Localization key and stereotactic surface projection map images of 4 example positron emission tomography scans with activity maps on the top row and z maps showing deviation from a normal control cohort on the second row. A, Localization key of brain regions as used by the raters. R indicates right; L, left. B, Scan of a 66-year-old healthy control subject and the color scale used for all positron emission tomographic images in the study. The local cerebral metabolic rate of glucose utilization (ICMRGlc) is indicated by the numbers along the top of the color scale, and the z score values are represented by the numbers across the bottom of the scale. C, Scan of a patient with Alzheimer disease with unanimous interpretations. D, Scan of a patient with frontotemporal lobar degeneration with unanimous interpretations. E, Scan of a patient with frontotemporal lobar degeneration with nonunanimous interpretations (votes: 7 for frontotemporal lobar degeneration, 5 for Alzheimer disease).
The number of scans and the degree of unanimity in the interpretation among 12 interpreters. Zero raters with incorrect interpretations indicates unanimous interpretations. Only 7 frontotemporal lobar degeneration (FTLD) scans (50%) had unanimous, correct interpretations; 27 Alzheimer disease (AD) scans (87%) had unanimous, correct interpretations. Of note, 2 of 4 AD scans with nonunanimous interpretations had only 1 of 12 raters in error.
Customize your JAMA Network experience by selecting one or more topics from the list below.
Womack KB, Diaz-Arrastia R, Aizenstein HJ, et al. Temporoparietal Hypometabolism in Frontotemporal Lobar Degeneration and Associated Imaging Diagnostic Errors. Arch Neurol. 2011;68(3):329–337. doi:10.1001/archneurol.2010.295
Frontotemporal lobar degeneration (FTLD) is the third most common degenerative dementia, behind Alzheimer disease (AD) and dementia with Lewy bodies.1 It is a heterogeneous disorder with at least 3 recognized clinical presentations,2 multiple histopathologic subtypes,3,4 and familial cases associated with mutations in 4 different genes5-9 with an additional genetic linkage on chromosome 9p.10-12
Despite the existence of consensus clinical diagnostic criteria, patients with FTLD are commonly misdiagnosed as having AD or a psychiatric illness.2,13-15 These mistakes are understandable given the insidious, progressive nature of both FTLD and AD and their shared symptoms.16 Both illnesses may have prominent behavioral changes, which can overlap symptoms typically seen in psychiatric disorders.17-19 While amnesia as the initial symptom of a progressive dementing disease strongly favors a diagnosis of AD, it also occurs in some patients with FTLD.20 Frontotemporal lobar degeneration may present with language deficits, but prominent language deficits also occur in AD.2,21-24 The difficulty in obtaining a detailed and reliable clinical history in some situations is a further challenge to accurate diagnosis and highlights the value of validated diagnostic biomarkers.
Despite the difficulties, accurate diagnosis is critical because the clinical management differs for AD and FTLD. The US Food and Drug Administration currently has approved 5 drugs for the treatment of AD: 4 cholinesterase inhibitors and an N-methyl-D-aspartate channel modulator.25 In contrast, no drugs have been shown to be effective in FTLD, although serotonin reuptake inhibitors are often used.26 Cholinesterase inhibitors can worsen behavioral symptoms in patients with FTLD and are generally avoided.27-29 The treatment of FTLD with memantine has been the subject of a few small trials, but the open-label design of these trials prevents definitive conclusions from being drawn.30-32 The treatment approaches for AD and FTLD will likely diverge even further with the anticipated arrival of specific disease-modifying therapies for AD.26,33
Brain imaging provides an independent, objective, and quantitative measure of disease that complements clinical information and can aid in distinguishing FTLD and AD. Voxel-based morphometric analysis of structural magnetic resonance imaging can detect differences in regional atrophy between groups of patients with FTLD, FTLD subtypes, and AD and controls.34,35 However, visual interpretation of individual magnetic resonance imaging scans, while helpful, can be misleading.36 Positron emission tomography with fludeoxyglucose F 18 (FDG-PET) typically shows sufficient abnormalities that can be used to improve the accuracy of distinguishing AD from FTLD in individual cases.37 Patients with AD characteristically have reduced activity most prominently in the posterior temporoparietal and posterior cingulate cortices.38 By contrast, the FDG-PET scans of patients with FTLD have hypometabolism that is most prominent in the frontal, anterior temporal, and anterior cingulate cortices.39 Metabolic abnormalities are not limited to these regions, however. As the severity of dementia increases, the severity and topographic extent of hypometabolism also increase and begin to involve other regions. Likewise, there is considerable heterogeneity in the individual pattern of hypometabolism that reflects the patient's clinical symptoms. Consequently, considerable judgment is required for visual diagnostic interpretation. Analytic techniques such as stereotactic surface projection maps (SSP) that incorporate both metabolic and statistical information further improve diagnostic accuracy of FDG-PET scan interpretation as compared with standard transaxial images.40
In a previous study using the same series of SSP-processed FDG-PET scans that are used in this current analysis, individual raters were able to interpret the scans of patients with autopsy-confirmed AD with a very high degree of sensitivity (97.8%) and confidence. Scans from patients with autopsy-confirmed FTLD, however, had more variability of interpretation, resulting in reduced sensitivity (70.2%) and confidence; however, there remained in patients with FTLD a significant positive effect on diagnostic accuracy as compared with clinical assessment alone (positive likelihood ratio = 36.5).37 We decided to further evaluate the inconsistencies between individual raters for their interpretations of these FTLD FDG-PET scans to see what features were associated with inaccurate scan interpretation and to provide guidance in improving diagnosis. In standard clinical settings, a scan will typically be interpreted by a single individual without the benefit of a diagnostic consensus process often used in research. Identifying and describing features commonly found in the FDG-PET scans of patients with FTLD that are associated with inaccurate interpretations may improve the diagnostic accuracy of these scans in clinical practice.
The data for this analysis came from 2 different studies that evaluated the utility of FDG-PET to distinguish AD from FTLD. Each study used a group of 6 raters who reviewed the same series of FDG-PET scans.37 In both studies, the raters individually interpreted each scan as being most consistent with either AD or FTLD before any discussion took place and while blinded to all clinical information. This yielded 12 independent interpretations for each scan, from which we could observe the degree of discrepancy between the raters. The members of 1 group also rated each of 10 regions (5 regions on the right side and 5 regions on the left side) as normal or abnormal.
To simplify the comparisons, we classified a region as abnormal if it was judged to be abnormal on either the left or right side, yielding 5 regions.
A previously described group of 45 patients with dementia, FDG-PET scans, and subsequent postmortem histopathological diagnoses of either AD or FTLD was used for this study.37 This group comprised all patients meeting these criteria whose scans were obtained at the University of Michigan between December 1, 1984, and July 31, 1998, and for whom retrievable medical records as well as technically adequate parametric FDG-PET scans were available. A summary of the subjects' characteristics is provided in Table 1. Frontotemporal lobar degeneration is caused by several distinct pathologies. We did not have the information to categorize each of the pathologies but provide the pathologic classification from the autopsy report. We did not attempt to analyze the data by pathologic subtype because our sample size was not large enough to be subdivided and still retain statistical validity. A database of FDG-PET scans from 33 healthy elderly patients of a similar age were used for statistical comparison with patient scans as previously described.37
There were 2 different groups of raters used in this study. Each group consisted of 6 members, for a total of 12 raters. Ten of the raters were neurologists and 2 were psychiatrists. All had extensive experience in dementia care at 8 National Institute on Aging–funded Alzheimer's Disease Centers. The raters had variable experience with FDG-PET imaging, ranging from expert to novice. Each rated the scans independently, without knowledge of the opinions of the others and blinded to any clinical data.
The data used in these analyses were from the interpretation of SSP-processed FDG-PET scans. The SSP method is an automated analysis method that warps images into a common stereotactic space and allows for statistical analysis of individual scans as compared with a control group. This results in 6 surface projection maps that are displayed both as a metabolic map and as a statistical map showing surface pixel-by-pixel z scores derived from comparison with a control group. Examples of the maps are shown in Figure 1. (See the article by Foster et al37 for further details.)
All raters completed a 2-hour training session to establish a uniform approach to scan interpretation and to familiarize the raters with the SSP presentation of FDG-PET data.26 Interpretation was based on the evaluation of 5 regions of the cerebral cortex in each hemisphere and judging the relative degree of abnormality in regions typically affected in AD (temporoparietal and posterior cingulate cortices) and FTLD (anterior temporal, frontal, and anterior cingulate cortices). The raters were not instructed to weigh any particular region more heavily than another but rather to base their final interpretation on whether the preponderance abnormalities were in AD- or FTLD-associated cortical areas. The training used 25 scans from clinically diagnosed patients and healthy elderly controls (10 from patients with AD, 10 from patients with FTLD, and 5 from controls) that were not part of the experimental data set. The 5 regions were reviewed to establish consistent interpretation of the anatomical boundaries of each region (Figure 1).
Interrater reliability for the 6 raters judging regional abnormalities was assessed using κ statistics calculated for all possible rater pairs. The level of agreement based on the κ statistics was classified as fair (κ = 0.20-0.39), moderate (κ = 0.40-0.59), substantial (κ = 0.60-0.79), or almost perfect (κ = 0.80-1.00).41
If 4 or more of the 6 raters who rated regional metabolism thought that a region was hypometabolic, then it was considered abnormal. Associations between regional hypometabolism and a pathologically verified diagnosis of AD or FTLD were evaluated using a χ2 test with Yates correction. Sensitivity, specificity, odds ratios, and positive likelihood ratios were calculated for hypometabolism in the temporoparietal and posterior cingulate cortices for a pathologic diagnosis of AD, while these same measures were calculated for the frontal, anterior cingulate, and anterior temporal cortices for a pathologic diagnosis of FTLD. The positive likelihood ratio incorporates sensitivity and specificity into a single measure: (sensitivity)/(1 − specificity). This represents the probability of a positive test result in an individual with the disorder divided by the probability of a positive test result in an individual without the disorder. A positive likelihood ratio greater than 1 means that a positive test result is more likely to occur in patients with the disease than in those without the disease.
For the pathologically verified FTLD cases, associations between regional hypometabolism and lack of unanimity among the raters for their overall interpretation were evaluated with the Fisher exact test.
Interrater reliability for judging individual regions as normal or abnormal was substantial for the temporoparietal cortex and only slightly less so for the frontal and posterior cingulate cortices (Table 2). However, interrater reliability was only moderate for the anterior cingulate and anterior temporal cortices, which are typically affected in FTLD.41 As expected from previous research,38,42 our raters found hypometabolism in the temporoparietal and posterior cingulate regions much more frequently in AD than in FTLD (odds ratios, 14.5 and 7.2, respectively) (Table 3). Nevertheless, 7 patients with FTLD (50%) had temporoparietal hypometabolism. Temporoparietal hypometabolism was more sensitive, but posterior cingulate hypometabolism was more specific for AD.
Likewise, our raters found the expected higher frequencies of hypometabolism in the frontal, anterior cingulate, and anterior temporal regions in FTLD as compared with AD (Table 4). Despite what might be expected, the presence of frontal hypometabolism alone did not significantly increase the likelihood of FTLD (odds ratio = 3.3). Patients with AD with or without frontal hypometabolism did not differ significantly with respect to age (mean age, 64.6 vs 66.2 years, respectively; P = .72). On the other hand, anterior cingulate hypometabolism and anterior temporal hypometabolism were much more likely in FTLD cases. All FTLD scans had hypometabolism in at least 1 of the typical FTLD areas, and all but 1 of the FTLD scans had reductions in the anterior cingulate and/or anterior temporal cortices. Hypometabolism in the anterior cingulate and anterior temporal cortices had higher specificities and higher positive likelihood ratios for a diagnosis of FTLD than hypometabolism in the temporoparietal cortex had for AD. Even in the presence of temporoparietal hypometabolism, hypometabolism in the anterior cingulate cortex and hypometabolism in the anterior temporal cortex were each strongly associated with a diagnosis of FTLD rather than AD (Table 5).
The 12 raters who provided an overall interpretation of the scans were unanimous in their decisions 76% of the time (34 of 45 interpretations), and all unanimous decisions were also correct. Because nonunanimity would correspond to interpretation errors on the part of some raters, we looked to see what factors, if any, were associated with this subset of misdiagnosed scans (Table 6). Of the FTLD scans, 7 (50%) had nonunanimous interpretations with a range of 1 to 11 incorrect among a total of 12 raters (Figure 2). In contrast, only 4 AD scans (13%) lacked unanimity, demonstrating a strong association of nonunanimous FDG-PET interpretation with a diagnosis of FTLD (P = .02 by Fisher exact test; Pearson ϕ = 0.79). Clearly, raters had more difficulty with FTLD scans. Among the 4 AD scans with nonunanimous decisions, 2 had only 1 discrepant interpretation. In both of the AD cases that had more than 1 discordant interpretation, the posterior cingulate cortex was judged to be normal and at least 1 FTLD-associated area was judged to be abnormal. Because of the small number of these cases, we did not analyze them further. In the FTLD cases, hypometabolism in the temporoparietal cortex was significantly associated with nonunanimous interpretations, occurring in 6 of 7 nonunanimous scans and in only 1 of 7 unanimously decided scans (Table 6). Posterior cingulate abnormalities were not independently associated with nonunanimity beyond the trend level, and all FTLD scans that had posterior cingulate hypometabolism also had temporoparietal abnormalities. No individual FTLD areas were independently associated with unanimity. Five of the FTLD scans had hypometabolism in all 3 FTLD-associated areas, and all of these scans had unanimous interpretations.
Temporoparietal involvement in FTLD that is detectable by both magnetic resonance imaging and single-photon emission computed tomography has been noted previously, particularly with respect to its association with progranulin mutations.43-45 Corticobasal degeneration, which is part of the FTLD spectrum of disorders, frequently involves the parietal cortex as well.46,47 Parietal atrophy has also been demonstrated in patients with microtubule-associated tau protein mutations, although it is less than what is seen with progranulin mutations.48
In our sample, the presence of temporoparietal hypometabolism on FDG-PET imaging was a common finding in the FTLD cases. This raises concern from a diagnostic standpoint because many use hypometabolism in the temporoparietal region as a reliable sign of AD. While we found the sensitivity of temporoparietal abnormalities to be quite good for AD (93.6%), the specificity was only 50%. This reduced specificity had consequences because temporoparietal hypometabolism had a disproportionate effect on interpretation errors for patients with FTLD. All of the FTLD scans with temporoparietal abnormalities also had hypometabolism in at least 1 or more areas associated with FTLD, and most had abnormalities in at least 2 FTLD regions. This suggests that evidence for AD may have a tendency to trump evidence for FTLD in FDG-PET interpretation. Our findings demonstrate, however, that hypometabolism in the anterior cingulate and anterior temporal regions should carry at least as much weight for a diagnosis of FTLD as temporoparietal hypometabolism carries for a diagnosis of AD, even when this is seen in the presence of temporoparietal hypometabolism. While we found associations of anterior cingulate and anterior temporal hypometabolism with FTLD, we did not find an association with hypometabolism of the frontal cortex (lateral and dorsolateral) with FTLD. These findings are consonant with other work, which has carefully looked at patterns of atrophy that distinguish FTLD from AD. Atrophy of the paralimbic fronto-insular-striatal network, of which the anterior cingulate cortex is a part, distinguishes FTLD from AD, while atrophy of the dorsolateral frontal cortex does not.49 These findings in turn mirror the distribution of the von Economo neurons. These neurons are found in the anterior cingulate and anterior insular cortices and are absent from the dorsolateral frontal lobes. They are preferentially and severely affected early in the course of FTLD and may underlie this specific distribution of atrophy50,51 or, in the case of our data, hypometabolism. Our data show that relying more on anterior temporal and especially anterior cingulate hypometabolism for a diagnosis of FTLD would improve the accuracy of scan interpretation.
Ultimately, interpretation of an FDG-PET scan to distinguish between AD and FTLD cannot be based on the presence or absence of hypometabolism in a single region. Instead, overreliance on findings in a single region of the cortex should be avoided by considering all likely affected regions and determining the relative degree of hypometabolism in each, in terms of both intensity and topographic extent.
There are several limitations to our study. Our sample size was relatively small, particularly with respect to the number of patients with FTLD. Optimally there would be similar numbers of patients with FTLD and patients with AD; however, obtaining such a group of patients with FTLD with both technically adequate FDG-PET scans and pathologic confirmation of their diagnosis would be difficult. We used the majority opinion of raters to define the presence or absence of regional hypometabolism. More objective measures of hypometabolism could give different results, but to be clinically meaningful, a finding must be perceptible to an interpreter. We thus believe that our approach provides more practical value for clinical applications. This is a convenience sample, with patients scanned at various points during the course of their illness. While this study provides some general guidelines for image interpretation, it is possible that different algorithms would be ideal for early diagnosis and when there already are severe deficits. Nevertheless, in current practice, determining the cause of dementia is often delayed and patients can be first scanned at any point in their illness.
The findings of this study are particularly relevant given the somewhat recent and growing use of FDG-PET in dementia evaluations. Although recently approved for this use by the Centers for Medicare and Medicaid Services in the United States, relatively few physicians have been trained to appreciate the complexity of FDG-PET patterns of hypometabolism seen in dementia. This may lead to reliance on an overly simplified interpretation scheme, such as the presence or absence of temporoparietal hypometabolism as the primary deciding factor between AD and FTLD. The results of this study indicate that such an AD-centric approach to FDG-PET interpretation may produce interpretation errors in a substantial proportion of patients with FTLD. The current Medicare guidelines for the use of FDG-PET in dementia recognize it as an appropriate study to distinguish between AD and FTLD when the clinical evaluation cannot. If this criterion is applied correctly by ordering physicians, then the proportion of patients with FTLD relative to patients with AD will be much larger in the subset of patients with dementia receiving FDG-PET scans than in the clinical dementia population.
The clinician will ultimately have to reconcile clinical, laboratory, and imaging data to make a final, accurate diagnosis. Imaging with FDG-PET improves diagnostic accuracy in dementia, but this effect is in turn dependent on accurate scan interpretation. Understanding the moderate specificity of temporoparietal hypometabolism for AD and the relatively high specificity and positive likelihood ratio of anterior cingulate and anterior temporal hypometabolism for FTLD may improve FDG-PET scan interpretation and therefore maximize the positive effect of these studies on diagnostic accuracy.
Correspondence: Kyle B. Womack, MD, Department of Neurology, University of Texas Southwestern Medical Center at Dallas, 5323 Harry Hines Blvd, Dallas, TX 75390-9129 (firstname.lastname@example.org).
Accepted for Publication: September 16, 2010.
Published Online: November 8, 2010. doi:10.1001/archneurol.2010.295 This article was corrected for errors on November 12, 2010.
Author Contributions:Study concept and design: Womack, Diaz-Arrastia, Gabel, and Foster. Acquisition of data: Aizenstein, Arnold, Barbas, Clark, DeCarli, Jagust, Leverenz, Peskind, Turner, Zamrini, Heidebrink, Burke, DeKosky, Farlow, Gabel, Kawas, and Foster. Analysis and interpretation of data: Womack, Diaz-Arrastia, Boeve, DeKosky, Higdon, Koeppe, Lipton, and Foster. Drafting of the manuscript: Womack, Aizenstein, and Foster. Critical revision of the manuscript for important intellectual content: Womack, Diaz-Arrastia, Arnold, Barbas, Boeve, Clark, DeCarli, Jagust, Leverenz, Peskind, Turner, Zamrini, Heidebrink, Burke, DeKosky, Farlow, Gabel, Higdon, Kawas, Koeppe, Lipton, and Foster. Statistical analysis: Womack, Higdon, and Koeppe. Obtained funding: Womack, Turner, DeKosky, and Foster. Administrative, technical, and material support: Arnold, DeCarli, Heidebrink, DeKosky, Farlow, Gabel, and Foster. Study supervision: Diaz-Arrastia and Foster.
Funding/Support: This work was supported by grants AG22394 and AG30006 from the National Institutes of Health; an anonymous private donation to the Center for Alzheimer's Care, Imaging, and Research; the pilot cooperative project grant AG16976 from the National Alzheimer's Coordinating Center; and grants from the following National Institutes of Health Alzheimer's Disease Research Centers: Michigan (grant AG08671), University of California at Davis (grant AG10129), University of Pennsylvania (grant AG10124), University of California at Irvine (grant AG16573), Duke University (grant AG028377), Indiana University (grant AG10133), University of Pittsburgh (grant AG05133), and University of Texas Southwestern Medical Center at Dallas (grant AG12300).
Additional Contributions: Sid Gilman, MD, FRCP, Henry Buchtel, PhD, and R. Scott Turner, MD, PhD, made images from their research available for this study, and Angela Y. Wang, PhD, provided valuable assistance with Figure 1.
Financial Disclosure: Dr Farlow receives research funding from Bristol-Myers Squibb, Danone, Elan, Eli Lilly, Forest, Janssen, Medivation, Pfizer, Novartis, OctaPharma, and Sonexa; is a scientific consultant for Adamas, Accera, AstraZeneca, Astellas, BioRx, CoMentis, Cortex Pharmaceuticals, Eisai, Dainippon Sumitomo Pharma, Eli Lilly, GlaxoSmithKline, Medivation, Merck, Novartis, Noven, OctaPharma, QR Pharma, Sanofi-Aventis, Schering-Plough, Suven Life Sciences Ltd, and Toyama Chemical Co; is a speaker for Eisai, Forest, Janssen, Pfizer, and Novartis; receives royalties for intellectual property from Elan; and has a spouse who is employed by and receives a salary from Eli Lilly.
Create a personal account or sign in to: