Table 3.

Proposed analysis standards for neuroimaging features of small vessel disease

	Measures of interest	Qualitative analysis standards	Quantitative analysis standards	Study design	Accuracy, reliability, feasibility	General comment
Recent small subcortical infarct	Number (multiplicity might indicate other causes);85,86 size (maximum diameter); volume; location (anatomical region, vascular territory); shape (round, ovoid, tubular); swelling (indicates recent, not old)	Various coding schemes available for location: anatomical (eg, centrum semiovale, corona radiata, basal ganglia, thalamus, internal capsule, external capsule, optic radiation, cerebellum, and brainstem), and the vascular territory (eg, middle cerebral artery, posterior cerebral artery, internal carotid artery, and basilar artery)	Possible, but impractical for size and volume	Cross-sectional and longitudinal: recent small subcortical infarcts are typically detected in the setting of an acute clinical event, but can also be an incidental finding	Easy to identify on DWI, reliability depends on time between infarct and imaging; more difficult when using other sequences or CT without longitudinal data	Mimics include acute inflammatory multiple sclerosis plaques; acute lesions generally have increased signal on DWI and reduced signal on apparent diffusion coefficient images
Lacune of presumed vascular origin	Number (one or many); size (maximum diameter); shape (round, ovoid, tubular, other); location (anatomical region); evidence of previous haemorrhage; ex-vacuo effect	Various coding schemes available for shape and location: anatomical (eg, lentiform nucleus, thalamus, internal capsule, centrum semiovale, brainstem); prominent ex-vacuo effect indicates lesion was originally larger (eg, striatocapsular infarct)20	Protocols for quantitative measurement available, need manual correction	Cross-sectional and longitudinal: particular care is needed to differentiate lacunes from perivascular spaces; longitudinal: difference in imaging helps to identify incident lacunes	Differentiation from perivascular spaces can be difficult; high observer agreement should be achieved before undertaking actual ratings	Hypointense rim on T2*-weighted imaging suggests previous small deep haemorrhage
White matter hyperintensity	Volume; location (anatomical region); number	Various coding schemes available for anatomical location (eg, periventricular, deep, subcortical, brainstem; or centrum semiovale, corona radiata, internal capsule, external capsule, optic radiation, brainstem; or frontal, temporal, parietal, occipital)	Various visual rating scores87–92 and protocols for quantitative measurement93,94–97 are available; the two approaches are complementary;82,83 outputs should be visually reviewed by an experienced rater for mimics, artifacts, focal infarcts, and mislabelling	Cross-sectional and longitudinal: consider masking recent small subcortical lesions, lacunes, and perivascular spaces when measuring volume of white matter hyperintensity to avoid inflating the volume; longitudinal: difference imaging might help to identify new white matter lesions	Inter-rater and intra-rater reliability for both qualitative and quantitative analysis of white matter hyperintensity is high if done by trained raters, with intraclass correlation coefficients generally above 0·90; visual rating scores might have ceiling or floor effect so performance can differ with extent of disease	Careful visual checking is needed at all stages of computational analysis to avoid difficulties from excess lesion distortion by, for example, bias field correction; regular recalibration against standard examples is needed in rating large numbers of scans
Perivascular space	Number (multiplicity); location (anatomical region); size (maximum diameter); shape	Anatomical: midbrain, hippocampus, basal ganglia, centrum semiovale	Visual scores used to rate number of lesions in basal ganglia, centrum semiovale, midbrain;42,45,46 various threshold-based methods are in development	Cross-sectional: consider masking perivascular spaces when measuring volume of white matter hyperintensity, although this might be difficult; longitudinal: little experience	Difficult to determine, especially when numerous, and in the presence of white matter hyperintensities	Can be difficult to distinguish from lacunes; giant perivascular spaces can be greater than 2 cm, and are most commonly located below the putamen
Cerebral microbleed	Number (few or multiple); location (lobar, deep, or infratentorial; anatomical region); size	Some semi-automated approaches segment cerebral microbleeds as an extra tissue class or radial symmetry and mask areas of mineralisation,98,99 but these are experimental at present, and need validation	Several visual scores are available;60,100,101 no methods available for automated detection	Cross-sectional and longitudinal: consider use of visual scores; longitudinal: no specific scores available for longitudinal studies	Inter-rater agreement for the presence or absence of one or two microbleeds varies, but agreement (ie, 0·8) between the numbers of microbleeds is reasonable; reliability can be improved through the use of standardised scales60,100,101	Lobar and deep cerebral microbleeds might have different risk factors and causes (eg, lobar cerebral microbleeds are associated with cerebral amyloid angiopathy)
Brain atrophy	Whole brain should be adjusted for intracranial volume; regional (hippocampus, specific gyri, lobes should be adjusted for whole-brain volume); cortical or subcortical; superficial or deep (sulcal or ventricular enlargement; whole-brain volume adjustment needed)	If scans are not suitable for volumetric techniques or if such techniques are not available, qualitative rating scales could provide an alternative102,103	Automated or semi-automated quantitative methods are preferred but visual checking and manual editing are commonly needed to avoid including the orbits and excluding the brainstem from the whole-brain volume;69,93,104 regional or subregional brain volume computational methods are in development, but their reliability, especially in individuals with disease, is still to be determined78,105	Cross-sectional: brain atrophy can be estimated by comparison with the inner-skull volume (an estimate of maximum brain size in patients at around age 20 years); all intracranial contents must be included in the intracranial volume, including veins and meninges, which expand into space left by shrinking brain;76 longitudinal: serial brain volumes can be measured; a registration-based approach is preferred, although the discipline is advancing rapidly106	Computational approaches have high reliability; visual rating is more varied but can be improved with reference to a standard visual template;103 subcortical and cortical vascular lesions affect the reliability of automated volumetric techniques,79 particularly in subjects with a high lesion load	Consider masking recent small subcortical lesions, lacunes, and perivascular space when measuring brain volume;78,105 specific standards are emerging for hippocampal volume measurement107

DWI=diffusion-weighted imaging.