Despite numerous clinically based research studies, the clinical relevance of SH remains to be fully elucidated. The most well-documented risk factors for presence of SH are aging, hypertension, and other cerebrovascular risk factors [36–39]. SHs have also been associated with cognitive decline, particularly speed of information processing and executive functions [40–43], as well as physical disability, particularly gait disorders and poor motor dexterity [12, 44–47]. In neuropsychological studies, correlations with poor attention and reduced speed of mental processing have been consistent across many series [36, 48, 49].
In patients with AD, SHs have been previously correlated with MMSE scores and Clinical Dementia Rating scores, even after adjusting for potentially compounding variables such as age, sex, and cardiovascular risk factors . Additionally, periventricular WM changes have been reported in 48% to 100% of AD cases [51, 52]. However, when simultaneously considered with measures of GM and WM atrophy, SHs often explain only a small proportion of the variance, usually in relation to executive function measures [13, 53, 54].
The main results of our study support the hypothesis that regional signs of small vessel disease may be differentially correlated with executive function, mental processing speed, and memory. Whereas numerous reports from the literature show mixed results with whole-brain SH volumetrics and associations with cognitive functioning, limited reports have demonstrated brain-behavior relationships with hyperintensities in specific brain regions [6, 11, 40, 55–60]. However, one study which examined the effects of focal lesions on verbal fluency found that strategic frontal lobe damage may modulate executive function . More specifically, patients with damage in the medial frontal region, a region which corresponds to the SABRE parcellated MMF region (Figure 2), were shown to perform poorly on the FAS ‘letter-based’ verbal fluency task . Additionally, several studies examining strategic SH on the cholinergic WM tracts have demonstrated correlations with executive, visuospatial, and memory function [63–67]. A novel study (on a group of normal elderly patients) using a combination of structural and functional MRI demonstrated that an increase in SH within the dorsolateral prefrontal cortex was associated with decreases in prefrontal cortex activity during a working memory task . Another study using FLAIR imaging to examine older stroke patients found differential associations with SH, demonstrating that SH volumes in the frontal lobe were associated with cognitive processing speed and attention, while SH volumes in the temporal lobe were associated with memory impairment . Finally, a recent study using a voxel-based analysis method found specific clusters of SH to be associated with executive function and episodic memory .
Although the results of our study support this handful of studies reported in the literature, it is unclear why more groups have not reported similar findings, given the widespread availability of imaging techniques which allow SH quantification.
It is likely that the limited number of reports found in the literature is due to methodological differences in quantification of SH as well as a lack of standardized definitions for periventricular, deep white, and lacunar-like, cystic fluid-filled infarcts [2, 15, 69, 70]. Although the increasing popularity of thinner-slice 3D FLAIR images has allowed more automatic segmentation of SH, it has also introduced more variability as FLAIR images have been shown to have lower sensitivity in different brain regions . Various approaches have also been applied for the separation of pvSH from dwSH, where some groups have employed proportional distances from the ventricles to the dura matter , whereas other groups employ an arbitrary distance from the ventricles . Our study applied a 3D connectivity algorithm, similar to that of van den Heuvel and colleagues  (2006), whose results will clearly vary from the previously mentioned approaches. Additionally, many older studies were based on semi-quantitative data obtained from visual rating scales to estimate SH load [63, 73–75] or with the use of volumetric estimates derived by adding slice-by-slice semi-quantitative ratings with spherical shape assumptions for each lesion . Furthermore, a recent meta-study examining differences in lacunar lesion definitions revealed a wide variation in the literature regarding the detection and classification of lacunar lesions and thus recommended a consensus for imaging definitions for small vessel disease , now achieved by the STRIVE recommendations .
In a recent study examining early- and late-onset AD, a relationship was demonstrated with frontal SH and mental processing speed by using the Trails A score . Unfortunately, cross-comparisons with our study are difficult as this study did not examine the relationship with mental speed using the periventricular and deep white delineations. However, correlations with mental processing speed and SH in the periventricular region have been reported in at least one major longitudinal study, which demonstrated that progression of pvSH was associated with increased time to complete a Stroop test in a large sample of non-demented elderly patients . Although our study was cross-sectional, a similar association was found by using the Trails A trail-making test in a group of demented elderly patients. Interestingly, the volumetric table in the study by van den Heuvel and colleagues  reports a mean volume for pvSH at baseline of (4.12 cc), which is comparable to the normal elderly pvSH volume in the current study (4.5 cc).
A possible explanation for this decrease in mental processing speed related to pvSH has been proposed, implicating the long association WM tracts [10, 76]. Disruptions from SH in the periventricular region may affect communication between distant multiple cortical brain areas and may result in an overall decrease in speed of processing. In contrast, dwSH may affect communication along the so-called “U-fibers” that connect adjacent areas of the brain and thus are less likely to translate into a significant decrease in processing speed . In this sample, the overall volume of dwSH was small relative to pvSH volumes in both AD and NC (AD: pvSH = 4.5 cc, dwSH = 0.6 cc; NC: pvSH = 1.8 cc, dwSH = 0.3).
This may suggest that the decrease in mental processing speed found in patients with AD may be due in part to some subcortical vasculopathy in the periventricular region related to venous collagenosis . Pathological correlates suggest that pvSH may reflect a form of vasogenic edema resulting from venous insufficiency due to veno-occlusive disease of the deep medullary veins in the periventricular region, which in turn may impair interstitial fluid circulation and exacerbate amyloid angiopathy that is commonly associated with AD . This theory is based on brain pathology imaging work that examined the periventricular venous system in a broad age range (25 to 95 years) of patients at autopsy . The major periventricular and subependymal veins and venules were selectively identified from arteries by using alkaline phosphatase microvascular staining combined with a modified Masson trichrome collagen counterstain. Assessment of this vasculature revealed that the periventricular veins from 65% of subjects over 60 years old had at least 50% stenosis of the periventricular veins because of non-inflammatory collagenous thickening of the venous walls. Moreover, regression analyses revealed that greater venous disease was associated with more severe leukoaraiosis, which was observed as hyperintensities on MRI. Subsequent studies led the authors to conclude that the thickening of the walls of periventricular veins and venules, with collagen subtypes I and III, results in stenosis and occlusion which may restrict venous outflow . This venous collagenosis was believed to be a possible significant contributor to leukoaraiosis, as this venous pathology was likely a progressive and age-related pathology associated with increased WM signal intensity in the periventricular region on MRI of elderly, AD, and vascular dementia patients [80, 81].
In our study, no brain-behavior relationships were demonstrated with lacunar lesion subtypes. As stated previously, there is a large variation in detection and identification of lacunes, consensus-driven definitions have yet to be established, and recent controversies in this matter are yet to be resolved [70, 82]. For clarification purposes, lacunar volumes in the current study were defined as any hypointense (CSF intensity) voxels on T1 that are found within hyperintense SH-defined voxels on PD/T2. A large number of participants in our sample population had no signs of lacunar infarction as defined by this segmentation method. Although patients with AD in our study had a larger volume of lacunar infarcts compared with NC, it is likely that the wide variation and regional distribution for lacunar lesion volumes in both groups limited correlative power in the analysis.
The results from our study also suggest that poor memory performance is correlated with SH volumes in the left temporal lobe. A recent study report from a Dutch group examining patients with AD (n = 107) found a similar association with hyperintensities in the temporal region and memory by using the Visual Association Test . Another group used a similar region of interest-based method examining SH volumes in a stroke population and found an association with right temporal SH and memory by using a numerical working memory task . In contrast, a composite score was used in our study, which was composed of several memory tests that included a visual reproduction task as well as a verbal memory test. Despite these varied memory components, a correlation was demonstrated with SH volume in the left temporal lobe. Unfortunately, the SABRE VOIs for the temporal lobes were relatively large. Future releases of this software will include further subdivisions of the temporal lobe.
As a final comment, it is important to note that the regional correlations with SH volumes demonstrated in our study were relatively small compared with the large proportion accounted for by atrophy, as indicated by BPF. In large longitudinal studies, such as the Alzheimer’s Disease Neuroimaging Initiative (ADNI), vascular markers have been shown to be associated with brain atrophy. When the Boundary Shift Integral was used as a measure of annualized change in brain volume, a relationship between atrophy and SH was demonstrated in pre-dementia stage participants enrolled in ADNI . However, in the ADNI sample, this relationship was demonstrated in NC participants but not in participants with AD or mild cognitive impairment (MCI). The opposite was demonstrated in our study, which showed SH to be associated with atrophy measures in AD patients but not in NC. This could be due to sample differences and the inclusion of an MCI sample, where a larger AD sample was analyzed in our study (Sunnybrook AD: n = 265 versus ADNI AD: n = 146), and a larger NC and MCI sample in the ADNI study (ADNI NC: n = 197, ADNI MCI: n = 331, versus Sunnybrook NC: n = 100).
Ventricular expansion is another commonly used MRI-based biomarker of atrophy . Longitudinal results from the large Cardiovascular Health Study showed significantly greater vCSF volumes in dementia cases compared with MCI, and MCI compared with NC . Additionally, it was suggested that larger vCSF volumes in NC at baseline may be used to indicate future dementia progression. Future longitudinal analyses examining baseline measures of SH and atrophy may further elucidate the predictive cognitive outcomes of these MRI markers of disease progression.
As our study was cross-sectional, it would be interesting to examine volumetric change over time, particularly with SH subtypes and other brain tissue compartments, to determine whether these changes predict progression patterns in comparison with ADNI data sets. Future directions will examine the inclusion of additional patient groups, including those with MCI, to further examine the relationships between disease progression and mixed pathologies. Diffusion tensor imaging measures of fractional anisotropy and diffusivity and arterial spin labeling perfusion imaging obtained in a subset of these participants will also be added to the correlative modeling.