Dementia imaging in clinical practice: a European-wide survey of 193 centres and conclusions by the ESNR working group

An online questionnaire was designed using Google forms open-access toolbox (Google.​com, Mountain View, CA, USA). Questions were assembled by the members of the ESNR Working Group (authors MWV, FBP, FB, SH) and further modified/added by the ESNR Diagnostic Committee (MWV, TY, MS, NB) as well as by 2 members of the European Association of Neurology (EAN) committee (GF and RS, the latter acting as co-chair of the Panel of Dementia and Cognitive Disorders of the EAN). The questionnaire featured 42 items, divided into multiple choice, single best choice and free text questions on personal practice as well as clinical scenarios (see supplement for entire list of questions).

Survey invitations were emailed to all ESNR members (n = 1662) and non-members who had expressed their interest in ESNR activities in the past (n = 6400), European national neuroradiological societies (the UK, the Netherlands, Belgium, Turkey), and distributed via professional social media channels (LinkedIn, Twitter, Facebook). The survey was open for 2 months from 1 May to 1 July 2017. To avoid duplicate bias, participants were instructed to supply institution details or confirm they were the only person answering from their centre.

A number of questions included the option ‘other’. If this was answered by < 5% of individuals, percentages are not quoted in the results.

The results of the survey were presented by the members of the working group at the 2017 annual meeting of the ESNR in Malmo in a dedicated session that was attended by around a hundred meeting participants. Matters arising in the plenary discussion of that session were considered when interpreting survey results and formulating conclusions and discussed further with the EAN representatives.

The ESNR survey on imaging in dementia had a reasonable number of responses (193 unique European centres), with a good geographical spread including Eastern-Europe. We should consider however that the low response rate (less than 5% of all invitations) will have led to selection bias, likely towards an overrepresentation of those who are very active and potentially have more expertise in this field. Indeed, the majority of respondents were subspecialty-trained neuroradiologists and working in academic centres, seeing roughly five scans per week for patients with dementia. Referrals tended to be from dedicated memory centres, though in non-academic centres, this was balanced with primary care referrals. We tried to take the academic overrepresentation into account by stratifying several results for academic versus non-academic centres and found that these did not differ, pointing towards generalisability and representativeness. Also, the response rate in this survey was very similar to an earlier pan-European survey conducted by the ESNR on glioma imaging [19]. Overall, we feel that the responses provide a reasonably realistic picture of neuroradiology practice for dementia in Europe.

The imaging modality of choice among the survey respondents clearly is MRI and, in those cases where CT was used, motivation varied and included reasons such as limited access to MRI or lack of clinical impact. Some respondents also reported that the information obtained with CT is sufficient, which is indeed justified for exclusion of other treatable causes of cognitive impairment, and to a certain extent also for atrophy patterns and more extensive vascular pathology when using multi-detector CT with coronal and sagittal reformats [20]. Yet, it should be noted that pathology such as microbleeds, areas of diffusion restriction, and more subtle patterns of vascular lesions (white matter pathology) and cortical atrophy are better evaluated on MRI.

There was very high convergence on the MR sequences used, with over 90% of respondents conducting the earlier suggested minimum set of MR imaging sequences for the evaluation of memory clinic patients (Box 1) [3, 4]. This indicates that MR imaging among the European centres that participated in the survey is already fairly homogeneous and meets current acceptable standards in the field.

Box 1. Proposed minimum MR imaging protocol for dementia imaging (adapted from [3, 4])

There was a limited use among survey respondents of advanced sequences, such as DTI and fMRI, in the majority due to lack of interest among clinicians or radiologists (> 70% of non-users), which is probably due to the lack of proven diagnostic performance or clinical impact as well as time intensiveness of the acquisition and post-processing in relation to lack of reimbursement. ASL was used most frequently, probably reflecting its potential to replace fluorodeoxyglucose (FDG)-PET [21] while being non-invasive. In addition to uncertain value of many of the advanced sequences, prolonged acquisition times and lack of tools/support to interpret the data were hurdles that were mentioned by the survey respondents. Inter-vendor differences in acquisition and analysis methods likely play a role as well in the low rates of implementation of advanced sequences.

Interpretation of MRI scans in patients with dementia was often (75% of all institutes) supported by the use of visual rating scales which were used even more often by non-academic centres. The most popular scales used were the MTA for hippocampal and medial temporal involvement (related to Alzheimer’s disease) and the Fazekas scale for white matter lesions of presumed cerebrovascular origin; both can be applied to CT as well [20]. Less frequently used were scales for GCA and posterior atrophy (as can be seen in younger-onset AD patients). Most respondents felt quite comfortable in applying the visual rating scales, although there clearly is a need for further training [15]. Although cut-off values for abnormality are age (and sex)-dependent [22], visual rating scales perform well in discriminating major dementing disorders [14] and have good inter-observer variability [23]. Box 2 summarises the most frequently used visual rating scales.

Box 2. Frequently used visual rating scales

Quantitative methods such as MR volumetry are currently mainly used in research settings. As indicated in our survey, only 5.7% of centres performed volumetric analyses on a regular basis, despite the fact that for example voxel-based morphometry (VBM) was introduced already in 2000 [24], and more user-friendly automated segmentation algorithms such as FreeSurfer were introduced over 20 years ago.

Theoretical advances of implementation of volumetry are its potentially higher sensitivity to subtle changes, yielding an earlier detection of abnormality, an objective observer-independent assessment and higher sensitivity to change over time, e.g. in clinical trials. Yet, evidence from clinical practice supporting these advantages is still scarce and not unequivocal [25‐29].

Main hurdles identified by the respondents were the limited availability of software, lack of time to use (offline) workstations and difficulties in the interpretation. The lack of transition of volumetric techniques into clinical routine is in our opinion furthermore mainly due to key issues that are common to the introduction of new biomarkers. These comprise lack of standardisation, lack of validation, concerns about specificity and the difficulty to translate research findings on a group level to the individual patient. For example, with respect to hippocampal volumetry, the rate of atrophy of AD patients is around 4.25% per year, yet only 1.25% per year in controls, i.e. 3% difference for a follow-up imaging of 1 year [30]. To be able to detect these small differences and apply these findings in a clinical setting would call for standardisation of image acquisition (field strength, vendor type, imaging parameters) as, for example, done in ADNI [31]. In addition, in a single time point setting, the use of reference populations or control groups representative for the population an individual patient was derived from is needed. Likewise, there are multiple software analysis tools and multiple data analysis parameters which may influence the obtained results.

In most instances, referrals were not accompanied by detailed clinical information and only a minority of respondents participated regularly in multidisciplinary meetings where such information would have been available. This most likely reflects the previous practice of performing structural imaging in suspected dementia only with the aim to exclude a surgical lesion. The role of neuroimaging in dementia has however changed importantly with imaging being incorporated in research and diagnostic criteria of several dementia subtypes [5, 7]. This stresses the need for close communication between referring clinicians and radiologists to obtain the most valuable information with respect to differential diagnosis.

In contrast, a general lack of clinical information may also drive a radiologist to scrutinise the images in a systematic fashion (rather than to look for a confirmatory finding) and may be more objective than being potentially misled by the clinical information. For example, in the clinical diagnosis of AD, about a third of cases cannot be confirmed and finding some degree of hippocampal atrophy could be misinterpreted when other diagnoses are not systematically considered (e.g. FTD).

Participation of multidisciplinary team (MDT) meetings was variable, although if these happened, radiologists were frequently included. Recognising that MDT meetings require time for preparation and attendance, we strongly recommend radiologists to participate, as it will allow them to address specific clinical information that may have arisen after the request form was issued or the patient was examined in more detail and the radiological findings can now be interpreted in light of additional laboratory (e.g. CSF) findings. Although numbers are not known for dementia MDT meetings, studies from the field of oncology show that radiological input during MDT meetings results in direct influence on patient management in up to 20% of discussed cases [32, 33]. MDT meetings are also important to generate feedback on the quality of the radiology reports and provide a mechanism to verify the accuracy of the conclusions and the appropriateness of the wording of reports. Finally, they provide an excellent training environment for younger colleagues.

Literature points towards the fact that semi-structured reporting may help to more systematically evaluate scans, remove ambiguity, lead to more clarity towards clinicians and facilitate research [34]. Yet in the majority of responding institutes (72% out of 193), radiologists did not apply structured reporting in dementia imaging, with as most common reasons not being supported by dictation software, being time-consuming or not helpful. Despite this, the frequent use of visual rating scales indicates there at least is a semi-structured approach to the interpretation of images. In fact, use of a standard set of visual rating scales (see Box 2) and measurements is a way to structure the analysis path and build the report systematically. In addition, the use of a common language may further be facilitated by the use of visual rating scales, which are more objective than subjective terms such as ‘mild’, ‘appreciable’ or ‘significant’. Interestingly, the use of semi-quantitative scales seemed to be adopted more widely in non-academic centres for reasons not further explored in the survey.

Finally, a report should always be placed in context of the provided information and question. The conclusion can mention whether there is neurodegeneration (and vascular) pathology beyond expectation for ageing, but, unless specifically asked to confirm a specific diagnosis, should only provide a possible differential diagnosis, to avoid anxiety and confusion among referrers and patients. This caveat becomes increasingly important as patients are getting direct access to reports through electronic health records and may even read the report before seeing the clinician.

Box 3 indicates a suggested structure for reporting in dementia imaging [3, 4].

Box 3. Suggested structure for reporting in dementia imaging (adapted from [3, 4])

Given the survey findings, we conclude that current practice in dementia imaging is fairly homogeneous across Europe, in terms of image acquisition and image interpretation. With respect to protocol set-up, relatively easy adaptation towards the best clinical practice recommendations can be expected, although some variations throughout Europe are likely to remain, depending on reimbursement strategies, practical and logistical setups and availability of scanning facilities. Several hurdles have been identified through the survey that may help progress the field further. These mainly lie in the areas of training on the use of visual rating scales, translation of research findings (advanced imaging) to a clinical setting, implementation of volumetric assessment, structured reporting and communication.