Subcortical vs. periventricular WMLs
Based on their proximity to ventricles, WMLs were classified as subcortical or periventricular in seven studies [24–28, 33, 34]. The results show that more studies have found an association between periventricular WMLs with the cognitive domain of executive function, than subcortical WMLs.
Subcortical WMLs are believed to primarily disrupt short connections, and thus impairing cognitive performance supported by the specific brain region . For example, dexterous hand and arm movements are generally thought to be primarily supported by the motor cortex. Therefore, subcortical WMLs in this specific region can result in reduced performance in hand and arm dextrous movements . In contrast, periventricular WMLs disrupt longer connections to spatially distant cortical areas, and thus can cause cognitive performance decline in multiple domains [24, 27]. For example, executive function tasks typically used in research experiments depend on multiple brain regions (i.e., frontal and non-frontal) which are not necessarily located spatially close to each other . Therefore, any disruption in long white matter tracts traversing from periventricular areas may initially reduce the axonal transmission speed , and later cause impaired executive function. In summary, cognitive function depends on intact connections within subcortical areas and between cortical and subcortical structures, and any disruption in these connections may impair cognitive function.
We categorized all included studies into two major cognitive domains which are sensitive to aging: 1) memory; or 2) executive function/processing speed. The latter category was a combination of two cognitive domains based on the idea that they are not mutually exclusive, and one needs to control for their mutual relationship before examining their unique effects .
For memory, out of seven studies, three studies [24, 27, 33] found a significant association between periventricular WMLs and memory performance, two studies [26, 28] found a significant association between subcortical WMLs and memory performance, and two studies [25, 34] did not find any association.
For executive function/processing speed, out of six studies, three studies [24, 27, 34] found a significant association between periventricular WMLs and executive function/processing speed, while only one study  found a significant association between subcortical WMLs and executive function/processing speed. Two studies [25, 33] did not find any association.
Thus, our overall results show that greater number of studies found an association between cognitive impairment (in both domains of memory and executive function/processing speed) and periventricular WMLs, compared with subcortical WMLs. Moreover, greater number of studies showed an association between impairment in the domain of executive function/processing speed with periventricular WMLs, compared to subcortical WMLs.
As highlighted earlier, periventricular WMLs may impact multiple domains of cognition because they disrupt distant connections. Hence, our findings concur with the general knowledge that the domain of executive function/processing speed may depend on multiple brain regions and spatially distant connections [37, 40].
Seven studies [14, 21, 29–32, 35] investigated regional WMLs. No common pattern was evident secondary to the heterogeneity of regions studied.
The following regions demonstrated significant associations between WMLs and cognitive function: cerebral white matter, cerebellar white matter, and basal ganglia ; frontal (dorsolateral frontal and prefrontal) [29–31], parietal, occipital, and temporal lobes [29, 31, 35]; internal and external capsule ; posterior corona radiata, and splenium corpus callosum . This systematic review provides researchers with a summary set of brain regions in which an association have been found between WMLs and cognitive performance. To better understand the role of anatomical location in the association between WML and cognitive function, future studies should examine the spatial distribution of WMLs on the whole brain, or specific set of brain regions identified in this review as being highly associated with cognitive dysfunction.
The discrepancies between the results may be due to the heterogeneous study methodologies and the quality of included studies.
Different MRI sequences, WML quantification methods, and neuropsychological batteries
The included studies were heterogeneous in MRI sequences for WML detection (i.e., PD, T1, T2, or FLAIR), WML quantification method (i.e., scoring or volume measurements), and components of neuropsychological batteries. This likely contributed to variability in our results.
Moreover, two different methods were used for WML quantification: 1) scoring [24, 25, 33, 34]; and 2) volume measurement [26–28]. Scoring measures are usually done manually, and show a higher accuracy for selection of subtle WMLs, compared to automatic volumetric methods. However, these methods vary significantly in terms of lesion classification and severity scoring. Moreover, each scoring method has its own specific limitations.
For WML volume measurement, there are two steps. The first step is identifying lesions, which can be done either manually by an expert radiologist or automatically. After the WMLs are identified manually or automatically, one can proceed to the second step, which is measuring WML volumes automatically. It has been shown that both scoring and volumetric quantification methods are reliable for measuring WML load [41, 42]. However, periventricular and subcortical WMLs quantified by these two quantification methods are differently associated with cognitive function . Out of three studies which used volume measurement, two studies [26, 28] showed a significant association between the subcortical WMLs and cognitive performance. Out of four studies which used scoring, three [24, 33, 34] showed a significant association between periventricular WMLs and cognitive performance. These results suggest that scoring might have biased the results toward periventricular WMLs. Conversely, volume measurement might be problematic for periventricular WMLs due to their similar appearance to CSF on some MRI sequences (e.g., T2 or T1) .
Modifying effect of cardiovascular risk factors
There is a growing recognition that WMLs are associated with age and cardiovascular risk factors [8, 44]. However, all but one included study  considered the modifying effect of cardiovascular risk factors in the statistical analysis. We recommend that future studies consider including cardiovascular risk factors in their analysis.
Quality of studies and lack of sample size calculation
One study  did not demonstrate a significant association between any type of WMLs and any of the cognitive tasks. Based on our quality assessment, this study is the only study categorizing WMLs locations as subcortical and periventricular that did not consider age or education as potential confounders. Therefore, we concluded that this study did not provide strong evidence for the lack of correlation between WMLs and cognitive function.
Moreover, the lack of sample size calculations in all of the included studies might have resulted in possible type II errors. However, we do recognize that the lack of sample size calculations may be due to the dearth of data in this research area .