This was an investigator-initiated prospective single-center study. We reviewed the demographic, clinical, laboratory and radiological data of all patients who received brain MRI scan but had no diagnostic intracerebral lesions such as trauma, infection, acute stroke, and space-occupying lesions in our department from January 2010 to April 2016.

Inclusion criteria were: (1) WMHs visible on FLAIR images; (2) age above 30; (3) had written informed consent. Exclusion criteria were: (1) patients with secondary causes of white matter lesions, such as demyelinating, metabolic, immunological, toxic, infectious, and other causes; (2) patients with abnormal brain MRI findings such as head trauma, hemorrhage, infarction (except lacunes) and other space-occupying lesions; (3) patients with definitive peripheral neuropathy, spinal cord disease; (4) evidence of calcification on CT scans or encephalomalacia in the deep gray matter structures since it may influence the observation of DMVs.

Twenty normal control subjects with matched age and gender were recruited from elderly volunteers for the visibility of DMVs after giving written informed consent, who were volunteers of our ongoing functional MRI projects. Their clinical data (for common vascular risk factors such as hypertension, diabetes mellitus and hyperlipidemia), laboratory examinations, and radiological examinations (both MRI and CT) were normal.

All subjects had given written informed consent prior to the study, and the protocols had been approved by the local ethics committee. All clinical investigation has been conducted according to the principles expressed in the Declaration of Helsinki.

All subjects underwent multi-model MRI by a 3.0 T MR (MR750, GE Healthcare, United States) scanner using an 8-channel brain phased array coil, including T1, T2, T2 FLAIR, and SWI sequence. In order to minimize head motion, foam pads were inserted into the space between the subject’s head and the MRI head coil. Axial T2 FLAIR sequence was used to measure the WMHs volume with the following parameters: repetition time = 8400 ms, echo time = 150 ms, FOV = 24 cm × 24 cm, matrix size = 256 × 256, inversion time = 2100 ms, slice thickness = 4.0 mm with no gap (continuous) between slices, and in-plane spatial resolution of 0.4688 mm/pixel × 0.4688 mm/pixel. The whole brain was imaged. The SWI sequence was in an axial orientation parallel to the anterior commissure to posterior commissure line and covered the whole lateral ventricles, using a three-dimension multi-echo gradient-echo sequence with 11 equally spaced echoes: echo time = 4.5 ms [first echo], inter-echo spacing = 4.5 ms, repetition time = 34 ms, FOV = 24 cm × 24 cm, matrix size = 416 × 384, flip angle = 20°, slice thickness = 2.0 mm with no gap between slices, and in-plane spatial resolution of 0.4688 mm/pixel × 0.4688 mm/pixel. Flow compensation was applied. Magnitude and phase images were acquired.

The raw data were transferred to a separate workstation (ADW4.4, GE), and a custom built programwas used to reconstuct the magnitude and phase images.

We assessed DMVs on five consecutive periventricular slices (10 mm thick) of SWI phase images from the level of the ventricles immediately above the basal ganglia to the level of the ventricles immediately disappeared for each patient, considering that these slices cover most of the DMVs. According to medullary venous anatomy (Lee et al., 1996), six regions including frontal region, parietal region and occipital region (bilateral, respectively) were separated on the above five slices and the characteristics of the DMVs were then evaluated in each region, respectively (Figure 1).

The following four-point score was used for the evaluation of DMVs: Grade 0 - each vein was continuous and had homogeneous signal; Grade 1 – each vein was continuous, but one or more than one vein had inhomogeneous signal; Grade 2 – one or more than one vein were not continuous, presented with spot-like hypointensity; Grade 3 – No observed vein was found continuous (Figure 2). Two neurologists (RZ and YZ, both with 3 years of neuroimaging review experience), who were completely blinded to the subjects’ clinical data and disease state, visually assessed the vascular changes.

The axial T2 FLAIR images were processed for the quantification of WMHs volume after it was coregistered to the phase images, according to the published methods (DeCarli et al., 1992, 1995). The segmentation threshold for WMHs was determined a priori as three standard deviations in pixel intensity above the mean of the fitted distribution of brain parenchyma as described previously (DeCarli et al., 1995; Wen and Sachdev, 2004). Then the automatically segmented lesions were manually checked and corrected by two experienced neuro-radiologists (XY and MZ) who were blinded to all other imaging and clinical data. The manual correction process included: (1) Correction of non-white matter area being labeled as WMH; (2) WMH area not properly labeled as WMH or normal appearing white matter falsely labeled as WMH. Since the intracranial volume (ICV) of each individual was different, we measured the corrected WMHs (cWMHs) volume for further analysis with the following equation: cWMHs volume = (WMHs volume × mean ICV)/ICV. The mean ICV is the mean value of ICV of all patients. The volume of periventricular hyperintensities (PVHs) and deep white matter hyperintensities (DWMHs) were measured, respectively (Fazekas et al., 1993; O’Sullivan et al., 2002), corrected by the ICV, too. PVHs volume was calculated 10 mm away from everywhere around the lateral ventricles for confluent type (Young et al., 2008). Additionally, the same six regions used for the above measurement of DMVs were used for the volumetric assessment of regional PVHs, which were defined as the sum of PVHs in each region, corrected by the ICV, too (Figure 1).

Microbleeds were identified according to a field guide of microbleeds detection and interpretation. SWI magnitude images were used to identify microbleeds. Briefly, microbleeds should be small, rounded or circular, well-defined hypointense lesions within brain parenchyma with clear margins ranging from 2 to 10 mm in size (Greenberg et al., 2009). Signal voids caused by sulcal vessels, calcifications, choroid plexus, and low-signal averaging from adjacent bone were excluded (Vernooij et al., 2008).

T2 FLAIR images were used to identify lacunes. Lacunes were defined as cavities with signal intensities similar to cerebrospinal fluid on MRI and with a diameter of 3 to 10 mm which was different from the enlarged Virchow–Robin spaces by the size, shape, and rim (Gouw et al., 2008).

All the raters were blinded to all other clinical and imaging data of the patients. A single-trained observer (RZ) performed the quantitative assessments of 50 patients twice, at an interval of 3 months apart. Another observer (YZ) independently made the measurements on the same patients. The interobserver intraclass correlation coefficients (ICCs) was 0.94 for grading of DMVs (including each region). The intraobserver ICCs was 0.91 for grading of DMVs (including each region). ICCs were described in detail elsewhere (Lee et al., 1989).

Patients were dichotomized according to WMHs volume or DMVs scores at the median. Comparison between groups were assessed by using Student’s t-test for data that followed normal distribution, Mann–Whitney U test for data that did not follow normal distribution, and Fisher’s Exact test for categorical data. We performed partial Pearson’s correlation analysis to determine the correlation between log-transformed corrected cWMHs volume, PVHs volume, DWMHs volume, and DMVs scores, by adjusting for age, and number of lacunes. We also conducted logistic regression analysis to provide an odds ratio statistic in order to facilitate comparison with other known risk factors. Backward stepwise conditional method was used to avoid the unstability resulted from collinearity of the included variables. The log-transformed cWMHs volume and DMVs scores appeared to be acceptably normative. All analyses were performed blinded to participant identifying information. A p value of <0.05 was considered to be statistically significant. All statistical analysis was performed with SPSS package (14.0 for Windows).

Totally, 268 consecutive WMHs patients were enrolled in this study. The main reasons for admission of those patients were transient ischemic attack (TIA) or lacunar ischemic stroke (n = 140, 52.2%), dizziness (n = 43, 16%), cognitive impairment (n = 18, 6.7%), gait disturbance (n = 9, 3.4%), anxiety or depression (n = 5, 1.9%), incidental finding of WMHs on MRI without symptoms (n = 53, 19.8%). Their age were 66.6 ± 11.5 years and cWMHs volume were 14.1 ± 16.3 mL. Among them, 160 (59.7%) subjects presented with microbleeds, and 110 (41.0%) subjects presented with lacunes.

There were no significant differences in age and gender between WMHs patients and controls. The DMVs score in WMHs patients was significantly higher than controls [8 (6–11) vs. 0 (0–2), p < 0.001].

Table 1 showed the baseline characteristics of patients for comparison. We classified the patients into mild cWMHs group (n = 134) and severe cWMHs group (n = 134) according to the cWMHs volume dichotomized at the median (8.36 mL).

Patients with severe cWMHs had higher DMVs scores (p < 0.001), more numbers of CMBs (p < 0.001) and more number of lacunes (p < 0.001), compared with those with mild cWMHs. Glucose and total cholesterol level were lower in patients with severe cWMHs.

Age, glucose level, total cholesterol level, DMVs score and number of lacunes were included in the primary logistic-regression model, and age, glucose level were then removed from the backward stepwise conditional model. The DMVs score was an independent factor for severe cWMHs (OR = 1.359; 95% CI: 1.224 to 1.508; p < 0.001), after adjusting for total cholesterol level and number of lacunes.

We classified the patients into high DMVs scores group (n = 165) and low DMVs scores group (n = 103), according to the DMVs scores dichotomized at the median (8) (Table 2).

Patients with high DMVs scores had higher cWMHs volume, higher PVHs volume, and higher DWMHs volume, compared with those with low DMVs scores (all p < 0.001). Patients with high DMVs scores had more numbers of microbleeds (p = 0.018) and lacunes (p < 0.001). Glucose level were lower in patients with high DMVs scores (p < 0.001), and high-sensitivity C-reactive protein (hs-CRP) level were higher in patients with high DMVs scores (p = 0.008).

Since cWMHs volume, PVHs volume, and DWMHs volume were highly correlated with each other, they were included in the logistic-regression model separately. Finally, we found cWMHs volume (OR = 1.048, p = 0.001), PVHs volume (OR = 1.166, p < 0.001) and DWMHs volume (OR = 1.051, p = 0.007) were all independent factors for high DMVs scores, after adjusting for age, glucose level, Hs-CRP level and number of lacunes.

Significant correlations were found between DMVs scores and cWMHs volume (Pearson r = 0.463, p < 0.001), PVHs volume (Pearson r = 0.539, p < 0.001) and DWMHs volume (Pearson r = 0.366, p < 0.001), respectively, after adjusting for age, glucose level, Hs-CRP level and number of lacunes.

Additionally, as Table 3 shows, the correlation coefficient between DMVs scores and regional PVHs volumes in the same brain region was highest, except left frontal and right occipital region. Figure 3 was an example figure illustrating the regional correlation between DMVs and PVHs.