Which neuroimaging techniques?
Exclusion of a potentially (surgically) treatable cause of dementia (e.g. tumour or subdural haematoma) and evaluation of the presence and extent of cerebrovascular disease can be ascertained using computed tomography (CT). However, MRI offers benefits over CT for the demonstration of markers of specific diseases, particularly cerebral atrophy patterns [e.g. hippocampal atrophy for AD; very focal temporal and/or frontal atrophy for frontotemporal dementia (FTD) and midbrain atrophy for progressive supranuclear palsy (PSP)] [10-12]. Therefore, MRI should be considered the ‘preferred modality to assist with early diagnosis’ in a subject with dementia [2, 3].
The essential MR sequences that will provide the important minimum set of information required to be addressed in a subject suspected of having dementia are 3D T1-weighted gradient echo; turbo/fast spin echo T2-weighted and fluid attenuated inversion recovery (FLAIR) and T2*-gradient echo . If 3D T1-weighted techniques are unavailable, coronal oblique 2D images can serve as an alternative. A multiplanar reformatting tool can be applied to such images to assess specific brain regions (e.g. to reslice the data on the anterior/posterior commissure line or perpendicularly to the long axis of the hippocampus). Two other MR techniques that are very frequently used in a clinical setting include diffusion-weighted imaging (DWI) and post-contrast 2D T1-weighted spin echo images . DWI can be useful to identify recent infarcts in patients with vascular dementia, transient global amnesia or in the context of vasculitis and to identify neocortical or striatal abnormalities in patients with Creutzfeldt–Jakob disease (CJD). Post-contrast T1-weighted images are recommended in those patients, typically younger, where there is a suspicion of infectious (e.g. herpes simplex virus encephalitis) or inflammatory disorders (e.g. vasculitis, sarcoid, multiple sclerosis).
Single photon emission computed tomography (SPECT) and PET both rely on the detection of radioactive signals from a labelled compound that selectively binds in the brain. The most commonly used tracer to examine cerebral blood flow (CBF) using SPECT is 99m-Tc-hexamethylpropylene (HMPAO). FDG serves as a marker of cerebral glucose metabolism for PET. SPECT is technically less demanding and more widely available, whilst PET is more sensitive, mainly because of its higher resolution [13-16], but comes at the cost of more complex detector system and tracer production facilities. In general, the magnitude of hypometabolism seen with FDG PET is greater than the amplitude of hypoperfusion seen with CBF SPECT . SPECT and PET images can be analysed for diagnostic purposes by means of visual inspection. However, the visual method greatly depends on the observer's experience and lacks a clear cut-off between normal and pathological findings. To overcome these limitations, FDG PET images can be assessed using software that analyses the pattern of tracer uptake voxel-wise by comparing the subject's scan with a reference data set of normal ageing. T-maps (statistical maps of level of significance based on t tests) are created that allow better recognition of the pattern of hypometabolism compared with visual interpretation .
Structural MRI findings
Vascular brain diseases
Cerebrovascular disease can be detected by CT and structural MRI. Although both modalities perform relatively well in depicting large-vessel infarcts, MRI is more sensitive to subtle small-vessel vascular changes than CT. T2-weighted and FLAIR sequences are highly sensitive for detecting major strokes as well as small strategic infarcts and small-vessel ischaemic white matter damage. FLAIR performs less well than T2-weighted sequences in detecting thalamic infarcts . Extensive white matter changes visible as diffuse hyperintense abnormalities on T2 and FLAIR, predominantly involving the periventricular and deep white matter, but relatively sparing of the U-fibres, are the imaging correlate of Binswanger's disease . Marked hypointensity on T1-weighted images usually represents tissue destruction in the presence of a complete large-vessel infarct, whereas white matter changes are usually not prominently hypointense on T1-weighted images. Finally, lacunar infarcts are focal complete infarcts of deep small vessels, which are hyperintense on T2-weighted images and markedly hypointense on T1 and FLAIR images . On FLAIR, lacunae are often surrounded by a hyperintense rim.
According to the National Institute for Neurological Disorders and Stroke Association pour la Recherche l'Enseignement en Neurosciences (NINDS-AIREN) international criteria , structural brain imaging is an essential element for the diagnosis of vascular dementia, and without it vascular dementia will be ‘possible’ at best. In addition, the operational radiological definitions for the NINDS-AIREN criteria provided indications on the topography and severity of vascular lesions . Bilateral infarcts in the area of the anterior cerebral artery, infarcts in the area of the posterior cerebral artery, association areas or watershed regions are thought to be causative of large-vessel vascular dementia . Extensive white matter lesions involving at least 25% of the white matter, or multiple basal ganglia, thalamic and frontal white matter lacunar infarcts, or bilateral thalamic lesions, are considered relevant radiological lesions associated with small-vessel vascular dementia . Research criteria to specifically diagnose subcortical ischaemic vascular dementia have also been proposed : the presence of extensive periventricular and deep white matter lesions and lacunar infarcts in the deep grey matter or multiple lacunae in the deep grey matter and at least moderate white matter lesions, in the absence of cortical and/or cortical–subcortical (non-lacunar) territorial infarcts, watershed infarcts, haemorrhages and other specific causes of white matter lesions.
Different methods can be used to measure the extent of white matter changes to diagnose subcortical ischaemic vascular dementia. Visual rating of white matter hyperintensities is relatively easy, and several scales [22-24] are available with good reproducibility. Volumetric studies use semi-automatic techniques that may provide more information on location and size, as well as continuous data, but are time consuming . With the Age-Related White Matter Changes (ARWMC) scale , for instance, a score of 3 in at least two regions and a score of 2 in two other regions could be sufficient for a diagnosis of subcortical vascular dementia . A conversion table amongst some of the most popular scales to rate white matter lesions can be found in Frisoni et al. .
Cerebral microbleeds (or microhaemorrhages) are small, rounded, dot-like lesions of low signal intensity in the brain that can be observed on T2*-weighted images, such as gradient echo [27-29]. Susceptibility-weighted imaging has considerably increased microbleed detection rates compared with gradient echo sequences [30, 31]. The sensitivity to detect microbleeds is also dependent on slice thickness and magnetic field strength . Microbleeds in deep brain regions are most likely to be associated with vasculopathy owing to hypertension, whilst their distribution is mostly lobar in specific disorders such as sporadic cerebral amyloid angiopathy (CAA) . Sporadic CAA is the most common cause of lobar intracerebral haemorrhages in the elderly and results from cerebrovascular deposition of β-amyloid protein in the media and adventitia of small- and medium-sized vessels of the superficial layers of the cerebral cortex and leptomeninges, with sparing of the deep grey matter nuclei . According to a set of validated criteria (termed Boston criteria) , a diagnosis of probable CAA can be reached in elderly patients with at least two acute or chronic lobar haemorrhagic lesions (including microbleeds), in the absence of other definite cause of intracerebral haemorrhage.
The presence of extensive white matter lesions and multiple bilateral, lacunar infarcts on T2-weighted and FLAIR images is critical for the diagnosis of cerebral autosomal dominant arteriopathy with subcortical infarcts and leucoencephalopathy (CADASIL), a genetic form of vascular dementia in relatively young people . The most distinctive MRI features suggestive of CADASIL are the presence of T2 hyperintensity of the temporal pole, the U-fibres at the vertex and external capsule or insular region, and multiple microhaemorrhagic foci (basal ganglia, internal capsule, thalamus and pons), which can be seen on gradient echo scans .
Although an increased white matter lesion load suggests vascular disease, particularly in combination with lesions in the basal ganglia, a critical clinical challenge in subjects with vascular lesions is determining the relationship of cerebrovascular disease and cognitive symptoms. To appropriately diagnose vascular cognitive impairment or dementia, there should be a clear relationship in the severity and pattern of cognitive impairment and the presence of vascular lesions with neuroimaging . However, this association can be confounded by the frequent co-occurrence of depression on a cerebrovascular basis, as well as by the frequent coexistence with neurodegenerative diseases, especially AD. Small-vessel disease is frequently observed in MRI scans of AD patients in the form of white matter hyperintensities, lacunae and microbleeds. Several studies have indicated that the prevalence of microbleeds in AD ranges from 15% to 32% [37-41]. Microbleeds are also considerably more frequent in AD compared with other neurodegenerative dementia .
In typical late-onset (arbitrarily defined as age at onset >65 years) AD, the medial temporal lobes (MTL), especially the hippocampus and entorhinal cortex, are amongst the earliest sites of pathological involvement [42, 43]. Other severely affected regions include the posterior portion of the cingulate gyrus and the precuneus on the medial surface [44-46] and the parietal, posterior superior temporal and frontal regions on the lateral cerebral surfaces [42, 47-49].
Structural MRI studies in mild cognitive impairment (MCI) have produced mixed results, in terms of hippocampal as well as posterior cingulate and parietal involvement (absent, unilateral or bilateral) [50-60]. The reasons for this variability may consist of different subject selection (i.e. diverse diagnostic inclusion criteria), small sample size (i.e. studies are not adequately powered to pick up differences even at the group level) and methodological differences. It is also worth noting that the largest source of variance in MCI studies is likely to be the intrinsic heterogeneity of the MCI population, because a relevant proportion of these subjects will not progress to dementia. MCI patients with predominant memory impairment (amnestic MCI), who are at increased risk of developing AD, have atrophy in a consistent set of cortical regions, the ‘cortical signature of AD’, including the MTL and temporoparietal cortex. Conversely, non-amnestic MCI shows a different pattern of atrophy characterized by a relative sparing of the MTL and a regional involvement which is typically highly consistent with the observed clinical deficits.
Of all the structural markers of AD, hippocampal atrophy assessed on conventional CT or coronal T1-weighted images is the best established and validated. MRI-autopsy studies have convincingly validated that hippocampal volumes measured from antemortem MRI scans correlate with Braak neurofibrillary tangle pathological staging [61-63]. Indeed, MTL atrophy is now one of the supportive biomarkers to make a diagnosis of AD in the presence of memory loss proposed by the new diagnostic criteria [6-8]. MRI is superior to conventional CT in the evaluation of MTL atrophy [10-12]. However, the possibility of evaluating the pattern of atrophy using CT has been improved with the advent of multi-detector row CT, owing to the availability of high-resolution coronally reformatted images . MTL atrophy can be detected by qualitative ratings based on visual scoring, or by linear measurements and quantitative volumetry of regions of interest (referenced to a well-characterized population with age norms). Several visual rating scales to quantify degree of MTL atrophy have been developed and are widely used [65-68]. On the contrary, at present, accepted standards for quantitative analysis are lacking . Manual hippocampal segmentation is the most validated procedure to estimate a quantitative hippocampal volume, but different laboratories use different anatomical landmarks and measurement approaches . Of note, a harmonized protocol for manual hippocampal segmentation is currently being developed by an international working group (EADC-ADNI), which will be available for general use in 2013 . Furthermore, the utility of structural imaging will certainly be increased further by automated segmentation algorithms developed in the last few years, which are the only feasible methods in the context of large studies [71, 72], such as clinical trials.
Clinical population studies have reported that hippocampal volumes in mild AD patients are 15–40% smaller than controls , and in MCI the volume is reduced by 10–15% . MTL atrophy can separate mild to moderate AD patients from normal controls with sensitivity and specificity >85% . Structural MRI estimates of tissue loss in MTL are predictive for conversion from amnestic MCI to AD [74-83]. A meta-analysis estimated that MTL atrophy, as assessed on structural MRI, has 73% sensitivity and 81% specificity for predicting whether patients with MCI will convert to AD . Subjects with amnestic MCI who progress to AD also have a greater degree of grey matter atrophy at baseline beyond the MTL than MCI who do not, including the medial and inferior temporal lobes, temporoparietal association neocortex and frontal lobes [50, 52, 85, 86].
It should be emphasized that MTL atrophy may occur in other diseases as well [87-92]; thus, MTL atrophy alone lacks the specificity to confidently exclude other dementia, in particular in patients at the MCI stage.
Early-onset AD patients (i.e. subjects showing onset of symptoms before the age of 65 years) showed less prominent MTL atrophy and greater involvement of the parietal, lateral temporal and frontal regions compared with late-onset AD cases [91, 93-95]. Methods such as voxel-based morphometry (VBM) are a popular and successful way to test for groupwise differences in the topography of atrophy beyond the MTL. However, the statistical testing portion of VBM is not designed to provide diagnostic information at the single subject level. A specific visual rating scale has been designed, evaluating the posterior cingulate, precuneus and superior parietal regions . The utility of such a scale has been assessed in pathologically proven (mostly early-onset) AD and frontotemporal lobar degeneration (FTLD) patients . Thirty per cent of AD patients had posterior atrophy in the absence of abnormal MTL atrophy, whereas only 7% of the FTLD group had abnormal posterior atrophy score and normal MTL . Adding the posterior atrophy to the MTL visual rating score improved discrimination of early-onset AD from normal controls and all AD from FTLD cases .
Atypical presentations of AD
AD pathology can manifest itself with clinically atypical presentations, that is, in some patients memory is not the primary deficit but visuospatial and visuoperceptual and/or language disturbances are prominent symptoms [98-100]. Atypical presentations are more often seen in early-onset AD patients . In atypical, focal AD presentations, the MTL is relatively spared [99, 101]. Two recent quantitative MRI studies in pathology-proven AD cases suggested that a pattern of temporoparietal atrophy or cortical thinning may suggest AD pathology even in subjects presenting with non-amnestic clinical syndromes [101, 102]. The temporoparietal cortex volume also provided better discrimination between atypical AD and FTLD groups than the hippocampal volume (81% vs. 74% accuracy) . Two common progressive, focal cortical syndromes associated with AD pathology are posterior cortical atrophy (PCA) and primary progressive aphasia (PPA). Structural MRI scans of patients with PCA show atrophy of parieto-occipital and posterior temporal cortices [99, 103]. Compared with typical AD cases, PCA patients had greater right parietal and less left MTL atrophy . For a detailed description of structural neuroimaging features in AD cases with prominent language deficits, see the section on PPA.
Dementia with Lewy bodies
No clear signature pattern of cerebral atrophy associated with dementia with Lewy bodies (DLB) has been established so far. Similar to AD, a diffuse pattern of global grey matter atrophy including temporal, parietal, frontal and insular cortices may occur in DLB [104-106], but at the same time, a pattern of cortical grey matter loss restricted to frontal and parietal lobes has also been reported [107, 108]. On the whole, several volumetric studies have not found significant or disproportionate occipital atrophy in DLB [106, 107, 109].
A relatively robust MR finding in DLB is that of relative preservation of the MTL compared with AD of similar clinical severity [87, 104-108, 110-115]. This finding is supported by a prospective MRI study with pathological verification, which found that MTL atrophy on MRI has a robust discriminatory power for distinguishing AD from DLB (sensitivity of 91% and specificity of 94%) . Thus, a relative preservation of MTL structures on CT or MRI supports a diagnosis of DLB in the consensus diagnostic criteria .
Subcortical structural alterations in terms of putamen atrophy have been described in some cases of DLB relative to AD , whilst no significant atrophy was detected in caudate nucleus [117-119]. A pattern of relatively focused atrophy of the midbrain, hypothalamus and substantia innominata, with a relative sparing of the hippocampus and temporoparietal cortex, has been found in DLB compared with AD cases . However, whether these findings help in recognizing early suspected cases remains unknown. Furthermore, a substantial overlap between DLB and AD with regard to atrophy in these regions detracts from the usefulness of these markers in individual cases.
The terms FTLD and FTD describe a group of clinical syndromes which may be produced by a number of histopathologically distinct entities. In these guidelines, the term FTLD will be used to indicate cases with a pathological diagnosis, whilst the term FTD will be used to refer to the clinical syndrome.
The designation of probable behavioural variant of frontotemporal dementia (bvFTD) by the revised diagnostic criteria  restricts diagnosis to patients with demonstrable functional decline and typical neuroimaging findings, including frontal and/or temporal atrophy, and hypoperfusion or hypometabolism on PET or SPECT.
Structural MRI studies showed that classical bvFTD presents with a combination of medial frontal, orbital–insular and anterior temporal cortical atrophy [121-128]. Such an atrophy pattern can be readily appreciated on coronal T1-weighted MRI scans (knife-edge atrophy). The MTL is more affected anteriorly, that is, the amygdala is more affected than the hippocampus, and posterior hippocampus often appears normal. Nevertheless, the typical pattern is not necessarily present in all cases [129-131], particularly in patients with FTD and motorneuron disease, and the pattern of atrophy in bvFTD varies significantly across different cohorts [132, 133]. In some cases, bvFTD presents with a remarkable atrophy of the right anterior temporal lobe and a lesser involvement of the frontal regions [127, 134]. A large VBM study suggested that bvFTD may be divided into four anatomically different subtypes, two of which are associated with a prominent frontal atrophy (i.e. frontal dominant and frontotemporal variants) and two with prominent temporal lobe atrophy (i.e. temporal dominant and temporofrontoparietal subtypes) . Brain atrophy in bvFTD also involves several subcortical structures, such as the striatum [122, 125, 135], thalamus [125, 135, 136], bilaterally and hypothalamus . Significant atrophy was also found in the brainstem, including the midbrain and pontine tegmentum in some cases [125, 138, 139].
Despite variation and overlap of atrophy patterns, visual inspection of regional atrophy on MRI may aid in discriminating FTD from AD. A combined diagnostic criterion based on the finding of either severe frontal atrophy or asymmetry was highly diagnostic (sensitivity 71%, specificity 93%) of bvFTD compared with non-FTD dementia cases (i.e. AD and vascular dementia) . In a study investigating the diagnostic accuracy of visual inspection of MRI scans in patients with pathologically confirmed diagnosis, atrophy of the anterior, inferior and lateral temporal lobes was associated with the highest sensitivity (≥90%) for discriminating FTLD from AD patients, and anterior greater than posterior gradient and hemispheric asymmetry of atrophy were each at least 85% specific for FTLD versus AD . A prospective study in 134 patients with clinically suspected bvFTD demonstrated that evidence of frontotemporal atrophy on structural MRI scans, in the absence of corresponding changes in more posterior areas of the brain, had a sensitivity of 63% and specificity of 70% against a clinical diagnosis after 2 years .
Recent quantitative studies have suggested that a cortical thinning of the anterior temporal lobe and frontal lobe is indicative of the presence of FTLD pathology in patients with a clinical diagnosis of bvFTD or PPA during life . Conversely, a cortical thinning or atrophy in the posterior cingulate gyrus, parietal lobe and frontal pole is suggestive of AD pathology independent of clinical presentations (i.e. even in patients with behavioural or language deficits) [101, 102] (see also sections on atypical AD presentations and PPA).
Primary progressive aphasia
In patients clinically diagnosed with PPA, who are then divided into clinical variants based on specific speech and language features characteristic for each subtype, an ‘imaging-supported’ diagnosis can be made if the expected pattern of focal atrophy on structural MRI scans (or functional involvement on SPECT and FDG) is found .
Semantic variant PPA is associated with left anterior temporal atrophy (temporal pole), affecting the lateral and ventral temporal surfaces, as well as particularly the anterior hippocampus, amygdala and fusiform gyrus [121, 143-145]. Semantic patients may have left hippocampal atrophy that is at least as severe as that seen in AD patients [49, 90, 91, 144]. In these patients, the hippocampal atrophy is predominantly located anteriorly, with a relative preservation of the posterior hippocampal regions [49, 90]. Temporal lobe atrophy is also mainly inferior (often severe involvement of the fusiform gyrus) with relative sparing of the superior temporal gyrus [49, 90]. As the disease progresses, the right temporal lobe becomes more involved .
The non-fluent PPA variant is associated with a characteristic pattern of left anterior peri-Sylvian atrophy involving inferior, opercular and insular portions of the frontal lobe . Motor and premotor regions and Broca's area are also involved . In non-fluent patients, bilateral atrophy of the basal ganglia [135, 145], thalamus  and amygdala  was observed. Compared with controls, non-fluent patients also have atrophy of the left hippocampus ; however, it is less severe than that in AD patients .
In the logopenic PPA variant, the pattern of atrophy primarily affects the left temporoparietal junction, including the left posterior superior and middle temporal gyri, as well as the inferior parietal lobule [99, 145, 147, 148]. The involvement of the left MTL is reported less consistently . Such a posterior temporoparietal pattern of atrophy chiefly discriminates this syndrome from the other subtypes of PPA .
A major clinical challenge, to date, is the need to improve the prediction of the specific histopathology causing each of the PPA variants during life (FTLD versus AD). A prominent left frontal and anterior temporal atrophy is seen in semantic PPA patients who have FTLD pathology at autopsy [149-154]. Conversely, semantic patients with AD pathology mostly have hippocampal involvement and a lack of the knife-edge anterior temporal atrophy, as well as a less severe thinning of the temporal lobe in the regions of the collateral sulcus and fusiform gyrus . Some studies evaluating patients with non-semantic PPA associated with AD pathology demonstrated the presence of temporoparietal atrophy [151, 152, 155, 156]. Two recent studies investigated the value of clinical phenotyping, neuropsychological analysis and pattern of MRI atrophy in predicting underlying pathology of non-fluent and logopenic patients [155, 156]. In the first study, PPA cases with CSF findings consistent with AD showed a posterior superior temporal atrophy, whilst patients with FTLD pathology had frontal atrophy . The second study showed that a disproportionate or asymmetrical frontotemporal atrophy on structural MRI (knife-edge atrophy) was 100% specific for FTLD pathology but was only present in a minority of patients (sensitivity 40%) . Altogether, these studies in pathologically proven cases suggested that distinct patterns of tissue loss could assist in the in vivo prediction of underlying pathology. However, the results of these studies are limited by the small numbers of patients assessed.
Midbrain atrophy, better seen on sagittal T1-weighted images, dilatation of the third ventricle, atrophy of the superior cerebellar peduncle and frontal cortical atrophy support a diagnosis of PSP . T2-signal change in the superior cerebellar peduncle can be seen in PSP patients but is less sensitive . Quantitative MRI measurements of brainstem structures have been proposed as potentially useful markers to diagnose PSP on an individual patient basis. In particular, a ratio of linear measurements (e.g. the so-called MR parkinsonism index, which combines measurements of midbrain and pons areas as well as superior and middle cerebellar peduncle widths) has been shown to differentiate accurately PSP from Parkinson's disease (PD) and multiple system atrophy (MSA) cases [158, 159].
Other specific imaging signs may include bilateral striatal atrophy in Huntington's disease, sometimes many years before disease onset , and striatal or neocortical abnormalities in patients with CJD. In CJD, T2-weighted and especially FLAIR sequences can show a very characteristic pattern of hyperintense signal in the striatum and/or cortex . DWI can show focal changes in CJD not yet apparent on FLAIR images (up to 20% of cases) [161, 162]. In sporadic CJD, involvement of either the striatum or neocortex or both is usually found . In variant CJD, there is a selective involvement of the medial and dorsal (pulvinar) thalamic nuclei, leading to the so-called hockey stick sign .
Recommendations for structural MRI
- Structural imaging should be carried out at least once in the diagnostic work-up of patients with cognitive impairment and serves at least three purposes: to exclude other potentially treatable diseases, to recognize vascular lesions and to identify specific findings to help distinguish different forms of neurodegenerative types of dementia (good practice point).
- MRI is currently the imaging modality of choice for assessing subjects with suspected dementia. However, where MRI is not available or contraindicated, CT scans can usefully exclude major space occupying lesions, large infarcts and hydrocephalus (good practice point). Multi-detector row CT is the best alternative for patients who cannot undergo MRI (good practice point).
- A standard MRI protocol should include a high-resolution structural volumetric T1-weighted scan, transverse T2-weighted and FLAIR sequences and transverse T2*-gradient echo sequence (good practice point). Routine contrast administration is not indicated (good practice point). DWI can be useful to identify recent infarcts, as well as cortical and/or basal ganglia changes in CJD patients (good practice point).
- It is particularly difficult to attribute clinical significance to evidence of cerebrovascular disease in patients with cognitive impairment. Vascular changes on CT or MRI do not preclude a diagnosis of degenerative dementia, especially in older age. A diagnosis of vascular dementia should only be made where the vascular lesion(s) can explain the cognitive deficit (class II, level A). The ‘mixed dementia’ label should be reserved for those cases in which both clinical features and diagnostic markers point to a mixed aetiology (good practice point).
- T1-weighted images should be carefully evaluated to assess specific patterns of focal atrophy, especially in the MTL, biparietal regions and posterior cingulate cortex (as seen in AD), temporal pole and/or frontal lobes (as seen in FTD), parietal/occipital lobe (as seen in PCA), putamen, and midbrain and frontal lobe (as seen in PSP) (good practice point).
- Coronal T1-weighted sequence can be used to assess MTL atrophy to support a clinical diagnosis of AD compared with cognitively normal subjects (class II, level A). Prediction of subsequent AD in individuals with amnestic MCI can also be obtained with MRI volumetric measures of the MTL (class II, level A). However, at present, accepted standards for quantitative MTL volume measurement are lacking. Therefore, quantification must rely on local specific standards (good practice point).
- Combining MTL measures with other potentially informative markers, such as posterior cingulate cortex and precuneus volumetric measures, are likely to improve diagnostic confidence in AD patients (class II, level B), mainly in younger cases.
- In cases of atypical AD presentations, the involvement of the MTL is reported less consistently than that of lateral temporal and medial parietal regions (class III, level B).
- No established structural MRI pattern is characteristic for DLB (class II, level A). However, the absence of MTL atrophy on CT or MRI may be suggestive of a diagnosis of DLB compared with AD (class II, level A).
- The pattern of atrophy is more useful than atrophy of single regions in the differential diagnosis of FTD compared with AD: knife-edge, severe frontotemporal atrophy combined with dilatation of frontal horn, and an anterior greater than posterior gradient is suggestive of a diagnosis of FTD (class II, level A).
- A normal structural MRI scan should prompt the clinician to reconsider a diagnosis of bvFTD, if clinically severe, and semantic variant PPA (good practice point).
- Presence of knife-edge frontal and/or temporal lobe atrophy in patients with PPA is predictive of FTLD pathology, whilst the presence of temporoparietal atrophy is highly associated with AD (class III, level C).
Functional imaging findings
Cerebral blood flow SPECT and FDG PET scans of typical AD patients demonstrate predominant hypoperfusion or reduced glucose metabolism in the temporoparietal regions, including the precuneus and the posterior cingulate cortex . Functional frontal lobe involvement is also often reported in AD, but usually in conjunction with and characteristically less severe than temporoparietal involvement . Overall, hypoperfusion or hypometabolism in early-onset AD is much greater in magnitude and extent than that of late-onset AD patients with similar dementia severity [166-168]. Early-onset AD patients typically show more severe hypometabolism in parietal, frontal, occipital and subcortical areas [167, 168]. The primary visual and sensorimotor cortices, cerebellum, thalamus and basal ganglia are relatively spared in AD .
A few studies compared CBF SPECT and FDG PET in their ability to differentiate AD from healthy controls and other dementia, but it seems that FDG PET has both a higher sensitivity and a higher specificity than SPECT [13-16]. The best correspondence was in the temporoparietal and posterior cingulate cortices. However, tracer uptake reductions were significantly more pronounced with PET than with SPECT .
Most existing studies compared FDG PET to a clinical diagnosis. Five case–control FDG PET studies using clinical assessment as the reference standard [169-173] revealed an overall diagnostic accuracy of 93% for differentiating AD subjects from healthy subjects, with sensitivity of 96% and specificity of 90% . A prospective study of 102 individuals, presenting consecutively to a primary care centre for examination of suspected early-onset dementia, showed sensitivity of 78% and specificity of 81% of FDG PET scans against clinical diagnosis of AD . Some studies compared the accuracy of FDG PET with that of clinical and neuropathological diagnosis of dementia [165, 175-177]. A multicentre analysis in 138 patients with histopathological diagnoses reported that FDG PET correctly identified the presence or absence of AD in 88% of the cases, with a sensitivity of 94% and a specificity of 73% . A single-centre cohort study of 44 subjects with variable levels of cognitive impairment and autopsy confirmation showed that the diagnostic accuracy available with FDG PET at an initial clinical evaluation (sensitivity, 84%; specificity, 74%) was better than that of initial clinical evaluation alone (sensitivity, 76%; specificity, 58%) and was similar to that of longitudinal clinical diagnosis over approximately 4 years . The diagnosis of AD was associated with a 70% probability of detecting AD pathology, but with a positive PET scan this increased to 84%, and with a negative PET scan this decreased to 31% . A diagnosis of not-AD at an initial clinical evaluation was associated with a 35% probability of AD pathology, increasing to 70% with a positive PET scan .
FDG PET differentiates patients with MCI from healthy controls. Amnestic MCI typically shows regional hypometabolism consistent with AD, although the magnitude of reduction is milder than that in clinically probable AD cases [172, 178-183]. Longitudinal studies of patients with MCI found that if the baseline FDG PET scan suggests an AD-like pattern, the probability of clinical progression within several years is extremely high [184-187]. A meta-analysis estimated that an AD-like FDG PET pattern observed at baseline in MCI patients had a sensitivity of 89% and a specificity of 85% in distinguishing converters from stable subjects . However, several MCI patients do not have amnestic symptoms. The few CBF SPECT and FDG PET studies that considered amnestic and non-amnestic MCI patients separately provided evidence for a high variability in non-amnestic subjects [172, 179, 180, 188]. A large multicentre study examining FDG PET scans from 114 MCI subjects (amnestic and non-amnestic) found an AD-like PET pattern in 25% of subjects and a DLB- or FTD-like PET pattern in 10% of subjects . The AD pattern was found in the majority (79%) of the MCI patients with deficits in multiple cognitive domains, frequently with additional frontal hypometabolism, and in 31% of amnestic MCI patients . The remaining amnestic MCI patients showed primarily hypometabolism restricted to the hippocampus and posterior cingulate cortex . Non-amnestic MCI patients showed more variable FDG PET profiles, from no hypometabolism (9%) and isolated hippocampal deficits (18%) to widespread FDG uptake consistent with DLB (18%) or with AD and FTLD (9%) .
Atypical presentations of AD
The value of functional imaging biomarkers might also differ in the setting of atypical, focal AD presentations, in which the topographical distribution of functional abnormalities needs to be considered separately in each syndrome. Data from functional imaging studies using either CBF SPECT or FDG PET demonstrate a comparable involvement of temporoparietal cortex and precuneus in PCA and typical AD, with extension of hypoperfusion or hypometabolism into occipital and posterior temporal lobes in PCA patients [189-192]. In addition to posterior regions, FDG PET in PCA has indicated specific areas of hypometabolism in the frontal eye fields bilaterally, which can occur secondary to loss of input from occipitoparietal regions and be the cause of oculomotor apraxia in these patients [189, 191]. Only a few studies provide guidance about the degree of sensitivity and specificity of CBF SPECT or FDG PET in the diagnostic work-up of atypical AD cases. A retrospective study of 94 patients with a clinical diagnosis of MCI or dementia (typical or atypical), who had an FDG PET within 2 months of their diagnosis, showed that FDG PET findings significantly lowered the number of atypical/unclear diagnoses from 39% to 16% . For a detailed description of functional neuroimaging features in AD cases with prominent language deficits, see the section on PPA.
Dementia with Lewy bodies
Numerous studies reported predominant medial occipital cortex hypoperfusion or hypometabolism in DLB patients compared with AD, with a parietotemporal reduction common to both the diseases [114, 165, 172, 194-199]. Occipital lobe hypometabolism differentiated patients with DLB from AD in both clinically diagnosed [172, 196, 197, 199] and autopsy-confirmed [165, 194, 195] cohorts. One study comparing FDG PET findings with autopsy results found that occipital hypometabolism, particularly in the primary visual cortex, distinguished DLB from AD with 90% sensitivity and 80% specificity . Furthermore, the sensitivity in discriminating DLB and AD using FDG PET was greater than that with clinical diagnostic criteria applied retrospectively to the data from medical charts . However, on individual SPECT and FDG PET scans, the appearances of DLB and AD can be identical. Moreover, occipital hypometabolism is not a specific marker for DLB and can occasionally be associated with AD. In particular, occipital metabolism declines in advanced AD  and is associated with atypical AD with PCA [189-192]. The differential diagnosis between DLB and PCA can even be particularly difficult as in some PCA patients there is relative sparing of the hippocampus. In the light of these findings, the McKeith criteria  consider a generalized low uptake on SPECT/PET scan with reduced occipital activity as a supportive feature in the diagnostic criteria for DLB, that is, a finding that is commonly present in DLB but lacks sufficient diagnostic specificity to be categorized as core or suggestive. On FDG PET, the cingulate cortex in DLB appears to be relatively spared (‘the cingulate island sign’) compared with AD cases, but the clinical value of this finding has not been fully investigated [196, 201].
Unlike AD, DLB is characterized by nigrostriatal dopaminergic neurodegeneration, making dopaminergic imaging a potentially useful diagnostic tool in this disease . Reduction in striatal uptake of dopamine can be visualized with 18F-l-dihydroxyphe-nylalanine (18F-DOPA) or by imaging the dopamine transporter using 123I-2beta-carbometoxy-3beta-(4-iodo-phenyl)-N-(3-fluoropropyl) nortropane (123I-FP-CIT) SPECT (known commercially as the DaTSCAN™; GE Healthcare, Waukesha, WI, USA). In a multicentre Phase III trial of 123I-FP-CIT SPECT in 326 patients with a clinical consensus diagnosis of probable (n = 94) or possible (n = 57) DLB or non-DLB dementia (n = 147) , mean sensitivity of 123I-FP-CIT SPECT imaging for a clinical diagnosis of probable DLB was 78%, whilst the mean specificity for excluding non-DLB dementia (which was predominantly due to AD) was 90%, giving overall diagnostic accuracy of 86%. Follow-up clinical diagnosis at 12 months, when diagnosis had become clearer in nearly 60% of patients, confirmed the ability of 123I-FP-CIT SPECT imaging to discriminate DLB from non-DLB dementia . Of 44 patients with a clinical diagnosis of possible DLB at baseline, the diagnosis at follow-up remained as possible DLB in 18 but was changed to probable DLB in 19, 12 of whom had abnormal SPECT scans at baseline, and to non-DLB dementia in seven, all of whom had normal baseline scans . 123I-FP-CIT SPECT has demonstrated higher sensitivity and specificity for differentiating DLB from non-DLB than clinical diagnosis in a series of 20 patients who had post-mortem brain examination . In this study, sensitivity of an initial clinical diagnosis of DLB against autopsy diagnosis was 75% and specificity was 42%, in comparison with 88% sensitivity and 100% specificity with 123I-FP-CIT SPECT imaging . These results suggest that an abnormal dopaminergic imaging scan in individuals with possible DLB strongly supports the diagnosis. As a consequence, low dopamine transporter uptake in the basal ganglia demonstrated by SPECT or PET imaging has been included as a suggestive feature in the diagnostic criteria for DLB (one suggestive feature plus one core feature being sufficient to allow a diagnosis of probable DLB) . On the contrary, its negativity does not exclude a clinical diagnosis of probable DLB, as about 20% of probable DLB cases will have a normal or inconclusive scan .
Dopaminergic imaging cannot distinguish DLB from alternative nigrostriatal disorders, such as PD with dementia [206-210], MSA, PSP, corticobasal degeneration, vascular parkinsonism with dementia or FTD with parkinsonism, as all are associated with presynaptic dopaminergic deficiency .
Behavioural variant of frontotemporal dementia is identified on SPECT or PET scans by patterns of hypo-perfusion or hypometabolism in frontal, insular and anterior temporal regions that are typically quite asymmetrically centred into the frontolateral cortex [212-214]. The regions mostly impaired are the medial frontal cortex, followed by the frontolateral and anterior temporal cortices. A prospective study in 134 patients with suspected bvFTD demonstrated that a predominant frontal, anterior temporal or frontotemporal hypoperfusion or hypometabolism on initial SPECT or PET scans has a sensitivity of 90% and specificity of 75% against a clinical diagnosis after 2 years . However, metabolic abnormalities are not limited to these regions. As the severity of dementia increases, the severity and topographical extent of perfusion and metabolic impairments also increase and begin to involve other association areas .
The regional pattern of predominantly frontal functional impairment in FTD, with relative sparing of posterior brain regions, usually allows a clear distinction between these patients and those with AD [140, 213, 216-219]. Using an anterior-to-posterior CBF SPECT ratio (medial superior frontal gyrus/medial temporal lobes), patients with clinical bvFTD were successfully distinguished from AD patients, with a sensitivity of 87% and a specificity of 96% versus early-onset AD patients and 80% versus late-onset AD patients . A few studies have looked at the accuracy of SPECT or FDG PET findings in relation to the pathological diagnoses [177, 220]. A reduction in frontal CBF was more common in pathologically confirmed FTLD than in AD cases and was of diagnostic value (sensitivity 80%, specificity 65%) . When the pattern of bilateral frontal CBF reduction was not associated with a bilateral parietal CBF abnormality, the diagnosis was more accurate (sensitivity 80%, specificity 81%) . However, an overlap of abnormalities between the two conditions can be seen, as AD can involve frontal regions and FTLD may not spare temporoparietal cortex . In a study of autopsy-proven FTLD and AD patients, the addition of FDG PET scans particularly improved the accuracy of an FTLD diagnosis more than an AD diagnosis . However, FDG PET scans from FTLD patients had more variability of interpretation compared with those from AD cases, resulting in lower sensitivity (70% for FTLD vs. 98% for AD) and confidence . The disagreement in interpretation of scans in patients with FTLD largely occurred when there was a temporoparietal hypometabolism . Hypometabolism of anterior cingulate and anterior temporal regions indicates a high likelihood of FTLD, even when temporoparietal hypometabolism is present .
Primary progressive aphasia
In patients with semantic variant PPA, FDG PET studies showed asymmetrical hypometabolism of the temporal lobes, more marked on the left side [100, 222-224]. A functional deficit of the left frontal opercular regions of the brain has been reported in non-fluent variant PPA patients [100, 225-228]. In these cases, a functional involvement of bilateral caudate nuclei and thalami was also described . Logopenic PPA patients usually show a pattern of left posterior temporoparietal hypometabolism on FDG PET scans .
A study comparing semantic PPA and very early AD patients using structural MRI and FDG PET findings revealed hippocampal atrophy and hypometabolism in both groups, but a strikingly reduced metabolism in the posterior cingulate cortex in patients with AD that was not present in those with semantic variant PPA . A functional imaging study of non-fluent patients demonstrated that a pattern of bilateral temporoparietal involvement is predictive of AD pathology, whilst a unilateral (left), reduced temporoparietal cortex function can be seen in cases with FTLD pathology . On the contrary, a bilaterally normal temporoparietal cortical perfusion or metabolism was predictive of FTLD pathology .
Recommendations for functional imaging
- Although typical cases of dementia may not benefit from routine SPECT or PET imaging, these tools are recommended in those cases where diagnosis remains in doubt after clinical and structural MRI work-up and in particular clinical settings (class II, level A).
- Functional imaging can be of value to diagnose (or exclude) a neurodegenerative dementia in those subjects with cognitive impairment presenting with severe psychiatric disturbances (including depression and agitation) and in cases where proper cognitive testing is difficult, that is, with no language in common with the clinician (good practice point).
- Normal FDG PET scan findings, in the presence of the suspicion of dementia, make a neurodegenerative diagnosis less likely (class II, level A).
- The overall regional pattern of metabolic impairment of the posterior cingulate/precuneus and lateral temporoparietal cortices, more accentuated than frontal cortex deficits, together with the relative preservation of the primary sensorimotor and visual cortices, basal ganglia and cerebellum defines the distinct metabolic phenotype of AD (class II, level A).
- AD-like metabolic patterns in patients with MCI are predictive of conversion to AD within several years (class II, level A).
- Occipital hypometabolism, particularly in the primary visual cortex, may be more common in DLB than AD on a group basis (class II, level B). However, on individual scans, the appearances of DLB and AD can be identical. Moreover, occipital hypometabolism is not a specific marker for DLB and can be associated with AD (good practice point).
- Although an overlap of functional abnormalities between FTD and AD has been shown to occur, the presence of posterior temporal and parietal brain hypoperfusion or hypometabolism is predictive of a pathological diagnosis of AD, whereas a disproportionate reduction in frontal perfusion/metabolism is more common in FTD cases (class II, level A).
- In PPA patients, bilateral posterior temporoparietal hypometabolism (PET) or hypoperfusion (SPECT) is predictive of AD pathology; normal bilateral posterior temporoparietal function is specific for FTLD (class III, level C).
- Dopaminergic SPECT is useful to distinguish DLB from AD (class I, level A), especially when there are no clear extrapyramidal symptoms and signs. However, a negative 123I-FP-CIT scan does not necessarily exclude a diagnosis of probable DLB, as around 20% of individuals with probable DLB appear to have normal scans (class I, level A).
- Dopaminergic SPECT can be useful in differentiating DLB from long-term psychiatric patients on neuroleptic drugs, whose parkinsonism may be drug-induced (good practice point).