We collected database for consecutive patients with acute ischemic stroke (AIS) within seven days of symptom onset who were admitted between October 2018 and March 2021. Among these patients, we consecutively selected patients with anterior circulation infarction classified as small vessel occlusion (SVO) according to the Trial of Org 10172 in Acute Stroke Treatment (TOAST) classification system (24). Their National Institutes of Health Stroke Scale (NIHSS) and modified Rankin Scale (mRS) scores were less than three. All participants underwent brain magnetic resonance (MR), TCD and CrCP during admission. We excluded patients who had (1) a history of stroke; (2) arrhythmias that may affect the accurate evaluation of TCD and CrCP; (3) a history of radiotherapy for head and neck cancer; (4) a history of endovascular therapy for anterior circulation diseases; (5) high or medium risk of potential sources of cardiac embolism according to the TOAST classification system; (6) unsuitable temporal windows for conducting TCD and CrCP measurements; (7) Patients with hypercapnia or hypocapnia. The written informed consent for this study was obtained from the patient or his or her family members. The ethics committee of the Second Affiliated Hospital of Guangzhou Medical University approved the study protocol. Good clinical practice guidelines in accordance with the Declaration of Helsinki were used, and patient privacy was strictly protected.

We collected data on sex, age, NIHSS and mRS scores on admission, presence of risk factors [including hypertension, diabetes mellitus, coronary heart disease, current or recent smoking history, alcohol consumption history, body mass index (BMI), triglyceride, cholesterol, low density lipoprotein and glucose], hemodynamic parameters [including blood pressures, pulse pressure, blood flow velocities of middle cerebral artery (MCA), PI and CrCP], CSVD subtype and its score (Table 1). Hypertension was described as present if the subject had been previously diagnosed by a cardiology physician and were routinely receiving antihypertensive therapy. Patients were defined as having type 2 diabetes if they had known diabetes treated by diet, oral hypoglycemic drugs, or insulin before the stroke. Coronary artery disease included any history of heart attack/myocardial infarction, angina, or coronary heart disease. The hemodynamic parameters, such as CBF velocity, PI and CrCP, were measured using TCD. CSVD subtypes and scores were defined according to the standards for reporting vascular changes on neuroimaging (STRIVE) criteria based on the results of the brain MRI (23).

Brain MRI scans for all participants were acquired through a Siemens 3T MRI system (Erlangen, Germany) employing a 12-channel phased array head coil during admission. The study protocol included axial diffusion-weighted imaging (DWI, TE, 90; TR, 3,200) with b values of 0 and 1,000 s/mm², T2-weighted (TR 7,000; TE 102), fluid-attenuated inversion recovery (FLAIR, TR 7,000; TE 102), gradient echo, and sagittal T1-weighted sequences (TR 511; TE 8.5); FOV, 250 × 250 mm; Mmatrix, 256 × 256. slice thickness was 5 mm with 1 mm gap between slices. The total imaging time was ~45 min for each participant.

All MRIs were independently assessed blinded to clinical information by two experienced neuroradiologists, and their discrepancies were resolved by consensus. Structural image analysis of CSVD features was performed according to the STRIVE criteria (23). Recent small subcortical infarct was defined as a hyperintense area ( ≤ 20 mm) involving subcortical tissue on DWI, with T1-weighted hypointensity, T2-weighted and FLAIR hyperintensity. Lacune of presumed vascular origin was defined as a CSF-filled cavity (3–15 mm) surrounded by a hyperintense rim on FLAIR and T2, and no increased signal on DWI. White matter hyperintensities (WMH) of presumed vascular origin was defined as lesions with hyperintensities on T2-weighted and FLAIR imaging and hypointensities on T1-weighted imaging, and graded according to the modified Fazekas scale (25). We defined perivascular spaces (PVS) as small round (axial) or liner (parallel to vessels) space (<3 mm) with CSF-like signal on all MRI sequences without hyperintense rim on T2-weighted or FLAIR imaging, and they were rated on a semiquantitative scale from 0 to 4. Cerebral microbleeds (CMB) were defined as small (2–10 mm) hypointensity on gradient echo images in cerebellum, brainstem, basal ganglia, white matter, or cortico-subcortical junction. According to previous descriptions, we rated the total MRI burden of CSVD using the CSVD score. Each MRI feature of CSVD was given one point if present, for a maximum score of four.

Hemodynamic parameters were measured by one experienced technician using TCD, continuous blood pressure monitor and capnograph. Participants were in a supine position with their heads slightly elevated after resting for 15 min. TCD basal examination was performed using a commercial machine (DWL Elektronische Systeme GmbH, Sipplingen, Germany) by directing a 2-MHz transducer to the temporal window above each zygomatic arch to detect the blood flow in the middle cerebral artery (MCA). Time-averaged peak systolic velocity (PSV), mean flow velocity (MFV) and PI were generated automatically. Once the signal was optimized, the transducer was locked in place and fixed on the temporal window using a custom fixation device (Marc 600, Spencer probe fixation system, Spencer technologies, USA).

Continuous BP recording was made via a tonomatic continuous blood pressure monitor (CBM-7000; Colin corporation, Japan), with the participant's hand maintained at the same level as the head. The measurements were initially corroborated by standard measurements of BP with an automated arm cuff (Omega 1400 series; In vivo Laboratories Inc., Orlando, Florida, USA). End-tidal CO₂ was measured with an automated capnograph (Multi-Box, DWL, Germany). The exhaled air is introduced into the non-invasive capnograph through the double-inserted nasal oxygen tube, and the breathing rhythm and end-tidal CO₂ concentration were continuously and dynamically recorded. The signals of continuous BP and end-tidal CO₂ concentration were transmitted to the TCD machine through a dedicated cable, so that the BP, end-tidal CO₂ concentration curve and MCA blood flow spectrum envelope can be displayed in the monitoring trend window at the same time. After the three waveforms were stable (change rate per minute <10%), the trend graph was continuously recorded for 5 min, and the data were synchronized to the hard disk of the TCD machine.

According to Rune Aaslid's method (12), we used our own offline software to calculate the value of CrCP. We selected the cerebral blood flow wave spectrums with complete envelopes in 6 continuous cardiac cycles (at least greater than one respiratory cycle). To compensate for the time delay between pressure and flow velocity curves at the radial and MCA, flow velocity curves were shifted by an average of 54 msec. The correct compensation of time delay was calculated by iterative regression analysis until hysteresis of BP/FV plots completely disappeared. The least square method was used to analyze the BP/FV relationship line, and the pressure axis intercept of these BP/FV plots represents CrCP of the cerebral circulation. Six continuous cardiac cycles of each measurement period were randomly selected and extrapolated CrCP data of all heart beats within these cardiac cycles were averaged for further analysis.

Data were analyzed using the interclass correlation (ICC) coefficient, described in detail in the Statistical Analysis section below to determine the interobserver and intraobserver reliability for CSVD score measurment as per all the records measured by two experienced neuroradiologists.

Data are expressed as the mean ± standard deviation (SD) or median (25th and 75th percentiles) for continuous variables and as the frequency and percentage for discrete variables. Comparisons between patients with CSVD scores of different levels were performed by unpaired Student's t-test or the Mann–Whitney U test where appropriate for continuous variables and the chi-squared test for categorical variables. To evaluate the Determinants of CSVD burden, we performed multivariate logistic regression with adjustments for the variables in four models. To better understand the relationship between CrCP, PI and the CSVD burden, we performed multivariable linear regression analysis, receiver operating characteristic (ROC) curve analysis and established a fractional polynomial plot with 95% confidence intervals (CIs) for the CSVD scores according to the level of CrCP and PI on the basis of the generalized additive regression model. We also performed multivariable linear regression analyses assessing the relationships between CrCP and the baseline characteristics. The ICC coefficient was used as an index of interobserver and intraobserver reliability/agreement. The interobserver and intraobserver reliability were assessed by fitting two-way mixed effects model using the reliability analysis procedure in SPSS, where the value of CSVD score was modeled with neuroradiologist and subject entered as random effects. Statistical significance was established at P < 0.05. Statistical analyses were performed using SPSS 17.0 software for Windows (SPSS Inc, Chicago, IL, USA) and Stata 14.0 software for Windows (StataCorp. LP, Texas, USA).

In the univariate analysis (Supplementary Table 1), the factors associated with CSVD burden (P < 0.05) were the older age, hypertension, higher blood pressure, pulse pressure, and CrCP levels. In the multivariate logistic regression analysis, hypertension and CrCP remained independent determinants of CSVD burden (Table 2). After adjusting for age, sex, and significant (P < 0.05) variables from the univariate analyses in Model 1, only hypertension and CrCP were independently associated with CSVD burden. After additionally adjusting for the risk factors or hemodynamics, in Model 2 or 3, hypertension and CrCP remained independent determinants. Furthermore, we adjusted for all variables above, including age, sex, risk factors, hemodynamics, and significant variables from the univariate analyses in Model 4, hypertension [OR (95% CI): 1.06 (1.00–1.63); P = 0.020] and CrCP [OR (95% CI): 1.41 (1.17–1.71); P < 0.001] were still the independent determinants of CSVD burden.

In univariate linear regression analysis (Table 3), CVR and CrCP were associated with CSVD burden (all P < 0.05). After adjusting for age, sex, and significant (P < 0.05) variables from the univariate analyses (Table 3), we found that CSVD burden remained associated with CrCP [β (95% CI): 0.05 (0.04–0.06); P < 0.001], but did not have the similar relationship with PI [β (95% CI): −0.29 (−1.00 to 0.42); P = 0.416]. On visual inspection, we also found that only CrCP correlated with the CSVD score in the two hemodynamic parameters (Figure 1).

In ROC analysis, when the continuous value of CrCP was considered as a predictor of CSVD burden (Figure 2A), the AUC was 86.2% (95% CI, 78.6%-93.9%; P < 0.001). The sensitivity was 72%, and the specificity was 86%. However, the ROC curves of PI [AUC (95%CI): 50.7% (37.9–63.6%; p = 0.915)] did not present good predictive values for CSVD burden (Figure 2B).

In univariate linear regression analysis (Table 4), Age, systolic and diastolic BP, MBP and Pulse pressure were associated with CrCP (all P < 0.05). After adjusting for age, sex, and significant (P < 0.05) variables from the univariate analyses, age [β (95% CI): 0.27 (0.06–0.47); P = 0.012], BMI [β (95% CI): 0.61 (0.00–1.22); P = 0.049] and systolic BP[β (95% CI): 0.16 (0.09–0.23); P < 0.001] remained associated with CrCP.