Differentiating Cerebral Amyloid Angiopathy From Alzheimer’s Disease Using Dual Amyloid and Tau Positron Emission Tomography
Article information
Abstract
Background and Purpose
Although amyloid positron emission tomography (PET) might provide a molecular diagnosis for cerebral amyloid angiopathy (CAA), it does not have sufficient specificity for this condition relative to incipient Alzheimer’s disease (AD). To identify a regional amyloid uptake pattern specific to CAA, we attempted to reduce this overlap by selecting “pure CAA” (i.e., fulfilling the criteria for probable CAA but without tau PET AD signature) and “pure AD” (i.e., positive amyloid PET and presence of tau PET AD signature, but without lobar hemorrhagic lesions). We hypothesized that occipital tracer uptake relative to the whole cortex (WC) would be higher in patients with pure CAA and may serve as a specific diagnostic marker.
Methods
Patients who fulfilled these criteria were identified. In addition to the occipital region of interest (ROI), we assessed the frontal and posterior cingulate cortex (PCC) ROIs that are sensitive to AD. Amyloid PET uptake was expressed as the absolute standardized uptake value ratio (SUVR) and ROI/WC ratio. The diagnostic utility of amyloid PET was assessed using the Youden index cutoff.
Results
Eighteen patients with AD and 42 patients with CAAs of comparable age were eligible. The occipital/WC was significantly higher in CAA than AD (1.02 [0.97–1.06] vs. 0.95 [0.87–1.01], P=0.001), with an area under curve of 0.762 (95% confidence interval [CI] 0.635–0.889) and a specificity of 72.2% (95% CI 46.5–90.3) at Youden cutoff (0.98). The occipital lobe, frontal lobe, PCC and WC SUVRs were significantly lower in CAA than in AD. The frontal/WC and PCC/WC ratios did not differ significantly between the groups.
Conclusion
Using stringent patient selection to minimize between-condition overlap, this study demonstrated the specificity of higher relative occipital amyloid uptake in CAA than in AD.
Introduction
Sporadic cerebral amyloid angiopathy (CAA) is the main cause of spontaneous lobar intracerebral hemorrhage (ICH) and is an important contributor to cognitive decline in the elderly [1]. Accurate diagnosis of CAA has major implications for patient prognosis, management, and therapeutic trials. Positron emission tomography (PET) imaging with amyloid-binding tracers has potential as a molecular diagnosis of CAA because it detects not just parenchymal but also cerebrovascular Aβ deposition [2]. Previous studies, at the group level, showed significantly higher tracer retention in CAA compared to non-demented healthy controls or patients with hypertensive deep ICH, suggesting that a negative amyloid scan reasonably excludes CAA [3].
However, the specificity of high amyloid uptake for CAA has been questioned, compromising the potential routine application of amyloid PET as a rule-in for CAA [4]. This is thought to reflect the wide prevalence of incipient Alzheimer’s disease (AD). Indeed, the finding of positive amyloid PET steeply increases after age 60, in cognitively normal individuals [5]. CAA and AD do have pathological overlap, further complicating matters [2,4,6,7]. Apart from global cortical uptake or visual assessment, regional amyloid tracer uptake relative to the whole cortex (WC) has been investigated as a potential discriminator between CAA and AD, focusing on the occipital cortex, which shows the highest vascular amyloid burden in post-mortem studies [8]. Accordingly, higher relative occipital uptake has been reported in CAA as compared to AD, but sensitivity and specificity are considered insufficient for clinical application [2,3,9-13]. Besides, tau PET detects in vivo intracellular abnormal hyperphosphorylated tau [14], a histopathologic hallmark for AD, and affords around 90% sensitivity and specificity using an AD regional tau PET uptake “signature.” [15,16]
To reassess the value of amyloid PET for differentiating CAA from AD, we aimed in the present study to compare subsets of patients with CAA and AD expected to have the lowest possible chance of overlap. To achieve this, we selected patients fulfilling the Boston diagnostic criteria for probable CAA [17,18] without tau PET AD signature (“pure CAA”) and those fulfilling the current diagnostic criteria for probable AD, including positive amyloid PET and tau PET AD signature, but without lobar hemorrhagic lesions on magnetic resonance imaging (MRI) (“pure AD”). Our main goal was to assess the specificity of relative occipital cortex tracer uptake for differentiating pure CAA from pure AD. Our primary hypothesis was that this ratio would be significantly higher in the former. Based on the previously reported lower global cerebral uptake of amyloid tracers in CAA compared to AD [3], our secondary hypothesis was that cerebral cortex uptake, particularly the posterior cingulate cortex (PCC) and frontal lobe uptake, two regions known to have the highest amyloid uptake in AD [13], will have good sensitivity for differentiating pure AD from pure CAA.
Methods
Data availability
All data were obtained from the National Taiwan University Hospital (NTUH). Anonymized data not published in this article will be made available upon request by any qualified investigator.
Patient enrollment
Starting in October 2018, we prospectively enrolled patients attending the Memory Clinic or the Stroke Clinic of NTUH and NTUH Bei-Hu Branch who presented with cognitive complaints or spontaneous lobar ICH and fulfilled the clinical diagnosis of probable CAA based on the Boston criteria v1.5 [17] until May 2022 and the 2.0 criteria thereafter [18]. Patients who fulfilled a diagnosis of probable AD dementia based on the diagnostic criteria of the National Institute on Aging-Alzheimer’s Association [19] were recruited as the comparison group. All patients who agreed to participate in this study underwent brain MRI, amyloid PET, and tau PET. Patients with CAA were categorized as CAA-ICH or CAA-cognitive impairment (CAA-CI) according to their initial presenting symptoms of spontaneous lobar ICH or cognitive complaints with objective cognitive impairment, respectively.
The following demographic characteristics and clinical data were prospectively collected: age, sex, years of education, and diagnoses of chronic hypertension, diabetes, or hypercholesterolemia. At enrollment, cognitive status was assessed in each patient through a combination of history-taking and objective cognitive assessment by trained clinical psychologists and included the Mini-Mental State Examination (MMSE) and Clinical Dementia Rating (CDR) scale, a 5-point scale that assesses cognition and functional autonomy in subjects with neurocognitive disorders [20,21].
MRI acquisition and analysis
Brain MRI was performed using a 3-Tesla scanner (Siemens Verio, TIM, or mMR; Siemens Medical Solutions, Erlangen, Germany). The imaging protocol included T1-weighted magnetization-prepared rapid gradient-echo imaging (flip angle 9, repetition time/echo time=1,460/2.39 ms, field of view=25.6 cm, and slice thickness=1 mm), T2-weighted imaging (repetition time/echo time=3,530/83 ms, field of view=23 cm, and slice thickness=5 mm), fluid attenuated inversion recovery (FLAIR) (repetition time/echo time=10,000/89 ms, field of view=23 cm, and slice thickness=5 mm), susceptibility-weighted imaging (SWI) (flip angle 15, repetition time/echo time=28/20 ms, matrix number=221×320, field of view=23 cm, and slice thickness=2 mm), diffusion-weighted imaging, and apparent diffusion coefficient maps, as we previously described [22].
MRI markers were evaluated based on the Standards for Reporting Vascular Changes on Neuroimaging criteria [23]. Briefly, the number of cerebral microbleeds (CMBs) and the presence of cortical superficial siderosis (cSS) were evaluated using axial SWI sequences according to current standards [24,25]. CMBs in the lobar regions, deep regions, and cerebellum were counted. Lacunes were evaluated in the supratentorial region and defined as “round or ovoid, subcortical, fluid-filled cavities, from 3 to 15 mm in diameter.” White matter hyperintense (WMH) lesion volume was calculated based on FLAIR imaging of the ICH-free hemisphere and multiplied by two, as previously proposed [26]. Perivascular spaces (PVS) were evaluated on T2-weighted imaging and defined as sharply delineated structures measuring less than 3 mm following the course of perforating or medullary vessels [27]. High-degree PVS was defined as more than 20 visible PVS in the centrum semiovale or in the basal ganglia on the side of the brain with more severe involvement [27].
PET acquisition and analysis
Radiotracers were prepared at the Cyclotron and Radiopharmaceutical Laboratory at NTUH. PET images (Discovery ST; GE Healthcare, Chicago, IL, USA) were acquired over 30 min, starting 40 min after the injection of 10 mCi 11C-Pittsburgh compound B (PiB) or 18F-T807. As previously described [22], PET data were reconstructed via ordered set expectation maximization and corrected for attenuation.
For probable AD samples, amyloid scan positivity was determined by two nuclear medicine specialists using a previously proposed visual assessment [28]. Any discordant results were resolved by consensus.
Using the PMOD software (PNEURO module; PMOD Technologies, Zurich, Switzerland), high-resolution T1-weighted MRI was auto-segmented using Montreal Neurological Institute space/coordinates based on the maximum probability Anatomical Automated Labeling (AAL) atlas. PET data were matched and normalized to the corresponding T1 images. Standardized uptake values (SUV) were extracted from the cerebellar cortex, pons, and cortical ribbon regions of interest (ROIs). As per our primary and secondary hypotheses, for the PiB scans, the ROIs included the occipital cortex, frontal cortex, PCC, and WC. PiB images were semi-quantitatively analyzed and expressed as SUV ratios (SUVRs). We chose the pons as the reference region in preference to the cerebellar cortex, given that CAA could involve the cerebellar cortex, resulting in increased cerebellar PiB uptake [29,30]. We did not apply the centiloid method to quantify the amyloid load in the present study because this method is based on a voxel-of-interest mask that includes all brain regions with high uptake in AD [31], which may miss regions that are more sensitive to CAA.
Regarding T807 scans, a meta-temporal ROI was created as an AD signature, which included the entorhinal cortex, amygdala, parahippocampal gyrus, fusiform gyrus, and inferior and middle temporal gyri [32,33]. A cutoff of SUVR 1.26 (reference: cerebellar cortex) was applied to determine tau PET positivity based on results in previous cohorts using the mean and two standard deviations (SD) of T807 uptake in the meta-temporal ROI from amyloid-negative and cognitively unimpaired participants [21].
Statistical analysis
The baseline demographic information and neuroimaging variables of patients with CAA and AD were compared. Discrete variables are presented as counts (%) and continuous variables as mean (±SD) or median (interquartile range), as appropriate, based on their distribution. Categorical variables were analyzed using Fisher’s exact test, and continuous variables were analyzed using the Mann–Whitney U-test. For PiB scans, we compared the absolute and relative uptake ratios in the CAA-sensitive (occipital cortex) and AD-sensitive regions (frontal cortex, PCC). For the sensitivity analysis, we compared AD to the pre-specified CAA subgroups: CAA-ICH and CAA-CI.
We used the Youden index to determine the best cutoff value for the diagnostic utility of amyloid PET findings. The sensitivity, specificity, positive predictive value (PPV), negative predictive value (NPV), and the area under curve (AUC) were determined for these ROIs. We primarily focused on the specificity of the occipital to WC ratio to differentiate CAA from AD. Our secondary focus was on the frontal cortex and PCC to differentiate AD from CAA.
All statistical analyses were performed using SPSS version 25 (IBM Corp., Armonk, NY, USA). Significance was established as a two-tailed less than 0.05.
Standard protocol approvals, registrations, and patient consent
This study was approved by the Institutional Review Board (201903069RINB) of NTUH and was conducted in accordance with their guidelines. Written informed consent was obtained from all participants or their family members.
Results
Patients
A flowchart of the study is shown in Supplementary Figure 1. A sample of 60 patients fulfilled the eligibility criteria for the present study, including 42 patients with pure CAA (mean age, 73.8±6.9 years, 54.8% male; 24 with ICH and 18 with cognitive presentation) and 18 patients with pure AD patients (mean age, 73.4±8.2 years, 38.9% male). All pure CAA cases retrospectively fulfilled the Boston v2.0 criteria for probable CAA [18]. Figure 1 shows images of illustrative patients from the two groups.
Representative MRI, amyloid and tau scans in CAA and AD. (A) A patient with probable “pure” CAA with numerous lobar cerebral microbleeds but negative tau PET scan. (B) A patient with “pure” AD with positive amyloid and AD-signature tau PET scans. No lobar hemorrhagic lesions were found on SWI. MRI, magnetic resonance imaging; SWI, susceptibility-weighted imaging; PiB, Pittsburgh compound B; PET, positron emission tomography; CAA, cerebral amyloid angiopathy; AD, Alzheimer’s disease.
Table 1 presents the main demographic, clinical, and MRI data. The two groups had similar age and sex distributions. As expected, patients with AD had significantly lower MMSE scores (18.1±6.5 vs. 24.3±6.5, P=0.001) and higher CDR (1 [0.5–1] vs. 0.5 [0.5–0.5], P<0.001). On MRI, patients with CAA also as expected had more lobar CMBs (11.6±23.8 vs. 0.0±0.0, P=0.003) and a higher prevalence of cSS (26.2% vs. 0.0%, P=0.024), as well as more WMHs (13.6±11.9 mL vs. 4.2±3.4 mL, P<0.001).
Comparison of the demographics between the pure CAA and pure AD groups
Amyloid PET data
Amyloid PET data for the CAA and AD groups are presented in Table 2. Compared to AD, the occipital/WC ratio was significantly higher in CAA (1.02 [0.97–1.06] vs. 0.95 [0.87–1.01], respectively; P=0.001) (Figure 2A). The occipital cortex SUVR was significantly lower in CAA (0.78 [0.66–1.01] vs. 1.07 [0.81–1.19], respectively; P=0.002) (Figure 2B).
Whole cortex and regional PET amyloid uptake in pure CAA versus pure AD
Comparisons of amyloid PET uptake in the occipital cortex and whole cortex. Box plots show (A) relative amyloid uptake in the occipital cortex (occipital/ whole cortex ratio), (B) absolute amyloid burden in the occipital cortex, and (C) absolute amyloid burden in the whole cortex. SUVR, standardized uptake value ratio; AD, Alzheimer’s disease; CAA, cerebral amyloid angiopathy; PET, positron emission tomography. *P<0.05.
The WC SUVR was significantly higher in AD than in CAA (SUVR=1.15 [0.98–1.26] vs. 0.79 [0.64–0.99]; P<0.001) (Figure 2C). Likewise, the frontal cortex and PCC SUVRs were significantly higher in AD (frontal cortex: 1.16 [0.97–1.30] vs. 0.80 [0.63–0.98], respectively; P<0.001) and PCC: 1.35 [1.11–1.49] vs. 0.94 [0.73–1.19], respectively; P<0.001), but not the corresponding ROI/WC ratios (all P>0.05) (Supplementary Figure 2).
Regarding the pre-specified CAA subgroup analysis, the results were essentially unchanged from the whole sample regarding both the CAA-CI (n=18) and CAA-ICH subgroups (n=24), with slightly stronger P-values for the former (Table 3).
Whole cortex and regional amyloid uptake in CAA-CI and CAA-ICH versus AD
Diagnostic performance of amyloid PET in the differentiation between CAA and AD
Regarding the occipital/WC ratio, the best cutoff for discriminating CAA from AD based on the Youden index was more than 0.98. This cutoff provides a specificity of 72.2% (95% confidence interval [CI], 46.5–90.3) with AUC of 0.762 (95% CI 0.635–0.889) (Table 4). The diagnostic performance of the occipital/WC ratio in the two pre-specified CAA subgroups (CAA-CI and CAA-ICH) is shown in Supplementary Table 1. We observed similar specificities for differentiating the two presentation subgroups of AD.
Diagnostic performance of occipital/WC ratio on amyloid PET for CAA versus AD
The best SUVR cutoffs to differentiate pure AD from pure CAA for WC, frontal cortex, and PCC were >1.00, >1.07, and >1.20, respectively. The diagnostic performance of these ROIs is presented in Table 5. Overall, the SUVR cutoffs provided a sensitivity of 72.2%–77.8%, specificity of 78.6%–83.3%, and AUC of 0.803–0.823.
Diagnostic performance of SUVR cutoffs on amyloid PET for AD versus CAA
Discussion
In this study, we used dual amyloid and tau PET to optimize the distinction between “pure” CAA and AD groups, with presumably predominant vascular versus parenchymal Aβ, respectively. Consistent with our hypothesis, we found significantly higher relative PiB uptake in the occipital cortex in the CAA group than in the AD group. Furthermore, the occipital/WC ratio on amyloid PET provided fair diagnostic specificity (72.2%) for differentiating CAA from AD. Additionally, the WC, frontal cortex, and PCC exhibited markedly lower absolute PiB uptake in pure CAA than in pure AD, with SUVR cutoffs providing good diagnostic accuracy. Therefore, this study establishes that amyloid tracer uptake is much lower in absolute terms in CAA than in AD and furthermore shows a predominantly occipital relative distribution in CAA. In turn, these amyloid patterns may have diagnostic utility in clinical practice, particularly for the cognitive subtype where distinction from AD is more difficult and in which the performance of the current Boston criteria is being questioned [34].
Previous studies and a pooled meta-analysis suggested higher relative occipital uptake may differentiate CAA from AD, but the effect size was low and was considered insufficient for clinical application [2,3,11,12]. This was attributed to the presence of substantial incipient AD pathology in the CAA cohorts studied [5,35]. Despite the present attempt to reduce as far as possible the odds of such overlap by means of selecting patients with “pure” CAA and AD, the difference in relative occipital tracer uptake between CAA and AD, although statistically significant, remained small (mean value: 1.02 vs. 0.95, respectively), with individual values exhibiting substantial overlap. This situation may reflect a high topographical correlation between vascular and parenchymal Aβ pathologies [36]. A study in an aged cohort with available postmortem data reported that in vivo amyloid uptake correlated more closely with parenchymal than vascular Aβ [37]. However, amyloid tracers do significantly bind to vascular amyloid uptake in vivo, as shown by work in both asymptomatic and less than 60-year-old symptomatic patients with hereditary CAA, where almost pure vascular Aβ accumulation is expected [38,39]. Using strict selection criteria to reduce CAA and AD overlap as far as possible, our study has demonstrated that the occipital predominant uptake pattern on amyloid PET has value as a specific marker in CAA and may have clinical utility to differentiate CAA from AD. In addition to 11C-PiB PET, fluorine-18-labeled amyloid PET ligands such as 18F-florbetaben and 18F-flutemetamol have been shown to have similar global and regional signal retention characteristics as 11C-PiB in both healthy controls and patients with AD [40,41]. Therefore, we can safely assume that our results have direct applicability to these commercially available PET ligands using the AAL atlas.
The significantly higher amyloid uptake in AD as compared to CAA in the current study is in line with previous reports both in sporadic [3] and Dutch-type CAA [39] and indicates a globally higher Aβ load in AD. It might also partly reflect a lower tracer binding affinity for vascular Aβ in CAA [42]. and PET tracers that specifically target vascular Aβ plaque are being investigated [43-45]. On a clinical standpoint, our data suggest that, over and above the occipital/WC ratio, global amyloid uptake might help in the differentiation of AD from CAA as a two-pronged approach. For instance, in a patient with a cognitive presentation, a high occipital/WC ratio would help rule in CAA, while a concomitantly low amyloid SUVR would help rule out AD. Thus, in our cognitive phenotype cohort, post hoc calculations using both markers together showed PPV (rule-in) and NPV (rule-out) of 85.8% and 73%, respectively, for CAA.
Regarding the investigation of a regional uptake pattern specific to AD, and to avoid the issue of multiple statistical comparisons, we assessed two brain regions known to have the highest amyloid uptake in AD, namely, the PCC and frontal lobe. The results showed that these two regions have a similar diagnostic utility for WC uptake. As the precuneus is reported to show early involvement and high amyloid uptake in AD [46], we conducted a post hoc analysis, which revealed SUVR values very similar to those of the PCC and frontal lobe (pure CAA vs. pure AD: 0.87 [0.65–1.16] vs. 1.27 [1.01–1.48], respectively).
Potentially “hidden” biases are present in the samples. First, our “pure CAA” sample included patients with probable CAA with either lobar ICH or cognitive presentation, and the latter may have consisted of early AD pathology mixed with CAA [47], despite tau PET negativity. However, the results of our secondary prespecified analysis assessing these two phenotypes separately suggest that amyloid PET could achieve good accuracy for differentiating CAA from AD in patients with cognitive presentation, pointing to a potential role of amyloid PET in the memory clinic setting. Second, as already mentioned, the “pure AD” sample may have included patients with mixed AD/CAA. Third, the population studied was 100% Asian, more precisely of Chinese descent; thus far, no study has specifically investigated amyloid PET in CAA versus AD in this population. However, further studies are required to address this issue.
Our study has several limitations. First, it lacked pathological data, and our probabilistic categorization of CAA and AD was based on clinical symptoms and neuroimaging. A prospective study that includes patients with post-mortem histopathology is needed to validate our findings, although such a study would represent a major challenge. Second, our participants represented a selected population, as only patients who agreed to and successfully received amyloid and tau PET and 3D MRI were eligible, potentially affecting the generalizability of our results. Third, our sample size was relatively small. However, our study is the first to combine dual PET modalities for precise phenotyping of sporadic CAA, with a final sample size comparable to or even larger than that of most previous similar studies, yet without tau PET selection [4,9,10,12,48]. Lastly, we did not include age-matched healthy controls in the current study. However, as the primary focus was to investigate the specificity of amyloid PET for CAA versus AD rather than its already established sensitivity in detecting CAA, we believe that the lack of normal controls did not directly impact our findings.
Conclusions
The present study firmly established that higher relative occipital amyloid uptake in CAA than in AD is a robust feature. Accordingly, the occipital/WC ratio provides good specificity for differentiating CAA from AD, even among patients with cognitive dysfunction. We also underline the presence of markedly lower absolute amyloid uptake in CAA than in AD, which might also provide clinical utility over and above or on top of relative occipital uptake. Overall, although based on a highly selected population, our study suggests the potential diagnostic utility of amyloid PET for differentiating CAA from AD, which may have clinical and therapeutic implications.
Supplementary materials
Supplementary materials related to this article can be found online at https://doi.org/10.5853/jos.2024.02376.
Diagnostic performance of amyloid PET for CAA subgroups versus AD
Flowchart of patient enrolment. AD, Alzheimer’s disease; CAA, cerebral amyloid angiopathy; CMBs, cerebral microbleeds.
Comparisons of amyloid PET uptake in the frontal cortex and PCC. Box plots showing absolute amyloid burden in (A) frontal cortex, (B) PCC, and relative amyloid burden (ROI/whole cortex ratio) in (C) frontal cortex, (D) PCC. SUVR, standardized uptake value ratio; AD, Alzheimer’s disease; CAA, cerebral amyloid angiopathy; PCC, posterior cingulate cortex; PET, positron emission tomography; ROI, region of interest. *P<0.05.
Notes
Funding statement
This work was supported by grants from the Taiwan National Science and Technology Council (Tsai HH, 112-2923-B-002-001-MY3 and 113-2628-B-002-013-MY3) and the National Taiwan University Hospital (Liu CJ, 113-N0075 and 113-X0018).
Conflicts of interest
The authors have no financial conflicts of interest.
Author contribution
Study design: HHT, MP, AC, JCB. Methodology: HHT, MP, AC, JCB. Data collection: HHT, CJL, YCT, JSJ, LKT. Investigation: HHT, CJL, LKT. Statistical analysis: HHT. Writing—original draft: HHT. Writing—review & editing: MP, AC, JCB. Funding acquisition: HHT, CJL. Approval of final manuscript: all authors.
Acknowledgements
We thank the staff of the Fifth Core Lab, Department of Medical Research, National Taiwan University Hospital for their technical support during the study.