. The sequence was retrospectively gated with an acquired temporal resolution of 110.5 ms, and blood flow velocity was reconstructed at 14 time points of the cardiac cycle (ie, approximately 71 ms for a heart rate of 60 beats per minute). The scan time for the 2D PC sequence was approximately 3 minutes at a heart rate of 60 beats per minute.

• Download figure
• Open in new tab
• Download powerpoint

FIG 1.

Image processing. Section position of the 2D phase-contrast acquisition through the BG was planned on the T1-weighted sequence (A). Mean magnitude image and determination of the ROI in which vessel detection was performed (B). Circles mark detected perforating vessels in the ROI, distinguishing between upward (blue) and downward (red) blood flow directions. Only vessels with an upward flow direction were considered (C), to ascertain inclusion of arteries only (no veins). V indicates velocity.

Image Processing

Quality control was performed by visual inspection of the scans by T.A., especially with regard to motion artifacts (ghosting, blurring). If motion caused extensive ghosting and blurring that would hamper the analysis, the subject was excluded. To process and analyze flow-encoded PC images, we used a Matlab (2015b; MathWorks) tool, developed at the University Medical Center of Utrecht.²² This tool performs several steps as described in more detail elsewhere.^22,34 Briefly, perforating arteries were identified as voxels on the velocity map with nonzero mean velocity, using the following steps:³⁴ First, a phase correction was applied on the background to make the mean velocity of tissue 0 cm/s by median-filtering the time-averaged velocity map and subtracting it from the velocity map of each cardiac time point. Then, velocity SNR maps were calculated from magnitude SNR estimates that were based on the SD of the magnitude over the cardiac cycle. Furthermore, given the velocity SNR, the 2-sided 95% velocity CIs were estimated for the mean blood flow velocity (Vmean) to enable consistent selection of vessels with significant flow (ie, nonzero Vmean with a statistical significance of .05). All voxels inside the BG mask without 0 cm/s within their CI of Vmean and hyperintense signal in the magnitude image were considered significant. Every group of neighboring significant voxels was defined as belonging to the same perforating artery. The voxel with the highest mean velocity of such a group was taken as representative of the perforating artery and used for further analysis.

Detection of perforating vessels was confined to the BG on both sides by manually segmenting the area delimitated by the ventricles, the interhemispheric fissure, and laterally by the insula (Fig 1B). Multiple detections of a single vessel and tangentially intersected vessels were manually excluded from the analysis by a neurologist (V.P., with 4 years of experience), and only vessels with an upward blood flow direction (positive velocity) were considered, to avoid inclusion of veins. In each subject, the generated output of the imaging analysis listed all detected perforators (ie, number of vessels), together with the velocity curves. The location of the detected vessels was visualized on the magnitude image (Fig 1C). Image processing was performed by V.P.

Statistics

Differences in risk factors between groups were calculated using the 1-way ANOVA or Kruskal-Wallis test when appropriate.

The blood flow velocity of each perforator at each of the 14 time points of the cardiac cycle was registered and averaged over all selected vessels, obtaining a mean velocity curve over the cardiac cycle for each subject. Averaging this mean velocity curve over the time points measured resulted in a Vmean in the subject’s perforators of both hemispheres.

Besides the mean velocity curve, we also computed the mean normalized velocity curve by averaging over the normalized velocity curves of all selected vessels. This normalization was performed to give each vessel an equal weight before determining the PI from the mean normalized velocity curve according to the following equation:

In this equation, Vmax is the maximum velocity; Vmin, the minimum velocity, and Vmean, the mean velocity of the mean normalized velocity curve (Vmean = 1 for normalized velocity curves but is still shown for clarity). The use of the averaged normalized curve to perform the PI calculation instead of calculating it in each single vessel was based on previous work that showed that with this method, the overestimation of PI due to noise was greatly reduced.

We further aimed to investigate the relationship between CSVD downstream pathologies in the region of the BG and the blood flow velocity and pulsatility of the BG perforating arteries. Quantile regression analysis was, therefore, performed, setting the mean velocity, PI, and number of detected vessels, respectively, as dependent variables and age, sex, group (CAA, HA, controls), number of deep MBs, the presence of peri-BG WMH, number of deep lacunes, and the severity grade of BG EPVS as independent variables. Quantile regression was applied to the whole cohort as a method of regression analysis robust against outliers and was able to understand the distribution of a continuous variable across different quantiles. We applied this method to calculate the 25th, 50th, and 75th quantiles. To determine to which participants’ group the effects were pertinent, we calculated the Spearman correlations for significant factors in each group separately. P values were adjusted for multiple comparisons using the Bonferroni method, accounting for the number of predictors (n = 7) in the quantile regression models and the number of subgroups (n = 3) for the Spearman correlations. Results were considered significant at P < .05. For all analyses described so far, the software R, Version 3.6.0 (www.R-project.org), was used.

To investigate possible differences in blood flow velocity of the perforating vessels between groups, we created a combined mean velocity curve of all subjects within a group. To this end, the method previously described by Geurts et al²⁷ to synchronize all individuals was used and applied to the large vessels in the BG section. The large-vessel curves were obtained by selecting pixels in the image with the highest intensities, and the group mean velocity curve of the large vessels in the BG section was used for time lag synchronization. This mean large-vessel curve was iteratively updated with the synchronized personal velocity traces of the large vessels until no time lag changes occurred. These time lags were then used to synchronize each subject’s mean small-vessel velocity trace. After averaging, we obtained a group mean small-vessel velocity trace. Flow-velocity curves between subgroups with CSVD (CAA versus HA) and healthy controls were visually compared.