Higher cerebellum florbetapir uptake in cerebral amyloid angiopathy compared to Alzheimer's disease: A dual florbetapir and FDG PET study

Abstract

Background

Amyloid-β (Aβ) is the primary amyloidogenic protein involved in various diseases associated with cognitive dysfunction, including Alzheimer's disease (AD) and cerebral amyloid angiopathy (CAA).

Objective

This cross-sectional study aimed to investigate the characteristics of ¹⁸F-florbetapir PET, which detects Aβ deposition, and ¹⁸F-FDG PET, which measures glucose metabolism in patients with CAA and AD.

Methods

30 patients with AD, 37 with probable CAA, and 14 control subjects (CSs) underwent ¹⁸F-florbetapir and ¹⁸F-FDG PET imaging within a one-month period. Region of interest and voxel-wise analyses were performed to compare Aβ deposition and glucose metabolism patterns among the three study groups. Standardized uptake value ratios were calculated using brainstem as the reference region for ¹⁸F-florbetapir and ¹⁸F-FDG PET, respectively.

Results

Patients with CAA exhibited significantly higher ¹⁸F-florbetapir uptake in the cerebellum, global cerebral cortex, and various cortical regions compared to CSs. Compared to patients with AD, those with CAA showed predominantly higher ¹⁸F-florbetapir uptake in the cerebellum but lower uptake in the insular cortex and posterior cingulate gyrus. Glucose hypometabolism patterns in CAA did not differ significantly from those observed in AD.

Conclusions

Distinct Aβ deposition patterns, particularly the increased amyloid burden in the cerebellum, could serve as a valuable biomarker for differentiating CAA from AD.

Introduction

Cerebral amyloid angiopathy (CAA) is an age-related small vessel disease characterized by the deposition of amyloid-β (Aβ) in the walls of cortical and leptomeningeal small vessels. ¹ This condition can lead to lobar intracerebral hemorrhages (ICH), cognitive impairment, and ischemic injury. ¹ The diagnosis of CAA primarily relies on the modified Boston criteria, which have a sensitivity of approximately 80−90%.^1–3 As both Alzheimer's disease (AD) and CAA are age-related diseases, they share the common feature of Aβ deposition and exhibit similar clinical manifestations, including cognitive dysfunction.^1,3 Given the potential for these two diseases to coexist in the elderly population, the clinical differentiation between CAA and AD has garnered increasing attention. The primary distinction between CAA and AD lies in the location of Aβ deposition. In CAA, Aβ predominantly accumulates in the small cerebral vessels, whereas in AD, Aβ plaques form in the brain parenchyma. ⁴ Despite this difference, evidence suggests that the pathogenic pathways of CAA and AD intersect at several points, ⁴ highlighting the need for further research to improve differential diagnosis.

Positron emission tomography (PET) imaging using amyloid radiotracers such as ¹⁸F-florbetapir or Pittsburgh compound B (¹¹C-PiB) can visualize Aβ deposition in both AD ⁵ and CAA.^6–10 While prior studies have shown that amyloid PET has moderate-to-good diagnostic accuracy in differentiating patients with probable CAA from healthy controls or patients with deep ICH,^8,9 the distribution patterns of ¹⁸F-florbetapir uptake observed in CAA have been inconsistent.^{8,9 18}F-fluorodeoxyglucose (FDG) PET, a widely used imaging technique capable of providing metabolic information about the CNS, has been previously applied in both AD ⁵ and CAA. ¹¹ The characteristic neuroimaging profile of AD is characterized by glucose hypometabolism in the hippocampal, temporo-parietal, and posterior cingulate regions on FDG-PET, coupled with diffuse increased uptake throughout the cerebral cortex on amyloid-PET. ¹² While glucose hypometabolism has been previously observed in the posterior cortical regions of patients diagnosed with CAA, ¹¹ it remains uncertain whether this finding can effectively differentiate this condition from AD. Distinguishing between AD and CAA is still challenging, especially since previous studies have primarily relied on data from a single PET tracer. However, the unique information provided by different PET probes could potentially be leveraged to improve the differentiation between CAA and AD. This is particularly relevant given the current lack of radiotracers capable of distinguishing between Aβ deposition in the vascular wall and the brain parenchyma. ¹³

Given these considerations, we designed the current cross-sectional study to investigate the characteristics of ¹⁸F-florbetapir PET, which detects Aβ deposition, and ¹⁸F-FDG PET, which measures glucose metabolism in patients with CAA and AD. By combining the data from these two PET tracers, we sought to explore the potential of a multimodal imaging strategy in enhancing diagnostic precision for these two age-related conditions.

Methods

Participants and study design

Participants for this study were recruited from the Peking Union Medical College Hospital (PUMCH) from January 2021 to March 2024. We enrolled three distinct groups, i.e., (1) patients with CAA, (2) patients with AD, and (3) control subjects (CSs). Participants who underwent a comprehensive imaging protocol within a one-month period, which included 3.0-T magnetic resonance imaging (MRI) (T1, T2, T2-fluid-attenuated inversion recovery [T2-FLAIR], T2* or susceptibility-weighted imaging [SWI], diffusion weighted imaging [DWI], apparent diffusion coefficient [ADC]), ¹⁸F-FDG PET, and ¹⁸F-florbetapir PET were included in the study for the following screening steps. Both PET scans were conducted as part of the patients’ clinical evaluations. A flowchart detailing patient enrollment is shown in Supplemental Figure 1.

Patients with CAA, aged 55 years or older, were selected from a separate, longitudinal cerebral small vessel disease cohort study conducted at PUMCH. ¹⁴ Every participant in this group satisfied the criteria outlined in version 1.5 of the modified Boston criteria, ² qualifying them for a diagnosis of probable CAA, except for two patients between the ages of 50 and 54 satisfying the diagnostic criteria for CAA-related inflammation (CAA-ri)^15,16 who were also included into the study. The Boston criteria version 1.5 ² for probable CAA is provided here: clinical data and MRI or CT demonstrating: multiple hemorrhages restricted to lobar, cortical, or corticosubcortical regions (cerebellar hemorrhage allowed) or single lobar, cortical, or corticosubcortical hemorrhage and focal or disseminated superficial siderosis; age ≥ 55y; absence of other cause of hemorrhage or superficial siderosis. Specifically, for CAA diagnosis, we utilized T2* or SWI sequences of MRI to identify lobar hemorrhages, cerebral microbleeds, and cortical superficial siderosis. Additionally, T2-weighted or T2-FLAIR sequences were used to assess white matter hyperintensity, which are relevant for diagnosing CAA-ri. Of the 105 registered CAA patients, those who had undergone both PET-CT and MRI in one month were ultimately included (n = 37).

Patients with AD were recruited from a registry study of cognitive and movement disorder conducted at PUMCH. They had completed an extensive battery of cognitive assessments and met the 2011 National Institute on Aging and Alzheimer's Association (NIA-AA) core clinical criteria for probable AD, ¹⁷ or the 2018 NIA-AA Research Framework criteria. ¹⁸ A total of 46 out of 261 registered AD patients who had completed both PET-CT and MRI in one month were included. Among them, patients probably with other diseases including frontotemporal dementia (FTD), parkinsonian syndrome (PDS), progressive supranuclear palsy (PSP), corticobasal syndrome (CBS), idiopathic normal pressure hydrocephalus (iNPH), rapid eye movement sleep behavior disorder (RBD), epilepsy, syphilis, and stroke causing large areas of lesions were excluded (n = 10). Patients with MRI results suggesting the presence of any strict lobar cerebral microbleed (CMB), convexal subarachnoid hemorrhage (cSAH), cortical superficial siderosis (cSS), macro-hemorrhage were also excluded from the AD group (n = 6). A total of 30 AD patients were enrolled ultimately.

The CSs group comprised individuals from the neurology clinic who reported subjective cognitive complaints. They underwent a comprehensive suite of neuropsychological assessments, which did not support objective cognitive impairment. A total of 25 subjects who underwent dual PET-CT and head MRI within one month were included. Among them, subjects probably with diseases including AD, FTD, PDS, PSP, CBS, iNPH, RBD, epilepsy, syphilis, and stroke causing large areas of lesions were excluded (n = 10). Subjects with MRI findings indicating strict lobar CMB, cSS, ICH or any other abnormal lesion were also excluded from CSs group (n = 1). A total of 14 CSs were included finally.

Data collection

The following variables were collected from all participants: age, sex, presence of hypertension, presence of diabetes mellitus, smoking status, Mini-Mental State Examination (MMSE) scores, Montreal Cognitive Assessment (MoCA) scores, apolipoprotein E (APOE) genotype, as well as MRI and PET imaging data. Subjects with incomplete MRI or PET scans, or those with missing data, were excluded from the study.

Image acquisition

The ¹⁸F-florbetapir and ¹⁸F-FDG tracers were synthesized at the Department of Nuclear Medicine, PUMCH, using a PET-MF-IV-I ¹⁸F-multifunction synthesizer. The radiochemical purity of both tracers was maintained at over 95%. For ¹⁸F-florbetapir PET imaging, a scan was acquired over 60 min following an intravenous injection of the tracer (350−400 MBq). In the case of ¹⁸F-FDG PET imaging, a scan was acquired 45 to 55 min post-injection, with a dose of 3.7 MBq/kg. All patients were required to fast before undergoing FDG PET/CT scanning. After tracer administration, patients were instructed to rest in a quiet, dimly lit environment to minimize external stimuli and reduce physiological brain activity. Subsequently, PET/CT imaging was performed using a PoleStar m680 scanner (SinoUnion Healthcare, Beijing, China). A low-dose CT scan was initially acquired for attenuation correction, followed by a 20-min brain PET scan. The PET data were reconstructed using a 3-D ordered-subset expectation maximization (OSEM) algorithm with time-of-flight (TOF) information, employing 4 iterations, 10 subsets, a Gaussian filter with a full-width at half-maximum (FWHM) of 2.5 mm, and a zoom factor of 1. The reconstructed images had a matrix size of 512 × 512 and a voxel size of 1.18 × 1.18 × 1.87 mm³.

Visual analysis of ¹⁸F-florbetapir PET images

The ¹⁸F-florbetapir PET images were visually assessed by three independent observers, all of whom are experienced specialists in molecular imaging. To ensure an unbiased evaluation, the assessors were blinded to all individual information.

PET image processing

All ¹⁸F-florbetapir and ¹⁸F-FDG PET images underwent spatial normalization to the Montreal Neurological Institute (MNI) space using a method based on an adaptive probabilistic brain atlas. This MRI-free algorithm enables the spatial normalization of PET brain images acquired with various tracers. ¹⁹ Following normalization, the images were smoothed using an 8-mm FWHM Gaussian kernel.

For quantitative analysis, standardized uptake value ratios (SUVr) for both ¹⁸F-florbetapir and ¹⁸F-FDG PET images were calculated using the brainstem as the reference region. The brainstem region of interest was defined according to the Neuromorphometrics Atlas (Neuromorphometrics, Inc., http://Neuromorphometrics.com/), and its mean uptake value served as the denominator for SUVr calculations. The SUVr calculations were performed using an in-house script ¹⁹ developed by the research team that integrated in the SNBPI (https://github.com/IHEP-Brain-Imaging/Spatial-Normalization-of-Brain-PET-Images).

Voxel-wise analysis

In the voxel-wise analysis, after calculating the SUVr at each voxel, we first performed a one-way ANOVA across the three groups of CAA, AD, and CSs, generating an F-statistic map highlighting regions of significant group difference using Statistical Parametric Mapping (SPM) 12 (https://www.fil.ion.ucl.ac.uk/spm/). Gaussian Random Field (GRF) theory correction ²⁰ was then applied using DPABI software (https://rfmri.org/DPABI), with the following parameters: voxel-level p < 0.001, cluster-level p < 0.05. This step identified spatially significant clusters of group differences, which were saved as a binary mask. Subsequently, post-hoc pairwise comparisons were conducted exclusively within the mask derived from the previous step. For each pairwise comparison, a T-statistic map was generated using SPM12. GRF correction was again applied using DPABI with adjusted thresholds: voxel-level p < 0.001/3 (equivalent to a Bonferroni-corrected threshold of p < 0.001 for three comparisons), maintaining the cluster-level threshold of p < 0.05. In all of the one-way ANOVA and post-hoc pairwise comparisons，the age, sex, and APOE genotype (the carriage of APOE ε2 and the carriage of APOE ε4) were included as covariates to control for potential confounding effects. The resulting statistical maps from the pairwise comparisons were visualized using xjview version 19.0 program (https://www.alivelearn.net/xjview).

ROI-based analysis

In the region of interest (ROI)-based analysis, the WFU_PickAtlas ²¹ was employed in the MNI space for the extraction of specific regions. The ROIs included the frontal lobe cortex, parietal lobe cortex, occipital lobe cortex, temporal lobe cortex, insular lobe cortex, posterior cingulate gyrus (PC), and cerebellar cortex. A global cerebral cortex was generated by integrating frontal cortex, parietal cortex, occipital cortex, temporal cortex and insular cortex into a single mask. The SUVr values of the ROIs were calculated and compared across the three groups to analyze for differences. The occipital-to-posterior cingulate cortex (O/PC) SUVr ratio was also calculated and compared.

Statistical analysis

Continuous variables with a normal distribution were summarized using means and standard deviations (SDs). For skewed distributions, medians and interquartile ranges (IQRs) were reported. Categorical variables were expressed as frequencies (n) and percentages (%). The χ² test was employed for categorical variables. For comparisons of continuous data across the three groups, the one-way ANOVA test was applied for normally distributed data, and post-hoc pairwise comparisons were performed to identify which specific group differed significantly, using the Bonferroni method when variances were homogeneity and applying the Tamhane's T2 method when variances were heterogeneous. For continuous variables that were not normally distributed, intergroup comparisons were performed using the Kruskal-Wallis H test. If a significant difference was found, post hoc pairwise comparisons were conducted using Dunn's pairwise comparisons with Bonferroni correction for the p-values. Analyses were performed using the commercial software package SPSS, version 26.0 (IBM, Armonk, NY, USA). All tests were two-tailed, and a p value < 0.05 was considered statistically significant. Results were visualized using GraphPad Prism version 10 (https://www.graphpad.com).

Results

Participant characteristics

The general characteristics of the three study groups are summarized (Table 1). The CAA group had a significantly higher proportion of men compared to both the AD (70.3% versus 36.7%, p < 0.05) and CSs groups (70.3% versus 14.3%, p < 0.05). Similarly, the proportion of smokers was higher in the CAA group compared to the AD group (40.5% versus 6.7%, p < 0.05) and the CS group (40.5% versus 0%, p < 0.05). As expected, cognitive test scores (MMSE and MoCA) were significantly lower in both the CAA and AD groups compared to CSs. The CAA group had a higher proportion of APOE ε2 carriers (33.3%), whereas the APOE ε4 carrier status was highest in patients with AD (75.0%). In the CAA group, nine patients (24.3%) had symptomatic non-traumatic lobar hemorrhage, while the remaining 28 patients (75.7%) were identified as having non-ICH presentation, including six with CAA-ri and 22 with cognitive impairment or dementia. Furthermore, a comparison of baseline characteristics between the 37 included and 68 not-included CAA patients revealed no significant differences in sex, hypertension, diabetes, smoking, MMSE, or APOE genotype. However, we found that the included CAA patients were significantly younger (68 [64, 73]) than those not included (74 [68, 79.5]) (p < 0.008), which may be attributed to better medical compliance among younger CAA patients.

Table 1.

General characteristics of the three study groups.

	CAA, n = 37	AD, n = 30	CSs, n = 14	p#
Age, y median [IQR]	68 (64, 74.5)	70 (64.3, 74)	67 (62.3, 71.3)	0.221
Male sex n (%)	26 (70.3%)	11 (36.7%)a	2 (14.3%)a	<0.001
Hypertension n (%)	21 (56.8%)	13 (43.3%)	6 (42.9%)	0.476
Diabetes mellitus n (%)	9 (24.3%)	6 (20.0%)	3 (21.4%)	0.911
Smoking n (%)	15 (40.5%)	2 (6.7%)a	0a	<0.001
MMSE median [IQR]	21 (14, 26)b	21 (15, 26)b	29 (28, 29.3)	<0.001
MoCA median [IQR]	16 (7.5, 20.0)b	17 (12, 19.5)b	25.5 (23, 27.5)	<0.001
APOE ε2 carriage n (%)	12/36 (33.3%)	2/28 (7.1%)a	0a	0.005
APOE ε4 carriage n (%)	18/36 (50%)c	21/28 (75.0%)	4/13 (30.8%)c	0.018

Data are presented as medians with interquartile ranges for continuous data and as frequencies (n) with percentages (%) for categorical variables. Differences between the three study groups were assessed using one-way analysis of variance for continuous data and χ² test for categorical variables.

CAA: cerebral amyloid angiopathy; AD: Alzheimer's disease; CSs: control subjects; MMSE: Mini-Mental State Examination; MoCA: Montreal cognitive assessment; APOE: apolipoprotein E.

^# p value from one-way ANOVA (for normal distribution) or non-parametric test (for skewed distribution) across the three groups

^a Significant difference compared to patients with CAA (p < 0.05)

^b Significant difference compared to CS group (p < 0.001)

^c Significant difference compared to patients with AD (p < 0.05)

¹⁸F-florbetapir PET visual analysis

Among patients with CAA (n = 37), one demonstrated a negative ¹⁸F-florbetapir PET scan, while the remaining 36 exhibited positive results. All patients diagnosed with AD (n = 30) showed positive ¹⁸F-florbetapir PET findings, whereas all CSs (n = 14) tested negative.

Region of interest analysis of ¹⁸F-florbetapir SUVr values in the three study groups

The ROI analysis was conducted to examine the difference in ¹⁸F-florbetapir uptake among the three study groups across the global cerebral cortex, cerebellar cortex, and temporal, frontal, occipital, parietal, insular cortices and posterior cingulate gyrus. We compared the SUVr for these regions across the three groups (Figure 1, Supplemental Table 1). The global cerebral cortical SUVr was significantly higher in patients with CAA (1.00 ± 0.15) compared to CSs (0.88 ± 0.06, p < 0.001), with no significant difference between patients with CAA and those with AD. Furthermore, ROI analysis revealed significantly higher cerebellar SUVr in patients with CAA (0.87 ± 0.10) compared to those with AD (0.79 ± 0.07, p < 0.001) and the CS group (0.75 ± 0.06, p < 0.001). The occipital cortex SUVr was notably higher in the CAA group (1.05 ± 0.17), exceeding that of the CS group (0.88 ± 0.08, p < 0.001). Additionally, ¹⁸F-florbetapir uptake in the parietal (1.03 ± 0.16 versus 0.87 ± 0.06, p < 0.01), frontal (0.99 ± 0.15 versus 0.88 ± 0.06, p < 0.01), and temporal cortices (0.97 ± 0.15 versus 0.87 ± 0.06, p < 0.01) was significantly higher in patients with CAA compared to the CS group, with no significant differences compared to patients with AD. The insular cortex (0.92 [IQR: 0.83, 1.03] versus 1.02 ± 0.08, p < 0.01) and posterior cingulate gyrus (1.09 [IQR: 0.96, 1.17] versus 1.15 ± 0.08, p < 0.05) SUVr were both significantly lower in patients with CAA compared to those with AD, but neither showed significant differences when compared to the CS group. The ratio of O/PC SUVr was significantly higher in patients with CAA compared to the AD (0.96 ± 0.10 versus 0.88 ± 0.10, p < 0.001) and CS group (0.96 ± 0.10 versus 0.85 ± 0.05, p < 0.001).

Figure 1.

Comparative analysis of ¹⁸F-florbetapir standardized uptake value ratios in the three study groups. (A) Box plot displaying greater global cerebral cortex flobetapir SUVr in patients with CAA than CSs (p < 0.001), in patients with AD than CSs (p < 0.001), and no significant difference between CAA and AD. (B) Elevated cerebellar cortex flobetapir SUVr in patients with CAA relative to both CSs (p < 0.001) and patients with AD (p < 0.001). (C)-(F) Greater ¹⁸F-florbetapir SUVr in patients with CAA compared to CSs in (C) the occipital cortex (p < 0.001), (D) temporal cortex (p < 0.01), (E) frontal cortex (p < 0.01), (F) parietal cortex (p < 0.01). Greater ¹⁸F-florbetapir SUVr in patients with AD compared to CSs in these four regional cortices (p < 0.001). No significant differences were observed between CAA and AD in these regional cortices. (G) Lower SUVr of the insular cortex in CAA compared to AD (p < 0.01). Higher SUVr of that in AD relative to CSs (p < 0.001). No significant difference between CAA and CSs. (H) Lower posterior cingulate gyrus SUVr in patients with CAA compared to AD (p < 0.05). Higher SUVr of that in patients with AD than CSs (p < 0.01). (I) Higher O/PC SUVr of that in CAA compared to AD (p < 0.001) and CSs (p < 0.001). Statistical significance is denoted as follows: *p < 0.05, **p < 0.01, and ***p < 0.001. The upper and lower whiskers of the box represent the maximum and minimum values, respectively. Each point represents a sample. CAA: cerebral amyloid angiopathy; AD: Alzheimer's disease; CS: control subject; SUVr: standardized uptake value ratio; PC: posterior cingulate cortex; O/PC: Occipital cortex/Posterior cingulate cortex.

In patients with AD, ¹⁸F-florbetapir SUVr in global and regional cerebral cortices was significantly higher than that in the CS group, with the exception of the cerebellar cortex and ratio of O/PC.

Within the CAA group, no significant difference of ¹⁸F-florbetapir SUVr in the aforementioned ROIs was found between patients with and without symptomatic non-traumatic lobar hemorrhage.

Voxel-wise analysis of ¹⁸F-florbetapir uptake in the three study groups

Voxel-wise analysis of ¹⁸F-florbetapir uptake revealed diffusely greater uptake in the global cerebral cortex (primarily frontal cortex, parietal cortex, temporal cortex, and occipital cortex) and cerebellum of patients with CAA compared to CSs (Figure 2). When comparing AD and CSs, patients with AD exhibited significantly higher retention in the global cerebral cortex diffusely. Furthermore, voxel-wise analysis demonstrated that patients with CAA had greater ¹⁸F-florbetapir uptake compared to those with AD, primarily in the cerebellum.

Figure 2.

Voxel-wise analysis comparing ¹⁸F-florbetapir uptake in three study groups using Gaussian random field (GRF) theory correction of voxel-level p < 0.001, cluster-level p < 0.05. (A) Patients with CAA exhibited diffusely greater ¹⁸F-florbetapir uptake in the global cerebral cortex (primarily frontal cortex, parietal cortex, temporal cortex, and occipital cortex) and cerebellum compared to CSs (labelled in orange, white arrows for some regions). (B) Patients with AD demonstrated diffusely higher ¹⁸F-florbetapir uptake in the global cerebral cortex compared to CSs (labelled in yellow). (C) Patients with CAA exhibited higher retention of ¹⁸F-florbetapir primarily in the cerebellum (labelled in orange, white arrows), and lower uptake of that in the white matter compared to patients with AD (labelled in gray). All the color-coded regions in the figure depict areas of statistically significant differences in radiotracer uptake between groups. The age, sex, and APOE genotype (the carriage of APOE ε2 and ε4) were included as covariates to control for potential confounding effects, for 36 CAA, 28 AD, and 13 CS with APOE genotype available. The color bar in the figure reflects the parameters of t-values. The color-coded regions depict areas of statistically significant differences in radiotracer uptake between groups using Gaussian Random Field (GRF) theory correction of voxel-level p < 0.001, cluster-level p < 0.05. CAA: cerebral amyloid angiopathy; AD: Alzheimer’s disease; CS: control subject.

Region of interest analysis of ¹⁸F-FDG SUVr values in the three study groups

We subsequently applied ROI analysis to reveal the status of ¹⁸F-FDG metabolism in the three study groups. We compared the ¹⁸F-FDG SUVr in specific regions across the three groups (Figure 3, Supplemental Table 2). The ¹⁸F-FDG SUVr values were significantly lower in patients with CAA compared to CSs in the global cerebral cortex (1.29 ± 0.13 versus 1.41 ± 0.08, p < 0.001), temporal cortex (1.20 ± 0.14 versus 1.33 ± 0.08, p < 0.01), parietal cortex (1.35 ± 0.17 versus 1.51 ± 0.09, p < 0.01), occipital cortex (1.40 ± 0.20 versus 1.60 ± 0.14, p < 0.01), frontal cortex (1.27 ± 0.12 versus 1.35 ± 0.07, p < 0.05), and insular cortex (1.34 ± 0.13 versus 1.46 ± 0.10, p < 0.01). No significant difference of the FDG SUVr values was found in the global or regional cerebral cortex and cerebellum between CAA and AD. The SUVr of the posterior cingulate gyrus was significantly lower in patients with CAA compared to CSs (1.58 ± 0.24 versus 1.80 ± 0.17, p < 0.01), but similar to that observed in patients with AD. No significant difference of O/PC value was identified across the three groups. In patients with AD, FDG SUVr in these regional cerebral cortices was significantly lower than that in CSs.

Figure 3.

Comparative analysis of ¹⁸F-FDG standardized uptake value ratios in the three study groups. (A) Box plot displaying lower global cerebral cortex FDG SUVr in patients with CAA than CSs (p < 0.001), in patients with AD than CSs (p < 0.001), and no significant difference between CAA and AD. (B) The box plots demonstrate no statistical difference of cerebellar cortex FDG SUVr across the three groups. Lower FDG SUVr in patients with CAA compared to CSs in (C) occipital cortex (p < 0.01), (D) the temporal cortex (p < 0.01), (E) frontal cortex (p < 0.05), (F) parietal cortex (p < 0.01), (G) insular cortex (p < 0.01), (H) posterior cingulate cortex (p < 0.01). Lower FDG SUVr in patients with AD compared to CSs in (C) occipital cortex (p < 0.05), (D) temporal cortex (p < 0.01), (E) frontal cortex (p < 0.05), (F) parietal cortex (p < 0.01), (G) insular cortex (p < 0.05), and (H) posterior cingulate cortex (p < 0.01). (I) No significant difference of O/PC across the three groups. Statistical significance is denoted as follows: *p < 0.05, **p < 0.01, ***p < 0.001. The upper and lower whiskers of the box represent the maximum and minimum values, respectively. Each point represents a sample. FDG: fluorodeoxyglucose; CAA: cerebral amyloid angiopathy; AD: Alzheimer’s disease; CS: control subject; FDG: fluorodeoxyglucose; SUVr: standardized uptake value ratio; PC: posterior cingulate cortex; O/PC: occipital cortex/posterior cingulate cortex.

Within the CAA group, no significant difference of ¹⁸F-FDG SUVr of the specified ROIs was found between patients with and without symptomatic non-traumatic lobar hemorrhage.

Voxel-wise analysis of ¹⁸F-FDG uptake in the three study groups

Voxel-wise analysis of ¹⁸F-FDG uptake revealed that patients with CAA had glucose hypometabolism precuneus, inferior parietal lobule, angular gyrus, middle occipital gyrus, superior temporal gyrus, middle temporal gyrus, inferior temporal gyrus, and calcarine cortex compared to CSs, while the cerebellum exhibited normal metabolic activity (Figure 4). In contrast, patients with AD showed reduced metabolism in precuneus, inferior parietal lobule, superior frontal gyrus, middle frontal gyrus, inferior frontal gyrus, middle cingulate gyrus, angular gyrus, posterior cingulate gyrus, superior temporal gyrus, middle temporal gyrus, inferior temporal gyrus, hippocampus, para hippocampus, and temporal pole compared to CSs. No significant differences in glucose metabolism were observed between patients with CAA and those with AD.

Figure 4.

Voxel-wise analysis comparing ¹⁸F-FDG uptake in three study groups using Gaussian random field (GRF) theory correction of voxel-level p < 0.001, cluster-level p < 0.05. (A) Patients with CAA exhibited reduced glucose metabolism in precuneus, inferior parietal lobule, angular gyrus, middle occipital gyrus, superior temporal gyrus, middle temporal gyrus, inferior temporal gyrus, and calcarine cortex compared to CSs (labelled in green, white arrows for typical regions). (B) Patients with AD demonstrated significant glucose hypometabolism primarily in precuneus, inferior parietal lobule, superior frontal gyrus, middle frontal gyrus, inferior frontal gyrus, middle cingulate gyrus, angular gyrus, posterior cingulate gyrus, superior temporal gyrus, middle temporal gyrus, inferior temporal gyrus, hippocampus, para hippocampus, and temporal pole compared to CSs (labelled in blue, white arrows for typical regions). (C) No significant difference of FDG uptake was found between CAA and AD. The age, sex, and APOE genotype (the carriage of APOE ε2 and ε4) were included as covariates to control for potential confounding effects, for 36 CAA, 28 AD, and 13 CS with APOE genotype available. The color bar in the figure reflects the parameters of t-values. The color-coded regions depict areas of statistically significant differences in radiotracer uptake between groups using Gaussian Random Field (GRF) theory correction of voxel-level p < 0.001, cluster-level p < 0.05. FDG: fluorodeoxyglucose; CAA: cerebral amyloid angiopathy; AD: Alzheimer's disease; cs: control subject.

Discussion

To our knowledge, this study represents the first investigation in which patients with CAA and AD concurrently underwent both ¹⁸F-florbetapir and ¹⁸F-FDG PET imaging, enabling a simultaneous evaluation of Aβ deposition and cerebral glucose metabolism. Our research yielded the primary finding that patients with CAA exhibited significantly higher ¹⁸F-florbetapir uptake in the cerebellum, global cerebral cortex, and various cortical regions compared to CSs. Comparing to AD, patients with CAA had greater ¹⁸F-florbetapir uptake predominantly in the cerebellum and lower uptake in the posterior cingulate gyrus and insular cortex. Collectively, these findings suggest that the distinct patterns of Aβ deposition, particularly the increased amyloid burden in the cerebellum, could serve as a valuable biomarker for differentiating CAA from AD.

In our study, both voxel-wise and ROI-wise analyses indicated that ¹⁸F-florbetapir deposition in the cerebellar cortex was highest in patients with CAA compared to those with AD and CSs. According to the Boston criteria for CAA, hemorrhagic lesions in the cerebellum are classified as neither lobar nor deep hemorrhagic lesions. ¹ The involvement of the cerebellum in CAA has been overlooked in the past. While Johnson et al. ²² and Baron et al. ²³ found no significant differences in cerebellar ¹¹C-PiB uptake between patients with CAA and CSs, most amyloid PET studies on CAA have used the cerebellar cortex as the reference region to calculate global or regional relative radiotracer uptake.^8,9,24 Interestingly, when using the pons as the reference region, Tsai et al. ²⁵ reported significantly higher global cerebellar ¹¹C-PiB retention in CAA-related ICH compared to non-CAA-ICH. The potential involvement of the cerebellum in CAA is further substantiated by an autopsy study, which revealed a significant prevalence of concurrent cerebellar and occipital lobe CAA pathology. ²⁶ Furthermore, cerebellar hemorrhage is a relatively common occurrence in patients with CAA in clinical practice. ²⁷ Pathological studies also show that cerebellar Aβ deposition in AD is significantly lower than in the medial temporal, prefrontal and parietal cortices, and in early neuropathological investigations, Aβ plaques and neurofibrillary changes are typically undetected in the cerebellum of AD patients. ²⁸ We hypothesize that this divergence stems from distinct Aβ species and their clearance pathways. In CAA, the vascular Aβ is predominantly the more soluble Aβ_40, ²⁹ which may be drained via the bloodstream or perivascular space to deposit in distal vascular walls, including those of the cerebellum. Conversely, parenchymal plaques in AD are primarily composed of the less soluble Aβ₄₂, which is less prone to distant drainage. ²⁹ Confirming this hypothesis will require further correlative pathological and mechanistic studies. On the other hand, these findings suggest that using the cerebellum as the reference region in quantitative amyloid PET imaging may lead to less accurate results in patients with CAA, as the cerebellum itself can be affected by amyloid deposition. ²⁵ To address this issue and provide a more precise assessment of amyloid burden in individuals with CAA, we chose to use the brainstem as the reference region in our study. By selecting a reference region that is less likely to be directly impacted by the pathology of interest, our approach aims to enhance the reliability and validity of the quantitative results obtained from amyloid PET scans in this patient population.

Regarding cortical ¹⁸F-florbetapir results, and as expected due to the widespread Aβ deposition in the cortical vessels occurring in CAA, ³⁰ we found that these patients had markedly higher global and regional SUVr compared to CSs. This observation is largely consistent with the findings of Raposo et al., ⁸ who reported that patients with CAA-ICH exhibited higher ¹⁸F-florbetapir SUVr in the frontal, insular, temporal, parietal, and occipital cortices. Notably, the prominent ¹⁸F-florbetapir retention in the occipital cortex appears to be a distinctive feature in CAA and is in line with previous PET^8,22 and pathology studies.^31,32 No significant difference of ¹⁸F-florbetapir SUVr uptake in regional cerebral cortex was found between CAA and AD patients, except for the insular cortex and posterior cingulate gyrus, where the ¹⁸F-florbetapir uptake was lower in CAA. Tsai et al. demonstrated higher relative occipital ¹¹C-PiB uptake and lower posterior cingulate uptake in patients with CAA compared to AD, in a study utilizing Tau-PET to select “pure CAA” cases. ³³ Planton et al. ³⁴ found that patients with mild cognitive impairment due to AD had higher global ¹⁸F-florbetapir SUVr than those with CAA-ICH, with no differences in regional SUVr in ROI-based analysis. Johnson et al. ²² and Ly et al. ²⁴ both demonstrated that global ¹¹C-PiB uptake was increased in patients with CAA compared to CSs but less than that in patients with AD. The inconsistencies might be attributed, at least in part, to the distinct subtypes of CAA and the choice of reference regions. In our study, over 50% of patients with CAA exhibited cognitive impairment or dementia in the absence of macro hemorrhagic lesions. In contrast, Ly et al. ²⁴ and Planton et al. ³⁴ focused exclusively on patients with CAA-ICH, and their quantitative results may have been influenced by the presence of hemorrhagic lesions. While Johnson et al. ²² investigated amyloid PET in CAA not limited to CAA-ICH subtype, but they had a relatively small sample size. It is also important to consider the use of ¹⁸F-florbetapir instead of ¹¹C-PiB as a potential contributing factor to the observed differences. Furthermore, most CAA patients without lobar ICH are typically evaluated due to symptoms of cognitive decline, making it challenging to exclude a concomitant diagnosis of AD. This overlap could influence the current findings. To address these inconsistencies and to better understand the role of amyloid PET imaging in differentiating CAA from AD, future studies should aim to include larger, well-characterized cohorts of patients with various CAA subtypes and employ standardized methodologies for quantitative analysis.

In a voxel-wise analysis of ¹⁸F-FDG PET findings, we observed that CAA is characterized by hypometabolism in the parietal, occipital, lateral temporal cortex, and posterior cingulate gyrus, while cerebellar metabolism remains preserved. These findings agree with previously published literature. ¹¹ Notably, Bergeret et al. ¹¹ reported that significant FDG metabolic reduction in CAA predominantly affects posterior cortical regions, including the superior and inferior parietal, lateral temporal, and posterior cingulate regions, with the cerebellum and medial temporal cortex remaining unaffected. In the current study, we found that patients with AD exhibited glucose hypometabolism in the medial temporal cortex in addition to the parietal, lateral temporal cortex, and posterior cingulate regions. Even though neither voxel-wise nor ROI analyses revealed statistically significant differences in the metabolic patterns between CAA and AD, we observed that CAA patients were less likely to exhibit glucose metabolism involvement in the medial temporal cortex, providing some evidence to differentiate CAA from AD in clinical practice.

Moreover, we noted that the patterns of amyloid deposition and glucose hypometabolism were not entirely concordant in patients with CAA, particularly in the cerebellum. This discrepancy warrants further research to elucidate the mechanisms and impact of Aβ deposition in the vascular wall on regional brain metabolism. In clinical practice, CAA patients may present with motor discordance (such as postural imbalance and gait abnormalities, etc.), for which cerebellar pathology may play a partial role, although it is certainly not the only contributor.

In a nationwide multi-center CAA study in China, the prevalence of APOE ε2 carriage was 25%. ¹⁴ In another intracerebral hemorrhage study, the proportion of APOE ε2 carriage among patients with lobar hemorrhage was 20.3%. ³⁵ The slightly higher proportion in our CAA group may be attributed to the relatively small sample size. Regarding the influence of APOE ε2 on cortical Aβ deposition, a systematic review indicates that APOE ε4 carriers exhibit increased Aβ plaque deposition and an earlier onset of amyloid pathology compared to non-carriers, whereas APOE ε2 carriers demonstrate delayed deposition and less severe Aβ pathology. ³⁶ Evidence suggests that both APOE ε2 and APOE ε4 alleles exert certain effects on Aβ deposition. Based on this, in our voxel-wise analysis, we adjusted for the APOE genotype.

While this study presents promising findings, it is important to acknowledge its limitations. First, despite statistical adjustment for sex, residual confounding due to the significant sex disparity among the three groups and the limited sample size may persist. A study found that men with CAA seem to have an earlier onset and more hemorrhagic disease course compared with women. ³⁷ Our study included a higher proportion of male CAA patients, which might lead to an overall higher level of Aβ deposition. Besides, the female predominance in the control group increases the likelihood of including individuals with preclinical AD, as in China, the prevalence of AD and related dementia is higher in women than in men.^38,39

Second, we have to acknowledge that some CAA patients may have comorbid AD, particularly those presenting with cognitive impairment, as CAA and AD are highly likely to co-occur. Since most patients did not undergo cerebrospinal fluid biomarker testing or Tau-PET imaging, we were unable to accurately identify which CAA patients had concurrent AD. The presence of comorbid AD in CAA patients could potentially confound the imaging results and their interpretation. On the other hand, to minimize the confounding effect of co-existing CAA, we excluded AD patients who exhibited strictly lobar hemorrhagic lesions. While necessary, this deliberate exclusion may limit the representativeness of our AD cohort and introduce a potential selection bias. Besides, amyloid-PET was used for the diagnosis of some AD patients according to 2018 NIA-AA Research Framework criteria, which may probably lead to the selection of AD patients with a higher degree of cortical Aβ deposition.

Third, we did not specifically examine the impact of hemorrhagic lesions, which can result in areas of encephalomalacia that fail to retain PET tracers in some CAA patients, due to the relatively small proportion of CAA-ICH cases within the CAA group. However, lobar hemorrhage may make influence on the amyloid deposition and cause secondary hypometabolism, and thus affect image analysis. Furthermore, in an MRI-free pipeline, we were unable to perform partial volume effect (PVE) correction. We must acknowledge that varying degrees of brain atrophy likely exist between our CAA and AD patient groups, which may influence the analytical results. Besides, the presence of cortical and subcortical edema in CAA-ri patients may potentially affect tracer uptake. However, we believe this impact is limited, as our PET scans were typically performed when the patients’ condition had stabilized to a less acute stage.

Additionally, we noted that a significantly higher proportion of CAA subjects were smokers. This is a potential limitation, as cigarette smoke components may cause greater amyloid aggregation and lower cerebral glucose metabolism.^40,41 A large cross-sectional PET study from the Alzheimer's Disease Neuroimaging Initiative (ADNI) on cognitively normal elders reported greater florbetapir retention in smokers than in never-smokers within regions including the cingulate, temporal, parietal cortices and composite gray matter. ⁴⁰ Furthermore, a study by Timothy et al. found an interaction between smoking and APOE ε4 carrier status, where smoker APOE4 + showed lower composite glucose metabolism than never-smoking APOE4−, never-smoking APOE4+, and smoking APOE4−. ⁴¹ In addition, there was a 43–57% incidence of hypertension in all groups, which may impact the results. A PiB-PET and FDG-PET study demonstrated that in cognitively normal, late-middle-aged to older adult, systolic blood pressure and pulse pressure were positively correlated with cerebral amyloid in frontal, temporal and posterior-cingulate precuneus regions, and inversely correlated with cerebral glucose metabolism in frontal and temporal brain regions. ⁴² Although higher blood pressure may be associated with the amyloid deposition and glucose metabolism, the prevalence of hypertension showed no significant differences across our study groups.

Finally, the relatively small sample size may reduce the representativeness of CAA and AD, causing selection bias, and preclude the investigation of whether different subtypes of CAA could influence the value of ¹⁸F-florbetapir PET in differentiating this condition from AD.

Accurate differentiation between AD and CAA is essential in the era of anti-Aβ monoclonal antibodies because CAA patients are disproportionately prone to amyloid-related imaging abnormalities (ARIA-E/ARIA-H) when exposed to these therapies.^43,44 Consequently, incorporating MRI and PET together to identify or exclude concomitant CAA is critical for the participant selection and safety security. According to our study results, for example, if cerebellar Aβ deposition is detected, we must be alert to the possibility of comorbid CAA and exercise increased caution when administering anti-Aβ targeted therapies.

This study demonstrates that patients with CAA exhibit significantly higher ¹⁸F-florbetapir uptake in the cerebellum, global cerebral cortex and various cortical regions compared to CSs. Compared to patients with AD, florbetapir uptake was higher in the cerebellum and lower in the insular cortex and posterior cingulate gyrus in patients with CAA. However, the patterns of glucose hypometabolism in CAA did not differ significantly from those observed in AD, limiting the clinical utility of ¹⁸F-FDG PET imaging for distinguishing between the two disorders. These findings suggest that the distinct patterns of Aβ deposition, particularly the increased amyloid burden in the cerebellum, could serve as a valuable biomarker for differentiating CAA and AD, warranting further investigation in larger, well-characterized cohorts of patients with various CAA subtypes.