Quantitative structural MRI for early detection of Alzheimer’s disease

Abstract

Alzheimer’s disease (AD) is a common progressive neurodegenerative disorder that is not currently diagnosed until a patient reaches the stage of dementia. There is a pressing need to identify AD at an earlier stage, so that treatment, when available, can begin early. Quantitative structural MRI is sensitive to the neurodegeneration that occurs in mild and preclinical AD, and is predictive of decline to dementia in individuals with mild cognitive impairment. Objective evidence of ongoing brain atrophy will be critical for risk/benefit decisions once potentially aggressive, disease-modifying treatments become available. Recent advances have paved the way for the use of quantitative structural MRI in clinical practice, and initial clinical use has been promising. However, further experience with these measures in the relatively unselected patient populations seen in clinical practice is needed to complete translation of the recent enormous advances in scientific knowledge of AD into the clinical realm.

Alzheimer’s disease (AD) is a progressive, ultimately fatal, neurodegenerative disorder. It is the most common cause of dementia, accounting for 60–80% of cases [1]. Age is the strongest risk factor for late-onset, sporadic AD, with age-specific prevalence doubling every 5 years after the age of 65 years [2]. With the increase in average life expectancy, the prevalence of AD is rising at an alarming rate. In 2005, an estimated 24 million people around the world suffered from dementia, and that number is expected to double every 20 years. By 2040, it is predicted that over 81 million people worldwide will suffer from dementia [3].

Alzheimer’s disease inflicts a terrible toll on patients, their families and society in general. If present trends continue, the cost of caring for the expected increase in the number of AD patients will bankrupt public healthcare systems [1]. Current treatments for AD offer temporary symptomatic relief, but do not affect the underlying disease process. A huge effort is underway to develop disease-modifying treatments to prevent or slow disease progression. Numerous potential treatments, aimed at different underlying pathogenic mechanisms of AD, are under development [4]. There is hope that one or more of these approaches will be effective at altering the course of AD in the coming years [4].

Most experts agree that treatment will be most beneficial if applied early, before significant, potentially irreversible neurodegeneration and functional impairment has occurred [5,6]. Several large studies around the world have been initiated to identify biomarkers to aid in earlier detection of AD [7–10]. In the USA, the Alzheimer’s Disease Neuroimaging Initiative (ADNI), a ground-breaking publicly and privately funded, longitudinal, multisite study was launched in 2003 to examine the sensitivity of multiple biomarkers for the early detection and tracking of AD progression [11]. Data collected as part of this study has been made publicly available to the research community, resulting in a large and rapidly growing body of research on AD biomarkers. Results of many of these studies have recently been summarized in a series of articles published in a special issue of Alzheimer’s & Dementia [12–15].

The most promising biomarkers for early AD detection include cerebrospinal fluid (CSF) bio-markers, PET imaging of metabolism or cortical β-amyloid binding and structural MRI of brain atrophy. Of these, MRI is completely noninvasive and widely available. It is often used to rule out other underlying causes of impairment, and recent advances in the field have paved the way for transitioning research findings into routine clinical practice. This article will focus on recent studies that demonstrate the sensitivity of baseline and longitudinal quantitative structural MRI for the early detection of AD, and will describe the use and advantages of quantitative structural MRI in relation to emerging literature on other potential biomarkers. It will briefly discuss barriers to the adoption of quantitative MRI in clinical practice, the progress being made to surmount these barriers and results of initial clinical experience. This article is necessarily selective, as the number of articles examining MRI in AD has increased enormously in recent years. A literature search revealed a total of 586 publications with key words of “AD” and “MRI” in 1990–1999, 747 in 2000–2004, and 1515 since 2005, the period of focus for this article.

Alzheimer’s disease pathology

Although the etiology of AD is not clearly understood, the cascade of molecular changes associated with AD is becoming increasingly well delineated, opening up new avenues for potential intervention [16,17]. AD is characterized histopathologically by the presence of extracellular β-amyloid plaques and intraneuronal tangles of hyperphosphorylated tau proteins. The β-amyloid plaques first appear in basal neocortex then spread throughout the cortex [18], although neither the density nor distribution of plaques correspond to disease stage. Significant β-amyloid burden has been observed in cognitively normal individuals [19,20], but has been found to correlate with subtle cognitive dysfunction, suggesting that it may represent an early, preclinical stage of AD [20]. The role of β-amyloid in the etiology of AD is currently under sharp debate. Some suggest that it is an important contributing factor, if not the primary trigger, of the cascade of changes that lead to cell death in AD, while others contend that it is a downstream event that may even have neuroprotective benefit [21–24]. Regardless of its role, cortical β-amyloid plaque disposition is one of the defining histopathological features of AD.

The other defining histopathological feature of AD is the presence of neurofibrillary tangles (NFTs). Pathological phosphorylation of tau proteins leads to a sequence of events that results in the formation of NFTs and eventual death of the affected neuron. The tau pathology in AD correlates better with disease stage than β-amyloid burden does, and appears necessary for the clinical expression of the disease. NFTs are first observed in the transentorhinal area, then spread throughout the limbic area before appearing in widespread areas of association cortex, with sparing of primary sensory and motor areas until later disease stages [25,26]. Based on clinical diagnosis at the time of autopsy, Braak and Braak characterized the earliest, transentorhinal stage of the disease as a clinically silent period; the limbic stage, characterized by severe entorhinal and modest hippocampal damage, as clinically incipient AD; and the isocortical stage, when NFTs appear throughout association cortex, as fully developed dementia [25].

Alzheimer’s disease diagnosis

Currently, AD is clinically diagnosed as probable or possible AD based on the presence of progressive cognitive impairment of sufficient severity to interfere with activities of daily living, and by the absence of other neurological conditions that could account for the observed impairment [27]. A definitive diagnosis is available only through brain biopsy or postmortem examination, based on histopathological verification of the presence of β-amyloid plaques and NFTs. Several problems have been identified with the current criteria. One is that, given the insidious onset of AD, judgment of when functional impairment is of sufficient severity to warrant the diagnosis of dementia is subjective and somewhat arbitrary [28]. Second, reliance on the presence of dementia means that AD cannot be diagnosed until an individual has reached a relatively severe stage of the disorder, in which brain damage is widespread and likely irreversible. A third problem is the required absence of other explanatory neurological findings rather than the presence of characteristic biological features, or biomarkers, of AD [29]. Rapidly accumulating evidence indicates that AD is associated with specific patterns of cognitive impairment [30,31], abnormal levels of CSF biomarkers [32], increased cortical β-amyloid binding [33] and structural [34,35] and metabolic changes in the brain [36,37]. One or more of these biomarkers could be used to aid the earlier detection of AD. Working groups, supported by the National Institute on Aging and the Alzheimer’s Association, have been tasked with revising the diagnostic criteria for AD to reflect the strong advances in the field that have occurred since the current diagnostic criteria were introduced in 1984 [201], and their recommendations are expected to be published soon.

Mild cognitive impairment

In the quest to aid detection of AD at an earlier stage, mild cognitive impairment (MCI) has been proposed as a transitional state between normal aging and AD [38]. MCI is characterized by objective impairment in one or more cognitive domains, but is of insufficient severity to interfere with activities of daily living. When memory is one of the domains affected, MCI is associated with an increased risk of developing AD, with an annual conversion rate of approximately 10–20% per year, compared with 1–2% for the general population [38–41]. However, multiple etiologies can impair memory function. A sizeable minority of individuals with MCI develop other forms of dementia [42–44], whereas others remain cognitively stable or revert to normal cognitive function [38,45]. Thus, the MCI diagnosis by itself is not sufficient to identify individuals with a progressive neurodegenerative disorder, let alone to identify individuals whose neurodegeneration is due to AD. The addition of supporting biomarker evidence may help identify the subset of amnestic MCI individuals who are suffering from prodromal AD [5,28], and who thus represent the appropriate target for clinical trials [46] and, eventually, for early, aggressive treatment.

Quantitative structural MRI for detection of neurodegeneration in AD

Baseline atrophy in AD

The clinical symptoms of AD arise from progressive neuron and synapse loss, with the resulting tissue atrophy visible on high-resolution structural MRIs. As expected from the pathology and clinical expression of AD, significant atrophy is observed in early disease stages in the memory-related structures of the medial temporal lobe, particularly the hippocampus and entorhinal cortex [47,48], with the degree of atrophy correlating with memory impairment [48–50].

The initial use of MRI for the detection of atrophy in AD used manual tracing methods, necessitating focus on a few selected regions, such as the hippocampus or entorhinal cortex, or on global measures of atrophy, such as reduced global brain volumes and increased ventricular volumes. The advent of computer-based methods for quantitative MRI has allowed for efficient quantification of AD-related atrophy across the brain [51–55]. This revealed that even mild AD is associated with widespread atrophy of cortical association areas, with significant involvement of medial and lateral temporal, inferior parietal, posterior cingulate and prefrontal cortices (Figure 1) [47,56,57]. The degree of clinical impairment is associated with the severity of regional atrophy observed [57,58].

Figure 1. Regional atrophy patterns in mild Alzheimer’s disease and mild cognitive impairment.

(A) The average difference in thickness (mm) for 84 patients with mild AD relative to 139 controls are shown on medial (left) and lateral (right) views of the pial surface of the left hemisphere. Greatest thinning is observed in medial and lateral temporal areas, but significant thinning also appears across association cortices. (B) Average thickness difference for MCI subjects who were classified as ‘AD’ by a discriminant and control data. Although less severe, the atrophy pattern is very similar to that of AD subjects. (C) MCI subjects classified as ‘normal control’ by the discriminant model. These subjects show significantly less atrophy, particularly in medial temporal regions, subjects classified as ’AD’. For all brain images, the scale reflects thickness differences ranging from −0.3 (yellow)

(D) Average Mini-Mental State Examination score over time for MCI subjects classified as ‘AD’ and those classified with phenotypic AD atrophy showed significant, steady cognitive decline over a 2-year period, whereas those without the phenotypic AD atrophy pattern remained cognitively stable.

Brain images are reprinted with permission from [60].

AD: Alzheimer’s disease; BL: Baseline; MCI: Mild cognitive impairment; NC: Normal control.

Structural MRI measures can discriminate AD from healthy control data with high sensitivity and specificity, with the accuracy of discrimination improved when measures of atrophy beyond the medial temporal lobe are included in the analysis [59–61]. Importantly, it has been demonstrated that the degree of atrophy correlates well with disease stage determined from histopathology [62], and that the topographical pattern of atrophy is relatively consistent with the distribution of NFTs in later disease stages [63]. Regional atrophy patterns have also been found to differ between AD and other dementing disorders, such as frontotemporal dementia and dementia with Lewy bodies [64–67]. Thus, quantitative measures of atrophy from structural MRIs are sensitive to the neurodegeneration that occurs in AD, and although atrophy itself is nonspecific to AD, the topographical pattern of atrophy may be a sensitive and specific surrogate marker of AD pathology.

Baseline atrophy in preclinical AD

Consistent with the finding that histopathological changes occur prior to the clinical diagnosis of AD, significant brain atrophy is visible prior to the onset of dementia. Amnestic MCI, with its hallmark memory impairment, is associated with entorhinal and hippocampal atrophy, with thickness or volume measures intermediate between those of healthy controls and patients with mild AD [47,48,68,69]. However, even at this pre-dementia stage of the disorder, atrophy is not restricted to medial temporal areas, but extends to widespread areas of association cortex [47,57,68–70]. MCI individuals with isolated memory impairments, which may represent an even earlier stage of the disorder, show significant atrophy in areas beyond the medial temporal lobe, including thinning of lateral temporal, posterior cingulate, inferior parietal, precuneus and caudal middle frontal cortex [47].

Elderly individuals who do not meet objective criteria for memory impairment, but who report subjective memory complaints, are also at increased risk for developing dementia [71,72]. Hippocampal volumes and thickness of frontal cortex in these individuals are intermediate between those of elderly individuals without cognitive complaints and those with MCI [73]. These findings suggest that structural MRIs are sensitive not only to the neurodegeneration that occurs in MCI, but also to the neurodegeneration that occurs prior to the ability to reliably document cognitive changes with standard neuropsychological tests [73].

Prediction of conversion to AD

The atrophy in medial temporal lobe structures correlates with clinical decline and conversion to AD in individuals with MCI [74–76]. Nevertheless, retrospective comparisons between MCI patients who remained cognitively stable and those who progressed to a clinical diagnosis of AD, as well as prospective studies of MCI patients, have highlighted the importance of atrophy beyond medial temporal areas in predicting conversion to AD [50,59,60,68,70,77–79].

Recently, several groups have used multivariate analysis techniques to reduce the pattern of regional atrophy that best discriminates AD from healthy control data to a single numeric index [59–61,80]. These studies have demonstrated that the degree to which individuals with amnestic MCI express the characteristic AD atrophy pattern is predictive of a decline in cognitive function, progressive structural brain loss and conversion to AD [50,59,60,79]. Figure 1 shows an example of this approach. This study used semi-automated, high-throughput methods to derive individually specific measures of regional brain thickness and volumes from 139 healthy controls, 175 individuals with MCI and 84 with mild AD from the ADNI. Discriminant analyses applied to healthy control and AD data revealed that atrophy in medial temporal (hippocampus and entorhinal cortex), lateral temporal, orbitofrontal and isthmus cingulate cortex best differentiated AD from control data. Application of this model to MCI data demonstrated that the degree to which an MCI individual expressed the regional AD atrophy pattern was predictive of clinical decline and progressive structural brain loss [46,60]. MCI subjects whose data most resembled the AD atrophy pattern also experienced a significantly higher rate of conversion to AD (Figure 1) [60].

Recently, it was shown that this index can be converted into an estimate of an individual patient’s risk of imminent conversion to AD [81]. The 1-year risk of conversion to AD among 317 MCI patients from ADNI ranged from 3 to 41%. Individuals with the lowest scores showed a similar level of risk as that of healthy controls (1–2%/year), whereas those with the highest scores had more than double the risk of conversion (41%) associated with the diagnosis of MCI (which typically ranges from 10 to 20% per year [38–41]). These results demonstrate that information derived from quantitative structural MRI can provide useful information, not just for discriminating groups of subjects, but also for predicting individual patient prognosis [81].

Longitudinal MRI measures

Information to aid in the early clinical diagnosis of AD should ideally be derived from a single time point baseline MRI. However, due to the progressive nature of the disease and to interindividual variability in baseline brain structural measures, measures of brain change over time provide more sensitive detection of AD-related neurodegeneration and improved prediction of the conversion to AD in MCI [81–83]. Longitudinal measures are also likely to provide a valuable index for tracking disease progression and monitoring efficacy of disease-modifying therapies.

Studies comparing rates of whole-brain atrophy or ventricular expansion from serial MRIs have demonstrated that AD and MCI are characterized by significantly higher atrophy rates than that observed in healthy controls [84]. Despite the known regional specificity of AD pathology, several studies have found that these nonspecific measures of brain shrinkage over time were more sensitive to AD progression than atrophy rates in regions most strongly affected in AD [82,85,86]. However, recently developed semi-automated methods for quantifying regional change across the brain in individual subjects have indicated that atrophy rates in MCI and mild AD are highest in medial temporal areas, particularly the entorhinal cortex [58,87].

The ability to monitor region-specific atrophy rates is important since the rate of neurodegeneration in different brain regions is likely to impact different cognitive abilities [88] and to reflect different stages of disease severity [58]. For example, a study of 472 participants from ADNI found that atrophy rates varied as a function of brain region and disease severity, as characterized by Clinical Dementia Rating Scale Sum of Boxes score (CDR-SB) (Figure 2) [58]. At the mildest stages of functional impairment (CDR-SB scores of 0.5–1.0), significantly elevated atrophy rates, relative to the unimpaired elderly, were observed in medial and inferolateral temporal, inferior parietal and posterior cingulate regions. With increasing functional impairment (CDR-SB 1.5–2.5), elevated atrophy rates were observed in superior temporal, superior parietal, lateral occipital and widespread frontal regions. At the mild AD stage (CDR-SB > 2.5), elevated atrophy rates were observed across the cortex, with the exception of primary visual and auditory areas. The finding of higher atrophy rates with greater functional impairment is consistent with other studies that have demonstrated accelerated atrophy rates with disease progression [89,90]. However, the increase in atrophy rate with disease severity was not always linear. Atrophy rates increased monotonically with disease severity in hippocampal and lateral temporal areas. Greater acceleration in atrophy rates with disease severity was observed in prefrontal, parietal and anterior cingulate regions, whereas entorhinal cortex showed less acceleration with increased disease severity. These results are important for several reasons. First, they show that atrophy rates continue to accelerate with disease progression. No evidence was found in this, or other studies [89,90], that atrophy rates stabilize or decrease with increasing disease severity, suggesting that information on the rate of change will be a useful biomarker for tracking disease progression across the stages of the disease, during which good-quality MRI images can be obtained. They further suggest that if atrophy rates are used to monitor disease progression (or slowing of progression with treatment), different regions should be monitored at different disease stages [58,91], and that, when using measures of brain atrophy to differentiate AD from other dementing disorders [92–94], it may be crucial to take disease severity into account.

Figure 2. Annual atrophy rates as a function of degree of clinical impairment as assessed with baseline CDR-SB.

Mean atrophy rates are represented as a percent change in neocortical volume and mapped onto the lateral (left), ventral (middle) and medial (right) pial surface of the left hemisphere. These data demonstrate that atrophy rates are most prominent in medial and lateral temporal regions early in the course of disease, spreading to parietal and frontal regions as the level of impairment increases, with relative sparing of sensorimotor regions. The scale reflects annual percent change ranging from 0.5 (dark blue) to 3.0%. Note that the scale is optimized for showing disease-related change, rather than change in normal aging.

CDR-SB: Clinical Dementia Rating Sum of Boxes score.

Reprinted with permission from [58].

It is important to emphasize that although atrophy rates in MCI and mild AD significantly exceed rates observed in healthy older controls, subtle, but significant, changes in brain structure over time are detectable in healthy individuals [90,95], even over intervals as short as 6 months [84,96]. Using individually-specific regional measures of longitudinal change, atrophy rates in cognitively intact elders were found to vary across brain regions and to accelerate with increased age [95]. This acceleration was particularly apparent in regions most vulnerable to AD, and may be an early sign of impending cognitive impairment [95]. Consistent with this, a recent study demonstrated that 6-month atrophy rates in medial and lateral temporal areas in healthy elderly were predictive of 2-year memory decline [96]. Similarly, whole-brain and hippocampal atrophy rates increased 3 and 5 years, respectively, before the onset of AD in a small sample of pre-symptomatic autosomal dominant mutation carriers for AD [89]. The ability to differentiate mutation carriers from controls occurred 2 years earlier for atrophy rates than for cross-sectional comparisons of brain volumes [89], highlighting the sensitivity of longitudinal MRI measures for early detection of AD-related neurodegeneration.

In summary, cross-sectional and longitudinal studies convincingly demonstrate that quantitative MRI measures are sensitive to the neurodegeneration that occurs in prodromal and established AD, and suggest that structural loss can be detected prior to cognitive decline, providing firm support for the notion that structural MRI can be used to detect AD prior to the onset of dementia.

Structural MRI in relation to other biomarkers

Of the potential biomarkers currently under investigation for use in early detection and diagnosis of AD, structural MRI seems best suited for routine use as an adjunct to clinical diagnosis. It provides complementary information to that provided by the other biomarkers, and in isolation or in combination with other biomarkers, can provide valuable information to aid in AD diagnosis. The other promising bio-markers for early AD detection include PET measurements of glucose metabolism or cortical β-amyloid deposition, and CSF levels of β-amyloid and tau proteins.

FDG-PET

Neuron loss in AD, coupled with synaptic loss and dysfunction, results in reduced neuronal energy demand, evident as regions of hypometabolism on fluorodeoxyglucose (FDG)-PET images. AD is characterized by a distinct pattern of hypometabolism in posterior cingulate, precuneus and parietotemporal areas that correlates with clinical symptoms and predicts conversion to AD [36,37,97]. FDG-PET is currently used clinically to distinguish AD from frontotemporal dementia, and simple visual inspection of the images can reveal hypometabolism patterns characteristic of each disorder. As with MRI, FDG-PET appears to be sensitive to neuronal dysfunction prior to symptom onset. Significant hypometabolism in AD-related brain areas is observed in cognitively normal individuals at genetic risk for AD [98]. However, a recent study in which glucose metabolism in ADNI individuals was quantified in subject-specific regions of interest derived from coregistering an individual subject’s FDG-PET and MRI data, demonstrated that the largest structural effect sizes exceeded the largest metabolic effect sizes, not only in patients with mild AD, but also in those with MCI, including the subset of MCI patients with isolated memory impairment [99]. A recent longitudinal study of a small group of amnestic MCI subjects reported that hippocampal atrophy is an upstream event from posterior cingulate and medial orbital frontal hypometabolism [100]. These results contradict recent theoretical models of temporal ordering of relative biomarkers, which postulate that changes in FDG-PET metabolic measures precede changes in structural MRI measures and that FDG-PET measures begin to stabilize when measures of structural change begin to accelerate [35,91]. These findings imply that for early detection of AD in routine clinical practice, MRI may be just as sensitive, if not more so, than FDG-PET [99] and offers advantages of being free of ionizing radiation, less expensive, more widely available and more readily accepted for reimbursement by insurance companies.

CSF biomarkers & amyloid imaging

Elevated levels of CSF tau and phosphorylated tau proteins, as well as decreased levels of amyloid-β₄₂, are reliably associated with AD [44,101–103] and predictive of clinical decline in MCI [44,102]. Elevated tau levels are believed to be a nonspecific reflection of neuronal degeneration, whereas phosphorylated tau is thought to reflect NFT formation [104]. Decreased levels of amyloid-β₄₂, which are inversely correlated with postmortem measures of β-amyloid accumulation [105,106], are thought to reflect sequestering of β-amyloid in amyloid plaques in the brain, and thus serve as an indirect biomarker for brain amyloid pathology.

Brain amyloid pathology can also be directly measured in vivo through the use of PET tracers that bind to amyloid deposits in the brain [33,107]. PET imaging measures of β-amyloid binding are highly directly correlated with postmortem β-amyloid pathology [107], and highly inversely correlated with CSF amyloid-β₄₂ levels [108]. Like CSF amyloid-β₄₂ measures, PET measures of β-amyloid binding are very sensitive to AD and predictive of progression to AD in individuals with MCI and in healthy controls [109].

Using either CSF or PET imaging methods, estimates of β-amyloid pathology, indistinguishable from those of AD patients, have been observed in approximately 30% of healthy individuals, consistent with the frequency of findings of AD neuropathology in nondemented individuals at autopsy [106,110,111]. Current theories suggest that levels of β-amyloid pathology rise slowly, many years before clinical symptoms of AD emerge, and that by the time cognitive impairment and neurodegeneration are detected, β-amyloid levels have stabilized [35,91,112]. However, there is evidence to suggest that when individually-specific measures of brain structural changes are used, regional atrophy can be detected in cognitively normal individuals with abnormal levels of amyloid biomarkers [57,113].

Amyloid biomarkers are very sensitive in discriminating AD and MCI patients from healthy controls, and at discriminating individuals with MCI who will progress to AD from those who remain stable [110]. However, the specificity of amyloid pathology is uncertain, as it is not accompanied by dementia in approximately 30% of cases and may not inevitably lead to dementia in these individuals. Current research thus suggests that negative amyloid findings provide strong evidence for the absence of AD, whereas positive findings, in the absence of other biomarker evidence, do not necessarily imply that an individual will develop dementia.

Studies that have directly compared the predictive ability of amyloid biomarkers with structural measures have found that structural measures are better at predicting risk of decline in MCI over the near term, but that amyloid biomarkers provide important complementary information [79,112,114,115]. Taken together, these findings suggest that positive evidence of amyloid pathology in the presence of phenotypic AD brain atrophy would strongly suggest that an individual is in a prodromal stage of AD [112].

Translation to clinical practice

MRI is commonly used in current clinical practice for the evaluation of cognitive impairment. However, the translation of research results on the sensitivity of structural MRI to prodromal AD into clinical practice has been slow. As recently reviewed in detail [116], several barriers have prevented routine adoption of structural MRIs in the clinical assessment of suspected AD. These include a lack of standardized image acquisition parameters, spatial distortions and motion artifact in MRI data, lack of automated methods for individual-specific quantification of affected brain regions, lack of normative data and difficulties integrating quantitative MRI measures into the clinical workflow. Many of these barriers are being overcome with the aid of large-scale, multisite clinical trials, such as ADNI, and other large-scale efforts aimed at improving automated MRI processing, such as the Morphometry Biomedical Informatics Research Network (mBIRN) [117,118].

A critical first step in ADNI was the development of standardized image acquisition protocols that would be robust to intersite variation and would maximize the contrast between gray and white matter in the brain – a vital consideration for the use of automated image analysis algorithms [119]. Clinical MRI scans, currently used to rule out other causes of impairment in suspected AD cases, are suboptimal for use in quantitative structural analysis, due to their relative lack of sensitivity to gray–white contrast, and to their susceptibility to artifacts, such as spatial distortions related to gradient field nonlinearities, field inhomogeneities, and intensity nonuniformities due to B₁ motion-induced artifacts caused by patient movement. These artifacts do not pose a significant problem for visual detection of gross brain abnormalities, but can cause large errors in automated quantification algorithms.

Automated methods for spatial distortion correction and image intensity normalization that can be applied to post-acquisition data have been developed [118,120]. Techniques that can correct for motion artifact during image acquisition, with minor impact on acquisition time, are under development [121] and have shown very promising results in motion-prone populations [122]. High-throughput, automated methods have been developed for efficient segmentation and quantification of various brain structures in individual patients [52–54,123–126]. These methods, when applied to data acquired with ADNI-compatible protocols and that have undergone corrections for spatial distortion and intensity inhomogeneities, provide sensitive, reliable measures of global and regional volumes [50,126]. However, they do require qualitative visual review by a trained technician or imaging expert to detect cases where significant image artifacts or gross brain abnormalities result in segmentation errors.

Initial clinical experience

As was recently described [116,126], US FDA-approved image analysis software (NeuroQuant; CorTechs, Inc., La Jolla, CA, USA) currently exists that provides fully automated volumes of several brain structures, including cerebral white and gray matter, hippocampus, amygdala, basal ganglia nuclei and ventricular volumes from ADNI-compatible, 3D T1-weighted sagittal MRI scans. These measures have been validated against manual segmentation and shown to be sensitive to neurodegeneration associated with AD [50,126]. NeuroQuant software compares an individual patient’s regional brain volumes with those of a normative database, correcting for sex, head size and age. Visual quality review of the results by an imaging expert remains important, and the imaging expert, as well as the referring physician, have the ability to scroll through the segmented MRI images, in all three dimensions, to determine accuracy of the segmentation. NeuroQuant is currently in use in a number of clinics, including the University of California, San Diego (UCSD; CA, USA). During the past 2 years of clinical use of volumetry, Medicare reimbursement for the additional post-processing has been consistent. As advised by the local office for Center for Medicare and Medicaid Services, records are kept to document that the referring physicians have specifically requested quantitative segmental volume reporting and assessment (CPT code 76377) for the patient.

An example of a clinical report generated by this method is shown in Figure 3. The report provides axial, coronal and sagittal views of a patient’s MRI, with color-labeled structures to provide information on image and segmentation quality. Raw volumes and volumes expressed as a percentage of intracranial volume are provided, along with the age-specific normative range. Referring physicians have reported that the information contained in the reports have provided valuable complementary information to the history, neurological and neuropsychological findings of their patients, often providing a biological foundation for their clinical impressions [116].

Figure 3. NeuroQuant report from a mild cognitive impairment patient scanned longitudinally for 2 years as a volunteer in the Alzheimer’s Disease Neuroimaging Initiative study.

Ten scans (two from each timepoint) were analyzed using NeuroQuant. From top to bottom, the report consists of demographic information; a row of axial, coronal and sagittal MR images with a color overlay of the segmentation results; a table reporting the volume, the volume as a percent of intracranial volume and the volume’s percentile score based on the normative database for bilateral hippocampus, lateral ventricle and temporal horn of the lateral ventricle; and normative graphs plotting the patient’s hippocampus volume and temporal horn volume, corrected for intracranial volume, relative to the expected range (in white) across the age range included in the normative database. For this patient, the baseline hippocampal volume of 5.09cc –corrected for age, sex and intracranial volume – was more than two standard deviations below that expected for their age, and continued to atrophy at a rate faster than would be expected in normal aging. The baseline temporal horn volume of 4.25cc was at the 77th percentile of that expected for age, and continued to enlarge at a rate faster than would be expected in normal aging. The patient progressed to Alzheimer’s disease at year 2.

Quantitative structural MRI in AD clinical trials

In addition to its potential in clinical practice, quantitative structural MRI also has an important role to play in AD clinical trials. As an enrichment strategy, structural MRI offers the ability to selectively enroll individuals with the highest likelihood of experiencing significant decline over a 1–2 year period. Constraining enrollment to amnestic MCI individuals with a regional atrophy pattern characteristic of AD on a baseline screening MRI, as shown in Figure 1, would enable reductions in sample sizes by 40–60%, considerably reducing the expense of a trial [46]. As an outcome variable, structural MRI measures would also enable further reductions in sample size due to their lower inter-individual variability and higher sensitivity to change in early disease stages than standard outcome measures, such as the Alzheimer’s Disease Assessment Scale – Cognitive subscale (ADAS-Cog) [46,87,127–129]. Importantly, unlike standard clinical and functional outcome measures, which cannot distinguish symptomatic benefits from disease-altering effects, structural MRI measures can provide evidence of slowing of the progressive neurodegeneration associated with AD.

Structural MRI outcome measures are now being included in a number of clinical trials. Although no trials have yet demonstrated disease-modifying effects in planned primary analyses, secondary analyses of two amyloid agents have shown promising effects. In the Phase III trial of tramiprosate, an amyloid-β antagonist, secondary analyses revealed that tramiprosate treatment was associated with significantly less hippocampal volume loss and less decline on the ADAS-Cog (although effects on the latter were marginal) relative to placebo [130]. Similarly, secondary analyses in the Phase II trial of bapineuzumab, a passive amyloid-β immunotherapy, showed that with treatment, MCI individuals without the APOE ε4 genetic risk factor experienced a reduced rate of whole-brain atrophy and reduced decline on functional and cognitive measures, relative to the placebo group. Individuals with the genetic risk factor did not show any beneficial effect of therapy [131].

However, the interrupted trial of the active amyloid-β immunotherapy, AN1792, produced unexpected MRI results. Antibody responders showed increased ventricular expansion and whole-brain volume loss, but with less cognitive decline, than the placebo group [132]. Hippocampal loss did not significantly differ between groups. The basis for the contradictory MRI and cognitive results is not understood. The authors noted that increased neuronal degeneration in the responder group was unlikely, due to the beneficial effect of treatment on cognitive function. The observed decrease in CSF tau levels with treatment also argues against greater neurodegeneration in the treated group. The authors speculated that increased CSF outflow resistance, mediated by the mobilization of amyloid, could have resulted in ventricular enlargement and reduced brain volumes. Although not available in the study, information on subregional atrophy rates across the cortex may have been able to aid in the interpretation of the nonspecific effects and shed more light on the effects of treatment on AD-related neurodegeneration.

Expert commentary

Qualitative structural neuroimaging, available for more than two decades, has been revolutionary in identifying gross structural lesions, but has failed thus far in identifying and differentiating between neurodegenerative illnesses. Recent advances in quantitative structural neuroimaging, however, offer hope that this scenario is about to change. As discussed above, US FDA-approved image analysis methods, operating on MRI data collected using ADNI-compatible image acquisition protocols can currently provide computer-assisted detection of objective signs of neurodegeneration consistent with AD. This, coupled with clinical impression, can be a tremendous tool for assisting diagnoses in clinical settings where a myriad of causes may underlie a patient’s subjective or objective cognitive impairment.

In cases where the clinical impression points to neurodegenerative illness, quantitative neuroimaging can provide objective and direct supportive evidence of such neurodegeneration, allowing the physician to more confidently provide forewarning to the patient and family, with stronger justification for pursuing a more aggressive course of intervention. Currently, this might include earlier social, financial and safety planning, or recommendation for enrollment in clinical trials with a higher risk:benefit ratio, given the increased likelihood of poor outcome if no treatment is pursued.

In other cases, the clinical impression may be an absence of neurodegenerative illness, which is not uncommon when treating educated, knowledgeable and, often, very worried elderly individuals. Quantitative neuroimaging can be quite helpful in these cases as well, providing evidence of preserved brain structure consistent with the lack of clinical evidence of objective decline. This reassurance may itself be therapeutic in individuals for whom extreme anxiety and fear of developing AD contributes to their perceived symptoms. Although one can never ‘rule out’ prodromal disease, current evidence would suggest reduced immediate risk in such cases, warranting an approach of watchful waiting.

Finally, in cases where the clinical impression is consistent with early AD, but structural imaging results show minimal neurodegeneration in memory structures, the physician may be appropriately motivated to perform a more thorough search for less common or reversible causes of cognitive impairment, such as medication side effects or depression-related ‘pseudodementia’. In the rushed world of clinical practice, an elderly patient with memory impairment might currently be given only a brief evaluation before being prescribed an acetylcholinesterase inhibitor. Whereas inappropriate acetlycholinesterase treatment can be problematic, inappropriate treatment with future aggressive disease-modifying agents could be devastating. Current evidence suggests that such treatments may be accompanied by risk of significant side effects, such as brain edema and possible hemorrhage [133]. Their administration will only be warranted when it is clear that the potential benefit of treatment justifies the risk. The ability to make this determination will necessitate greater confidence in the diagnosis of early AD.

In preparation for such a time when patient safety will depend on accurate diagnosis of early AD, it is important to gain more clinical experience with the available biomarkers. While the validation of accuracy in large, public datasets of carefully screened subjects, such as ADNI, have been invaluable, more data must be collected in actual clinical settings, where improvement of diagnostic accuracy can be documented in relatively unselected clinical populations. Thus, a concerted effort to fund and support clinical research on the most promising biomarkers in actual clinic settings will be critical for completing translation of the enormous strides made in research into the clinical realm. Only in this way will the payoff from the large prior investment in this research be delivered directly to patients in the clinic.

Five-year view

Late stage clinical trials are evaluating the efficacy of medications that reduce the production of amyloid oligomers by inhibiting enzymatic cleavage of the amyloid precursor protein [134]. The targeted enzyme, γ-secretase, also has a role in the notch pathway, important for a number of normal biological functions [135]. Inhibition of the notch pathway was probably the cause of increased skin cancer in the treatment arm of the semagacestat Phase III trial and may have caused the decline in cognitive and functional abilities among treated patients that led to the trial’s recent stoppage. This has not discouraged other trials of γ-secretase inhibitors from moving forward, but has raised attention to the importance of minimizing notch pathway inhibition in attempts to target the amyloid cleaving function of γ-secretase.

Another therapeutic approach that is in late-stage clinical trials is through passive immunization against the amyloid protein. This approach also has risks that are perhaps associated with nonselective removal of both vascular- and plaque-based amyloid. The removal of vascular amyloid may predispose patients to vasogenic brain edema and hemorrhage [133]. A higher degree of risk might be accepted in the treatment of such a devastating disease, but results of the clinical trials to date highlight the importance of improving confidence in early diagnosis before exposing patients to the risks associated with potential disease-modifying medications.

Given the possible near-term availability of AD medications targeting amyloid, it will soon be important to identify not only whether an individual suffers from progressive neurodegeneration, but whether or not that degeneration is mediated by increased amyloid deposition. For this decision, the combination of an amyloid biomarker, either reduced CSF levels of amyloid-β₄₂ or elevated binding of amyloid-sensitive PET tracers in the brain, will complement structural imaging evidence of brain atrophy. Biomarker evidence of elevated amyloid, in the absence of ongoing neuro-degeneration, would likely be insufficient to warrant aggressive treatment that may be associated with risk of serious side effects, since such individuals may remain cognitively stable throughout the rest of their lives. An analogy from the cardiovascular realm seems appropriate: treating elevated cholesterol, a risk-factor for heart disease, would be appropriate only if the medicine is associated with minimal side effects, whereas treating angina, a sign of ongoing heart damage, warrants more aggressive, immediate therapy and even angiocatheterization. Much like elevated cholesterol, elevated amyloid is not necessarily associated with imminent decline, whereas, like angina or active myocardial infarction, ongoing neurodegeneration will likely warrant an aggressive approach, such as infusion of anti-amyloid therapy, once available. For other neurodegenerative illnesses, where complementary biomarkers may be lacking, there may still be hope for prodromal differential diagnosis using quantitative structural MRI. Although current automated and US FDA-approved techniques for quantitative structural neuroimaging have yet to be evaluated for assisting with differential diagnosis, strong evidence exists for selective vulnerability of cortical networks in various neurodegenerative diseases. Therefore, a more specific diagnosis is likely to be obtained through quantitative examination of the integrity of these networks. However, to inform differential diagnosis in clinical practice, the technique must be sensitive enough not only to identify the regional atrophy pattern in individual subjects, but to differentiate the early stage AD pattern from atrophy patterns associated with early stages of other neurodegenerative diseases. Furthermore, the technique has to be easily integrated into the workflow of clinical practice and be robust enough to be applied to a wide range of patients with variable compliance, with an acceptable, very low, failure rate. Understandably, the field might be more than 5 years from developing such a robust procedure for differential diagnosis based on structural imaging, but progress will be made in this direction.

Progress will also be made within the next 5 years to permit enhanced detection of progressive change. As reviewed above, measures of longitudinal change over time provide improved predictive prognosis in MCI. Risk stratification can occur at the initial visit, based on comparison to age-matched normative data, and later, as the patient is followed, the initial assessment can be adjusted based on rates of subregional change. Increased rates of change in AD-related regions may increase confidence in a diagnosis of early AD, and allow staging of the disease. Conversely, if atrophy rates are similar to those of age-matched controls, or increased atrophy rates are observed in unexpected areas, the initial diagnosis may be revised, prompting consideration of other causes of impairment. The ability to examine an overlay map of change projected on the three-dimensional cortical reconstruction might allow identification of the cortical pattern of neurodegeneration that would aid in differential diagnosis.

The future for use of quantitative structural MRI in clinical practice for early detection and monitoring of AD progression is highly promising. Future efforts aimed at more fully translating research results into clinical practice will result in improved diagnostic ability, allowing for more reliable identification of patients in very early stages of AD. This will provide physicians with better information on which to base risk/benefit decisions when deciding how aggressively to treat an individual patient, once disease-modifying treatments become available.

Key issues.

• There is strong interest in using biological markers to identify Alzheimer’s disease (AD) before the onset of dementia, since disease-modifying treatments, once they become available, are more likely to be successful when given early.

• Quantitative structural MRI measures are a promising potential biomarker to aid in early detection of AD since they are highly sensitive to the neurodegeneration that occurs in mild and prodromal AD.

• AD is associated with a characteristic pattern of subregional atrophy, and the degree to which this atrophy pattern is present in individuals with mild cognitive impairment is predictive of conversion to AD.

• Atrophy rates in regions vulnerable to AD may be able to identify individuals at risk of developing AD even prior to the onset of cognitive impairment.

• Barriers to routine clinical use of quantitative structural MRIs are being overcome through the aid of large-scale clinical trials, such as the Alzheimer’s Disease Neuroimaging Initiative.

• Initial use of US FDA-approved methods for obtaining objective evidence of atrophy in brain structures vulnerable to AD clinical setting has shown that these measures provide important complementary data to the history, neurological and neuropsychological findings.

• Since new disease-modifying therapies may be associated with significant side effects, quantitative structural MRI measures, which can predict risk of imminent decline, will be critical for determining whether potential benefits of preventing or delaying decline outweigh risk of significant side effects, such as brain edema or hemorrhage.