Table 4: Reporting MR Imaging Changes of Presumed Vascular Origin

Supplementary Web Appendix:

Table A1. Expanded summary of terms used to describe MR changes of presumed vascular origin: recent small subcortical infarcts, lacunes of presumed vascular origin, cerebral microbleeds.

Delphi principle used to develop the consensus document

Literature search methods and terms

Further details of recommendations for image analysis of SVD features on neuroimaging

Other vascular lesions that may be relevant to neurodegeneration

Figure A1. Overlap between lesion types

Figure A2: Recent small subcortical infarction (illustrating size in different planes).

Figure A3. Periventricular and deep WMH. EES suggestion example

Figure A4. WMH appearance on T1w MPRAGE and T2 - EES suggestion example

Figure A5. PVS example

References

1

Table A1: Expanded summary of terms used to describe recent small subcortical infarcts, lacunes of presumed vascular origin and cerebral microbleeds: Data were derived from structured literature search; for methodology see Supplementary Text

*Number of instances term was mentioned at least once in the abstract or in the title. The total number of instances was N=495 in a sample of 454 abstracts for recent small subcortical infarcts and lacunes of presumed vascular origin and N=387 in a sample of 370 abstracts for cerebral microbleeds.

1

Delphi principle used to develop the consensus document

The Delphi method is a structured communication technique, originally developed as a systematic, interactive prediction method, which relies on a panel of experts (references in We used the principle of the Delphi approach, adapted for face-to-face meetings:

We used a systematic, transparent, democratic approach, as unbiased as possible, with random allocation of participants to groups throughout. We convened a panel of experts known for their expertise in SVD and other forms of dementia and neurodegeneration such as Alzheimer’s disease. We aimed to identify experts from all major continents within funder criteria and budget limitations. We held two workshops and worked remotely in between. Writing group chairs were selected to chair SVD topic groups that were not their particular feature of interest to avoid strongly held views distorting the consensus.

First workshop, Edinburgh, March 2012:

• A series of questions were posed about each SVD feature to all present, to: summarise current definitions, terminology, image acquisition, image analysis and standards for reporting studies of SVD imaging features; propose consensus terms; and identify outstanding issues and points of contention that would impede consensus;
• Next, the group was split randomly into two subgroups with each subgroup discussing a specific SVD feature; this process was repeated twice in order to discuss each of the six SVD features of interest (each participant contributed to three SVD feature discussions);
• The results of these focussed discussions were then fed back to all participants and discussed;
• Contentious points were flagged for further discussion;

Participants were then randomly assigned to one topic writing group, with a chair, to prepare the draft text of their topic section. Gaps in the group were identified and additional experts identified to fill those gaps. The drafts were circulated prior to the second workshop. Six additional independent expert advisors were identified to attend the second workshop and independently critique the proposed standards.

Second workshop, Munich, November 2012:

• The draft statements were presented to all participants, discussed and consensus agreed.
• The independent expert advisors provided critique and suggestions for improvement.
• All participants, including those who were unable to attend the second workshop, provided written feedback and commented on the draft

Thereafter a revised draft of the consensus statement was circulated for comment and, after a further iteration, final sign off.
Literature search methods and terms

We searched the literature using the Pubmed search engine from Jan 1 1980 to April 17 2012, run on April 18 2012, restricted to human studies published in English (search terms see below) to cover the six SVD components.

The MeSH terms "cerebral small vessel disease" and "vascular dementia" were used to ensure potentially relevant articles, and the MeSH terms "lacunar stroke" and "leukoaraiosis" were used where relevant. All terms were exploded.

We adjusted the sensitivity and specificity of the terms to optimise the number of hits while verifying that known relevant articles were included. For most lesions this produced a manageable number of hits (<1000), the exception being acute and chronic lacunar infarcts which produced a very large number of articles (>2000). After an initial run, several additional terms were added to the search for terms related to acute and chronic lacunar lesions. The final search terms and their initial yields are as follows:

Recent small subcortical infarcts and lacuness – yields 2303 articles:

(Lacun* OR deep infarct* OR subcortical infarct* OR deep stroke* OR subcortical stroke* OR silent stroke* OR silent brain infarct* OR small vessel infarct* or small vessel stroke* OR lacunar stroke OR microinfarct* or microscopic infarct* OR etat crible) AND (brain OR cerebr* OR lacunar stroke OR cerebral small vessel disease OR vascular dementia OR stroke) AND (MRI OR computed tomography).

WMH - yields 967 articles

((White matter hyperintens*) OR (white matter lesion*) OR (white matter disease*) OR (white matter change*) OR (leukoaraiosis)) AND ((leukoaraiosis) OR (cerebral small vessel disease) OR (vascular dementia)) AND (MRI OR computed tomography)

Perivascular spaces - yields 256 articles

((Perivascular space*) OR (Virchow Robin) OR (Virchow-Robin)) AND (brain OR cerebr* OR leukoaraiosis OR cerebral small vessel disease OR vascular dementia) AND (MRI)

Microbleeds - yields 367 articles

((microbleed*) OR (microhemorrhag*) OR (microhaemorrhag*) OR ((“dot-like”) AND (suscept* OR hemosid*))) AND (brain OR cerebr* OR cerebral small vessel disease OR vascular dementia) AND (MRI)

Atrophy - yields 464 articles

(Atrophy OR brain volum* OR cerebral volum* OR volume loss) AND (leukoaraiosis OR cerebral small vessel disease OR lacunar stroke OR vascular dementia) AND (MRI OR computed tomography)

Note: expanding the search to include "dementia" and "Alzheimer's" as well as "vascular dementia" yields many more articles, 3,203

We distributed the search results to the work group chairs for use in formulating their recommendations.

Additionally, for four of the lesions, we performed an analysis of the frequency of terms. For recent small subcortical infarcts and lacunes there were 2303 potentially relevant abstracts identified. Two reviewers (KS, FD) evaluated 641/2303 (28%): the first 480 consecutive abstracts and every 10th abstract thereafter, which yielded 142 terms for small deep infarcts (not including 17 terms that had been identified in previous work3) giving a total of 159 terms. Amongst these the commonest were lacunar infarcts (127), lacunar infarctions (77), lacunar strokes (60) lacunes (50) and subcortical (50). “Small deep infarcts” occurred 10 times and there were more than 100 terms that have been used to describe lacunes of presumed vascular origin.

For WMH and microbleeds, one reviewer (EES) performed a similar analysis. For WMH, There were a total of 1,144 instances of 50 different terms for WMH used in the 920 abstracts identified in the literature search. The number of instances is greater than the number of abstracts because in some cases 2 different terms for WMH were used in the same abstract. Generally the terms were grouped around 5 major "families"--in order, leukoaraiosis (including mis-spellings), white matter lesions (and derivatives), WMH, leukoencephalopathy (mostly from older CADASIL literature) and white matter disease.

For CMBs, there were a total of 387 instances of 20 different terms for MB used in the 370 abstracts reviewed. ‘Microbleed’ was by far the most frequent, with "cerebral microbleed" appearing more commonly than "brain microbleed".

These results are summarised in main manuscript Table 1 (WMH) and Supplementary Table 1 (recent SSI and lacunes of presumed vascular origin).

Further details of recommendations for image analysis of SVD features on neuroimaging

Recent small subcortical infarcts: Visual assessment of acute small deep brain infarcts on DWI and structural sequences is the reference standard for image analysis of acute small deep brain infarcts at present. Observers should be trained in brain imaging interpretation and specifically in how to recognise acute infarcts. Various schemes have been described for classifying lesion location, size, shape, visibility on different sequences, etc, for quantification of acute as well as established SVD lesions in research (examples available at Points to consider in image analysis include recording the shape (eg ovoid, or tubular running the length of a perforating arteriole) (Supplementary figure 2). Lesion location should be specified (centrum semiovale; corona radiata, basal ganglia; thalamus, internal capsule, external capsule, optic radiation, cerebellum, brain stem). Multiplicity of acute lesions should be described. A small proportion of acute symptomatic lacunar lesions are accompanied by other acute lesions, the appearance of which indicates that they occurred simultaneously or at least within a few days. Multiplicity might suggest a proximal embolic source,2,3 although other mechanisms might also cause such appearances.4,5 Acute lesion volume may be a useful measure in future but the reliability and optimal method has not yet been fully explored.

Lacunes of presumed vascular origin: Image analysis of lacunes is very dependent on visual scoring; although lacune volume can be measured, there is as yet no established computational analysis approach. It is important to distinguish lacunes from PVS, given the likely difference in aetiology. This is poorly done at present;6 but might improve with a proposed size boundary of 3mm or more for lacunes and <3mm for enlarged PVS.

Visual scoring systems should document number, location, size of cavitated lesions, and evidence for previous haemorrhage associated with the cavitated lesions. T1 and T2 are more sensitive to cavitation than FLAIR and variation in cavitation rates may be due to whether lesions with a central “lacy” appearance due to several tiny punctate holes or only where the entire lesion is a hole are counted as cavities. Prominent ex-vacuo effect around a lacune should raise the possibility that the original lesion was much larger when acute, too large to have resulted from disease in a single perforator and more likely to have been striatocapsular in origin.7 Computational image processing includes 3D delineation of cavities8 but these are non standard and have so far only been applied in selected populations. The fate of lacunes in image segmentation algorithms are not known but it is suspected that they are classed as CSF9 and therefore contribute to assessment of brain atrophy. Whether this is appropriate or not requires further study.

White matter hyperintensities of presumed vascular origin: WMH of presumed vascular origin may be assessed using qualitative visual rating scales or computational quantitative image analysis measurements of lesion volume. Visual rating and volumetric measures both have advantages and disadvantages, but in direct comparisons, both are highly complementary.10 The decision as to which to use will depend on the type of research, number of scans to analyse, type of scans, resources available, etc. In general, volumetric analysis requires much more consistent, less flexible image acquisition than does visual rating, and volumetric analysis may be time consuming allowing time for visual correction of erroneously included tissue.

Numerous visual rating scales have been described. Some semiquantitative scales have the disadvantage of ceiling effects. Some do not differentiate periventricular from deep WMH.11 Some provide subregional rating of WMH which may be helpful in certain situations.12,13 Some convert the visual rating into an estimate of lesion volume.14,15 A wide range of scales have been compared.15,16 Visual rating should be performed by trained raters tested against established training sets and should aim for high observer agreement before undertaking actual ratings. Regular ‘recalibration’ against standard examples is a useful trick for maintaining consistency when rating large numbers of scans or when rating scans in different sessions. .

Numerous computational quantitative semi-automated methods have been developed.17-19 Most are used by a single research group without validation by other groups, and few have been compared with each other. None of these methods are able to distinguish reliably artifacts from true WMH, or infarcts from WMH and therefore all computational methods require the user to inspect and edit the images to avoid errors. This is likely always to be the case and emphasizes the need for such image analysis to be performed by raters trained in the identification of artifacts, stroke lesions, true WMH, etc. A few methods may differentiate periventricular from deep lesions.20,21

The inter and intra-rater reliability for both qualitative and quantitative analysis of WMH should be quite high if performed by trained raters, with ICCs generally above 0.90.

Perivascular spaces: An ideal and meaningful analysis procedure for PVS has not yet been identified. Thus far, the assessment of PVS has been done by visual grading, which concentrates on qualitative aspects, i.e. mostly PVS number, location and size. Several grading/scoring schemes have been employed to date.22-30 Broadly these reflect attempts to collect similar features from similar locations, usually basal ganglia and centrum semiovale, although some investigators also examine the midbrain and hippocampus. Most classify PVS frequency in a chosen anatomical location according to a three or four-point scale. However the specifics of which slice in which anatomical location have not yet been refined for maximum repeatability. The scales also differ in key respects (eg use of T1 only or T2). Many have not undergone independent validation and are only in use by the group where they were developed. More recently, dynamic changes in the size and volume of PVS in patients with MS have been determined using semi-automatic threshold-based image processing.31 PVS volume measurements in patients with vascular disease are in development.32

Cerebral microbleeds: Currently, the gold standard for assessment is by raters, well trained in the appearance of brain microbleeds and their mimics, with lesions quantified by number and anatomical location (especially “lobar” verses “deep”). There is some variability in agreement between raters on the absence/presence of one or two brain microbleeds, but reasonable intraclass correlation (i.e. 0.8) for numbers of brain microbleeds. Reliability can be improved through the use of standardised scales for their identification33,34 Semi-automated approaches, which segment brain microbleeds as an extra tissue class or radial symmetry and mask out areas of mineralisation, are being described35,36 but are experimental at the present time and require validation.

Cerebral atrophy: Quantitative volumetric techniques are preferred for the assessment of atrophy on brain MRI for research. For regional measurements, manual delineation is still the gold standard. For hippocampus volumetry, an international effort is under way to define a harmonized hippocampus protocol.37 A large variety of (semi-)automated techniques is available,38,39 and important in making it feasible to process large numbers of scans within a reasonable time frame. The selection of a suitable technique should match the scan parameters and should be guided by the intended measurements; i.e. single time-point versus serial studies, global versus regional measurements. Given the importance of the association between brain size and cognition, the variation in head size and the importance in relation to studies of cognition in neurodegeneration, we suggest that brain size should always be adjusted for ICV as a marker of original brain size. In this context, the accuracy and precision of the measurement of cranial volume is equally important as that of the brain, since both measures together the accuracy of the atrophy measure. Studies, whether cross-sectional or longitudinal, should adjust for intracranial volume -either by a measure of total intracranial volume or by registration of skull/scalp to a template to provide such an estimate. Regional effects can be assessed in isolation or relative to global brain volume (accessible through automated segmentation). For serial studies a registration-based approach is currently preferred although the field is rapidly advancing. Improvements in techniques – such as template based segmentation - mean that single time point delineation and volume estimation are becoming more reliable.40 Caution is required for longitudinal studies with unequal time intervals between measurements and some softwares are not well suited to this analysis. Importantly, subcortical and cortical vascular lesions affect the reliability of automated volumetric techniques, particularly in subjects with a high lesion load.41,42 Analytic methods for measuring brain volumes and change over time in AD have been studied in a number of studies for example the AD Neuroimaging Initiative (ADNI); however, to date there is no study of similar scope for the vascular-associated neurodegenerative changes.