Cerebellar cortical amyloid deposition biases ¹⁸F-Florbetapir PET quantification with whole cerebellum as reference region

Abstract

Objective

To investigate whether cerebellar cortical amyloid-beta (Aβ) deposition biases ¹⁸F-Florbetapir Aβ PET quantification when the whole cerebellum (WC) is used as the reference region, and to identify a stable reference region for quantification. Additionally, to establish an optimal diagnostic threshold for the detection of cerebellar cortical Aβ deposition.

Methods

This retrospective cohort study enrolled 359 Aβ-positive participants including 52 with and 307 without cerebellar cortical Aβ deposition. All participants underwent ¹⁸F-Florbetapir Aβ PET imaging, structural MRI, and cerebrospinal fluid biomarker analysis. Standardized uptake value ratio (SUVR) was derived from the standardized uptake value (SUV) by normalizing to WC, pons, and white matter (WM). Centiloid (CL) values were derived following standardized procedures. Bootstrap resampling was employed for group comparisons. Receiver operating characteristic (ROC) analysis was performed to determine the optimal SUVR threshold for identifying cerebellar cortical Aβ deposition, with validation conducted in an internal validation subset (n = 30). Two patients treated with lecanemab underwent longitudinal Aβ PET follow-up.

Results

Individuals classified as cerebellar cortical Aβ-positive exhibited significantly lower CL values than Aβ‑negative individuals (bootstrap P < 0.001). The cerebellar cortical Aβ‑positive group showed elevated SUV_WC (bootstrap P = 0.012) and reduced SUV_cortex (bootstrap P = 0.008), whereas SUV_pons showed no significant differences between groups (bootstrap P = 0.110). SUV_WM exhibited a significant intergroup difference (bootstrap P < 0.001), while SUVR_WM showed no significant between-group difference (bootstrap P = 0.288). An optimal cerebellar SUVR_WM threshold of 0.428 yielded excellent diagnostic performance for identifying cerebellar cortical Aβ deposition, with an area under the curve (AUC) = 0.931, sensitivity 86.5%, specificity 86.3%. Substantial agreement between quantitative and visual assessment was confirmed in the validation subset (Cohen’s kappa = 0.856). In the two cerebellar cortical Aβ-positive lecanemab-treated patients, CL values decreased over time, SUVR_WM showed modest alterations, while cerebellar SUVR_WM remained above the 0.428 threshold throughout follow-up.

Conclusion

Cerebellar cortical Aβ deposition raises WC reference signals and interferes with CL-based ¹⁸F-Florbetapir Aβ PET quantification, which systematically underestimates true cerebral Aβ burden. The WM represents a robust reference region, and the cerebellar SUVR_WM threshold of 0.428 is reliable and practical for identifying cerebellar cortical Aβ status.

Key Points

Cerebellar cortical Aβ deposition increases tracer uptake in the whole cerebellum reference region, interfering with Centiloid measurement and underestimating cerebral cortical amyloid burden.

Compared with the whole cerebellum and pons, white matter is a robust reference region and avoids bias in Aβ PET quantification.

A quantitative diagnostic threshold of cerebellar SUVR_WM = 0.428 was established and validated for detecting cerebellar cortical Aβ deposition.

Introduction

Alzheimer's disease (AD) is a progressive neurodegenerative disease and the leading global cause of dementia. It is characterized by progressive deterioration of cognitive function and behavioral impairment, placing a substantial socioeconomic burden on patients, caregivers, and healthcare systems [1, 2]. With ongoing global population aging, the prevalence of AD is steadily rising, rendering early and accurate identification critical for timely intervention and disease management. The pathological accumulation of amyloid-beta (Aβ) plaques is widely recognized as the upstream initiating event in AD pathogenesis. This pathological change occurs approximately 20 years prior to clinical manifestation and has become a core pathological hallmark for the definitive diagnosis of AD [3]. Aβ positron emission tomography (PET) enables in vivo visualization and quantification of cerebral Aβ pathology and has become a core biomarker for the diagnosis of AD [4]. ¹⁸F-Florbetapir (AV45) is one of the most commonly used Aβ PET tracers. Quantitative analysis of Aβ PET enables reliable identification of Aβ deposition and assessment of its overall burden. The standardized uptake value ratio (SUVR), calculated by normalizing tracer retention in cortical target regions to that in a designated reference region, is the primary quantitative metric for Aβ PET analysis. Nevertheless, SUVR values are tracer-specific, which limits direct cross-cohort comparison. The Centiloid (CL) method overcomes this limitation by converting tracer-specific SUVR measurements into a unified standardized scale anchored to 0 for young healthy controls and 100 for typical AD patients. This approach achieves harmonized Aβ PET quantification across multicenter cohorts and facilitates longitudinal monitoring of Aβ accumulation dynamics [5]. CL values have been increasingly utilized as a quantitative endpoint in clinical trials of anti-Aβ antibodies, including lecanemab [6], donanemab [7], and aducanumab [8]. This highlights their critical role in assessing treatment efficacy, serves as a valuable complement to visual assessment, and provides important insights for trial design and treatment management [9]. Multiple reference regions have been adopted for Aβ PET quantification across tracers and clinical cohorts, such as the whole cerebellum (WC), white matter (WM), and pons [10–19]. For CL scaling, however, the WC is specifically adopted as the standard reference region to ensure cross-tracer comparability and harmonized quantification [5]. Compared to other reference regions, the WC exhibits lower variability [18], greater robustness against brain atrophy and differences in image resolution [20], enhanced ability to detect subtle early Aβ accumulation [14], and more consistent results across different tracers [17]. Thus, the WC is generally considered a suitable reference region due to its minimal physiological Aβ deposition.

However, cerebellar cortical Aβ deposition has been reported across multiple populations. According to the Thal phase system, cerebellar cortical involvement is restricted to advanced AD (phase 5) [21, 22]. Furthermore, in autosomal dominant AD (ADAD), cerebellar cortical involvement occurs early in the disease course. A PET study of cognitively unimpaired Presenilin-1 (PSEN1) E280A mutation carriers detected cerebellar cortical Aβ deposition approximately one decade prior to symptom onset [23]. Crucially, a recent autopsy study in the oldest-old suggested that cerebellar cortical Aβ deposition may occur in older clinical populations, thereby potentially compromising the predictive accuracy of Aβ PET quantification when using cerebellar gray matter (GM) as the reference region [24]. When present, cerebellar cortical Aβ induces specific tracer binding within the reference region, artificially elevating the reference region signal in SUVR and CL calculations. This could result in a systematic underestimation of true cerebral cortical Aβ burden, which may lead to misclassification of Aβ status and confound clinical or research interpretations.

These observations underscore the critical need for reference regions that are robust to cerebellar cortical Aβ deposition. While the presence of cerebellar cortical Aβ has been well documented, its quantitative impact on PET measurements remains largely uninvestigated. Previous studies have predominantly focused on the pathological and clinical correlates of cerebellar cortical Aβ, including its associations with disease stage, genetic mutations, and cognitive phenotypes. However, the magnitude of the potential bias introduced by cerebellar cortical Aβ contamination in SUVR and CL values remains unquantified. Furthermore, the implications of such bias for determining Aβ positivity in individuals with cerebellar cortical involvement remain unexplored. Addressing these knowledge gaps is essential for improving the accuracy and reliability of Aβ PET quantification, which is pivotal for the diagnosis and monitoring of neurodegenerative diseases associated with Aβ pathology. Therefore, the present study aimed to (1) investigate whether cerebellar cortical Aβ deposition introduces bias into SUVR and CL measurements; (2) identify a reliable reference region for Aβ PET quantification in individuals with cerebellar cortical Aβ involvement; and (3) establish and validate a quantitative threshold for the detection of cerebellar cortical Aβ deposition.

Materials and methods

This study adhered to the Declaration of Helsinki and was approved by the Ethics Committee of Xuanwu Hospital, Capital Medical University. Written informed consent was obtained from all participants or their legally authorized representatives.

Participants and study design

A total of 576 participants were initially enrolled at the Department of Neurology, Xuanwu Hospital, from August 1, 2023, to November 30, 2025. Each participant underwent a comprehensive battery of assessments for diagnosis confirmation, including clinical neurological evaluation, standardized neurocognitive testing, structural magnetic resonance imaging (MRI) scans, and routine laboratory analyses. Additionally, all participants underwent ¹⁸F-AV45 Aβ PET imaging to determine Aβ pathology status. Of the 576 initially enrolled participants, 359 were identified as Aβ-positive (Aβ +) and constituted the primary discovery cohort. For the primary analysis, this cohort was stratified into two subgroups based on visual assessment of cerebellar cortical Aβ deposition: the cerebellar cortical Aβ + subgroup (n = 52, with deposition) and the cerebellar cortical Aβ-negative (Aβ −) subgroup (n = 307, without deposition). Furthermore, an internal validation subset of 30 Aβ + participants with equivocal visual assessment of cerebellar cortical Aβ deposition was selected from the discovery cohort. The flowchart of participant selection and stratification is presented in Fig. 1.

Fig. 1.

Flowchart of the present study. The internal validation subset was selected from the discovery cohort, comprising participants with equivocal visual assessments of cerebellar cortical Aβ deposition. Abbreviations: Aβ Amyloid-beta; PET Positron emission tomography

A total of 359 Aβ + participants were categorized into three diagnostic subgroups: mild cognitive impairment (MCI), AD dementia, and other neurological disorders. The diagnoses of MCI and AD dementia were based on the core clinical criteria according to the recommendations of the National Institute on Aging and Alzheimer’s Association (NIA-AA) [25, 26]. All remaining participants who did not meet the diagnostic criteria for MCI or AD dementia were classified into the other neurological disorders category. Cognitive function was evaluated in all enrolled participants using the Mini-Mental State Examination (MMSE) and the Montreal Cognitive Assessment (MoCA), which are widely used standardized tools for assessing global cognitive performance.

Participants were further stratified by apolipoprotein E (APOE) genotype into two groups: ε4 carriers (including ε3/ε4 and ε4/ε4 genotypes) and ε4 non-carriers (including ε2/ε3, ε2/ε2, and ε3/ε3 genotypes). Participants with the ε2/ε4 genotype were excluded from the subsequent analysis, considering the antagonistic effects of the ε2 and ε4 alleles on AD pathogenesis. Among the 359 Aβ + participants, APOE genotype data were available for 42 participants in the cerebellar cortical Aβ + group, of whom 15 (35.7%) were APOE ε4 carriers. In contrast, APOE genotype data were available for 251 participants in the cerebellar cortical Aβ − group, including 124 (49.4%) APOE ε4 carriers.

Cerebrospinal fluid AD biomarkers

Cerebrospinal fluid (CSF) samples were collected via lumbar puncture in strict adherence to international clinical guidelines [27]. Concentrations of the core AD biomarkers, including Aβ-42, Aβ-40, phosphorylated tau 181 (P-tau181), and total tau (T-tau), were available enzyme-linked immunosorbent assay (ELISA) kits (Fujirebio, Tokyo, Japan). The Aβ-42/Aβ-40 ratio and P-tau181/Aβ-42 ratio were then calculated subsequent analyses. All assays were performed in accordance with the standardized automated and manual assays [28]. All laboratory operations were carried out followed well-established protocols by researchers who were blinded to participants’ clinical information and other biomarker results to minimize detection bias.

Imaging acquisition and analysis

All integrated PET/MRI examinations were conducted on simultaneous time-of-flight (TOF) hybrid imaging systems (uPMR 790, United Imaging Healthcare, Shanghai, China; Signa, GE Healthcare, Waukesha, WI, USA). To minimize systematic quantitative differences between the two PET/MRI platforms, all PET reconstruction workflows were standardized across the two systems, including TOF modeling, ordered-subsets expectation maximization (OSEM) iterations with 32 subset divisions per iteration across 8 reconstruction cycles, and post-reconstruction Gaussian smoothing with 3-mm full-width at half-maximum (FWHM) filter. A unified 2-point Dixon attenuation correction sequence was used for all participants to eliminate systematic differences in attenuation maps derived from different hardware. For Aβ PET imaging in the discovery cohort, we used ¹⁸F-AV45 tracer including 359 participants. PET imaging acquisition began with 50–70 min post-injection and lasted for 20 min. MRI scans were obtained concurrently during the PET acquisition. The following pulse sequence configuration was used to acquire three-dimensional T1-weighted imaging (3D T1WI): flip angle = 10°, voxel size = 1.00 × 1.00 × 1.00 mm³, 192 slices, and repetition time/echo time = 7.2 ms /3 ms.

The PET and structural MRI data were additionally processed through partial volume correction (PVC) and spatial normalization using Statistical Parametric Mapping (SPM12, Wellcome Department of Imaging Neuroscience, Institute of Neurology, London, UK) within MATLAB R2024a (The MathWorks Inc., Natick, MA), as described in our earlier studies [29–31]. This correction was performed using an iterative reblurred Van Cittert method, with a 3D Gaussian kernel of 8 mm FWHM serving as the spatial smoothing function and a step length (α) of 1.5. The iterative process was terminated when the relative percentage change between successive corrected images fell below 1%, following the standard implementation of the algorithm [32, 33]. Given that cerebellar cortical Aβ deposition has been reported in a subset of cases and could potentially confound quantification, we employed a multi-reference approach to quantify regional Aβ burden. The mean standardized uptake value (SUV) was extracted from each participant’s scan for both the CL-based cortical region of interest (ROI) and the three reference regions: WC, pons, and WM. Volumetric masks (1 mm isotropic resolution) were predefined as follows: cortical mask (voi_ctx_1mm), WC mask (voi_WhlCbl_1mm), and pons mask (voi_Pons_1mm) were obtained from the Global Alzheimer’s Association Interactive Network (GAAIN; www.gaain.org). The WM mask was derived from the Johns Hopkins University (JHU) ICBM-DTI-81 atlas [34]. To reduce signal contamination from GM and CSF when using WM as a reference region, only core WM voxels distant from GM-WM boundaries were extracted after PVC and spatial normalization to Montreal Neurologic Institute (MNI) space [35, 36].

SUVRs for cortex were then calculated as the cortex-to-reference region uptake ratio. Specifically, we computed SUVRs using the reference regions defined above: SUVR_WC (cortex-to-WC), SUVR_pons (cortex-to-pons), and SUVR_WM (cortex-to-WM). A mean Aβ PET SUVR image was generated from the PET images of all Aβ + participants using the WC as reference region. To identify a reference region that remains stable in the presence of cerebellar cortical Aβ deposition, we compared SUV and SUVR values derived from each reference region between cerebellar cortical Aβ + and Aβ − subgroups. To reduce the confounding influence of inter-scan variability in tracer uptake on visual assessment, SUVR images were generated using the optimal reference region identified through the above comparison, enabling a more reliable visualization of longitudinal changes in Aβ deposition.

Conversion to CL scale

To allow quantification of Aβ burden on a common scale, SUVR_WC values were converted to CL. Standard GAAIN cortex and WC VOIs were used, with the CL scale anchored to reference Pittsburgh Compound B (PiB)-SUVR_WC values of GAAIN young controls (0 CL) and AD patients (100 CL) [5].

Calibration of AV45-SUVR_WC to the CL scale was performed using the GAAIN-provided PiB-AV45 paired calibration dataset (n = 46) [37], which includes both ¹¹C-PiB and ¹⁸F-AV45 scans from the same subjects and serves as the official reference for Centiloid calibration. Both tracers’ scans were processed with our in-house pipeline using SPM12 with PVC, as detailed above. This PVC-based calibration follows the Centiloid Project’s guidance for non-standard methods and has been validated for ¹⁸F-AV45 PET/MRI data [38]. Linear regression between PiB-SUVR_WC and AV45-SUVR_WC yielded the following two-step conversion equations:

This conversion pipeline and its validation have been detailed in our previous multicenter study [39].

Aβ PET images visual assessment

All Aβ PET images were initially interpreted by two readers (Reader 1, LG; Reader 2, SZY), who were blinded to clinical information, had completed relevant training, and had experience in assessing over 3000 scans. The readers first classified each image as Aβ + or Aβ − based on cortical deposition. Scans were classified as global Aβ + when cortical radiotracer activity equaled or exceeded WM activity in one or more cerebral lobes. For participants classified as Aβ + , they further assessed the presence of cerebellar cortical Aβ deposition. Elevated tracer uptake localized within the cerebellar cortex was defined as cerebellar cortical Aβ + , which stratified participants into cerebellar cortical Aβ + subgroup or Aβ − subgroup. Any discrepancies between the two readers were reviewed by Reader 3 (ZLZ), who has 20 years of clinical experience, and the adjudicated result was final. This visual assessment served as the gold standard for participant grouping and all subsequent analyses.

Statistical analysis

The normal distribution of continuous variables was evaluated using the Shapiro–Wilk test and visual examination of Q-Q plots. Between-group differences were analyzed using independent t-test or Mann–Whitney U test. Categorical variables were compared using chi-square test, with Fisher’s exact test applied when expected frequencies were below 5. To control the false discovery rate (FDR) across multiple comparisons, the Benjamini‑Hochberg procedure was applied to the resulting P-values. Scanner type was included as a fixed covariate in separate intergroup comparisons for each PET quantitative metric, namely SUV, SUVR and CL, to adjust for potential measurement bias introduced by different PET/MRI platforms.

To address the substantial sample size imbalance (1:6 ratio), robust between-group comparisons were performed. Effect sizes were reported as mean or median differences, as appropriate based on data distribution, and statistical inference was drawn from bootstrap resampling (n = 1000). Statistical significance was declared when the bootstrap 95% confidence interval (CI) excluded zero [40].

To establish a diagnostic threshold for cerebellar cortical Aβ deposition, receiver operating characteristic (ROC) curve analysis was performed using SUVRs calculated with the optimal reference region identified above. The optimal SUVR threshold was determined by maximizing the Youden index. All diagnostic metrics, including area under the curve (AUC), accuracy, sensitivity, specificity, positive predictive value (PPV), and negative predictive value (NPV), were calculated, with 95% CI derived using bootstrap resampling (n = 1000). Finally, Cohen’s kappa (κ) was used to measure the agreement between visual assessment (the gold standard) and SUVR-based classification of cerebellar cortical Aβ deposition.

All analyses were performed using R (version 4.5.2; R Foundation for Statistical Computing, Vienna, Austria), IBM SPSS Statistics (version 26; IBM Corp., Armonk, NY, USA), and MATLAB (version R2024a; MathWorks, Natick, MA, USA). A 2-sided α of 0.05 was set for all statistical tests.

Results

Demographics and clinical characteristics

The characteristics of the discovery cohort, stratified by cerebellar cortical Aβ − (n = 307) and Aβ + (n = 52), are summarized in Table 1. Notably, the cerebellar cortical Aβ + group exhibited a significantly lower CL value compared with the Aβ − group (P < 0.001, bootstrap 95% CI: − 73.31 to − 46.82, bootstrap P < 0.001). No significant between-group differences were detected for any other continuous variables (both P-value and bootstrap P-value > 0.05). Nevertheless, descriptively, the cerebellar cortical Aβ + group showed a trend toward a more severe clinical profile, which was characterized by lower cognitive scores and reduced CSF Aβ-42 levels. There were no significant differences in sex distribution (P = 0.071) or APOE status (P = 0.100) between the two groups.

Table 1.

Characteristics of the discovery cohort participants

Characteristics	Cerebellar cortical Aβ −	Cerebellar cortical Aβ +	Median difference (bootstrap 95% CI)	P	Bootstrap P
No. of subjects	307	52	-	-	-
Age (years)	64.00 (58.00, 70.00)	65.50 (59.50, 70.00)	1.50 (-1.00, 4.00)	0.886	0.438
Female, n (%)	194 (63.2)	26 (50.0)	-	0.071	-
MMSE score	18.00 (12.00, 22.00)	17.50 (11.00, 20.25)	-0.50 (-3.50, 2.00)	0.721	> 0.999
MoCA score	13.00 (8.00, 17.00)	9.50 (6.00, 15.00)	-3.50 (-5.50, 0.00)	0.446	0.120
APOE status	251	42	-	0.100	-
ε4 Carriers, n (%)	124 (49.4)	15 (35.7)	-	-	-
ε4 Non-carriers, n (%)	127 (50.6)	27 (64.3)	-	-	-
Aβ PET, CL	89.20 (60.53, 122.80)	24.47 (11.43, 38.23)	-62.88 (-73.31, -46.82)	< 0.001	< 0.001
CSF tests
Aβ-42 (pg/ml)	481.26 (407.86, 602.38)	437.64 (369.43, 582.60)	-43.61 (-77.69, 10.61)	0.645	0.124
T-tau (pg/ml)	464.79 (279.63, 701.41)	473.64 (321.42, 688.61)	8.86 (-64.80, 125.42)	0.721	0.664
P-tau181 (pg/ml)	85.78 (63.75, 135.60)	86.14 (64.75, 115.99)	0.36 (-15.94, 15.13)	0.721	0.952
Aβ-42/Aβ-40	0.04 (0.03, 0.06)	0.04 (0.03, 0.05)	0.00 (-0.01, 0.00)	0.446	> 0.999
P-tau181/Aβ-42	0.19 (0.13, 0.29)	0.20 (0.10, 0.30)	0.01 (-0.08, 0.07)	0.886	0.966
Diagnosis category, n (%)			-	0.847	-
AD	239 (77.8)	42 (80.8)	-	-	-
MCI	38 (12.4)	5 (9.6)	-	-	-
Othera	30 (9.8)	5 (9.6)	-	-	-

Data were presented with median (IQR). Median difference and bootstrap 95% CI were calculated as cerebellar cortical Aβ + group minus Aβ − group. Group comparisons: Mann–Whitney U-test (age, MMSE, MoCA, CL, CSF tests), chi-square test (gender, APOE status, diagnosis category)

^a Other include Vascular Dementia, Frontotemporal Dementia, Dementia with Lewy Bodies, Cerebral Amyloid Angiopathy, Parkinson’s Disease, paraneoplastic neurological syndrome, general paresis of the insane, pituitary adenoma, and other diseases with unknown etiologies

Abbreviations: CI Confidence interval; IQR Interquartile range; MMSE Mini-Mental State Examination; MoCA Montreal Cognitive Assessment; APOE Apolipoprotein E; Aβ Amyloid-beta; PET Positron emission tomography; CL Centiloid; CSF Cerebrospinal fluid; T-tau Total tau; P-tau181 Phosphorylated tau 181; AD Alzheimer Disease; MCI Mild cognitive impairment

Figure 2 presents the mean Aβ PET SUVR images of the discovery cohort. Representative cases with and without cerebellar cortical Aβ deposition were presented in Fig. 3, demonstrating a clear visual distinction between the two groups.

Fig. 2.

Mean Aβ PET SUVR images of discovery cohort, with the WC as the reference region. Axial (top row), coronal (middle row), and sagittal (bottom row) views are shown. The color scale indicates SUVR values. Abbreviations: Aβ Amyloid-beta; PET Positron emission tomography; SUVR Standardized uptake value ratio; WC Whole cerebellum

Fig. 3.

Representative Aβ PET images of two Aβ + participants, with and without cerebellar cortical Aβ deposition. a Visible Aβ deposition is observed in the cerebellar cortex indicated by the arrows (CL = 12.4). b No obvious Aβ deposition is detected in the cerebellar cortex indicated by the arrows (CL = 160.6). Abbreviations: Aβ Amyloid-beta; PET Positron emission tomography; CL Centiloid

Group differences of regional SUV and SUVR

Initially, we compared SUV across multiple regions, including the cortex, WC, WM, and pons, between the cerebellar cortical Aβ + and cerebellar cortical Aβ − groups. Most notably, significant differences between groups were observed in SUV_WC (P < 0.001, bootstrap 95% CI: 126.47 to 695.44, bootstrap P = 0.012), SUV_cortex (P = 0.044, bootstrap 95% CI: − 885.28 to − 190.68, bootstrap P = 0.008), and SUV_WM (P = 0.223, bootstrap 95% CI: − 1316.13 to − 296.73, bootstrap P < 0.001) (Fig. 4a, d and b). However, no significant differences were found in SUV_pons (P = 0.662, bootstrap 95% CI: − 815.55 to 48.46, bootstrap P = 0.110) (Fig. 4c).

Fig. 4.

Group differences in SUV and SUVR values. Box plots display the distribution of SUV_WC (a), SUV_WM (b), SUV_pons (c), SUV_cortex (d), SUVR_WC (e), SUVR_WM (f) and SUVR_pons (g) between subjects with and without cerebellar cortical Aβ deposition. Blue boxes represent the cerebellar cortical Aβ– group, and orange boxes represent the cerebellar cortical Aβ + group. (Bootstrap, * P < 0.05, ** P < 0.01, *** P < 0.001.). Abbreviations: SUV Standardized uptake value; SUVR Standardized uptake value ratio; WC Whole cerebellum; WM White matter; Aβ Amyloid-beta

The cerebellar cortical Aβ + group exhibited a significant reduction in SUVR_WC (P < 0.001, bootstrap 95% CI: − 0.681 to − 0.420, bootstrap P < 0.001) (Fig. 4e). When the pons was used as reference region, a significant reduction was observed in SUVR_pons (P < 0.001, bootstrap 95% CI: − 0.124 to − 0.003, bootstrap P = 0.030) (Fig. 4g). In contrast, no significant between-group difference was found in SUVR_WM (P = 0.012, bootstrap 95% CI: − 0.066 to 0.016, bootstrap P = 0.288) (Fig. 4f).

Determination and validation of SUVR threshold for cerebellar cortical Aβ deposition

SUVR_WM was comparable between the cerebellar cortical Aβ + and Aβ − groups, confirming that WM serves as a stable reference region for subsequent analyses. Therefore, WM was selected as the reference region to calculate cerebellar SUVR_WM, and the discriminative ability of cerebellar SUVR_WM for identifying cerebellar cortical Aβ deposition status was further evaluated.

ROC analysis was performed to evaluate the discriminative performance of cerebellar SUVR_WM (Fig. 5). The results demonstrated an excellent AUC of 0.931 (95% CI: 0.892 to 0.962). The optimal SUVR threshold was determined using the Youden index, which was 0.428. At this threshold, the diagnostic performance metrics were as follows: sensitivity of 86.5% (95% CI: 72.4% to 96.2%), specificity of 86.3% (95% CI: 78.7% to 97.7%), PPV of 51.7% (95% CI: 38.1% to 84.8%), NPV of 97.4% (95% CI: 95.0% to 99.3%), and overall diagnostic accuracy of 86.4% (95% CI: 80.2% to 95.3%).

Fig. 5.

ROC curve of cerebellar SUVR_WM for detecting cerebellar cortical Aβ deposition. The shaded blue area represents the bootstrap-derived 95% CI for the ROC curve. The optimal threshold value (0.428) was determined by the Youden index, with a sensitivity of 86.5% and a specificity of 86.3%. The AUC was 0.931 (95% CI: 0.892–0.962). Abbreviations: ROC receiver operating characteristic; SUVR Standardized uptake value ratio; WM White matter; Aβ Amyloid-beta; CI Confidence interval; AUC area under the curve

In the internal validation subset, participants were stratified into cerebellar cortical Aβ + and Aβ − groups based on the pre-determined cerebellar SUVR_WM threshold of 0.428. Specifically, 19 subjects were classified into the cerebellar cortical Aβ + group, while 11 subjects were assigned to the Aβ − group. Detailed demographic and clinical characteristics of the internal validation subset are summarized in Table 2. No significant differences were observed between groups for any demographic or clinical variables (all P > 0.05) except for cerebellar SUVR_WM, which was significantly higher in the Aβ + group than in the Aβ − group (P < 0.001). Furthermore, the agreement between visual interpretation (the gold standard) and SUVR-based classification was substantial (n = 28/30, Cohen’s κ = 0.856).

Table 2.

Characteristics of the internal validation subset participants

Characteristics	Cerebellar cortical Aβ +  (cerebellar SUVRWM ≥ 0.428)	Cerebellar cortical Aβ −  (cerebellar SUVRWM < 0.428)	P
No. of subjects	19	11	-
Age (years), mean (SD)	64.26 (7.51)	64.09 (10.01)	> 0.999
Female, n (%)	14 (73.7)	6 (54.5)	0.425
MMSE score, mean (SD)	17.53 (5.20)	14.45 (6.96)	0.539
MoCA score, mean (SD)	11.68 (5.49)	10.00 (6.63)	0.725
APOE status	15	10	0.442
ε4 Carriers, n (%)	10 (66.7)	5 (50.0)	-
ε4 Non-carriers, n (%)	5 (33.3)	5 (50.0)	-
Cerebellar SUVRWM, mean (SD)	0.45 (0.01)	0.40 (0.02)	< 0.001
CSF tests
Aβ-42 (pg/ml), mean (SD)	469.09 (115.30)	534.65 (126.61)	0.539
T-tau (pg/ml), median (IQR)	417.32 (245.92, 551.48)	419.79 (306.31, 547.20)	> 0.999
P-tau181 (pg/ml), mean (SD)	94.34 (39.15)	83.31 (44.00)	0.725
Aβ-42/Aβ-40, mean (SD)	0.04 (0.02)	0.05 (0.02)	0.929
P-tau181/Aβ-42, median (IQR)	0.19 (0.14, 0.24)	0.17 (0.09, 0.21)	0.691
Visual positive, n (%)a	18 (94.7)	1 (9.1)	-

Data were presented as mean (SD) or median (IQR). Group comparisons: independent t-test (age, MMSE, MoCA, cerebellar SUVR_WM, CSF Aβ-42, CSF P-tau181, CSF Aβ-42/Aβ-40), Mann–Whitney U test (CSF T-tau, P-tau181/Aβ-42), chi-square test (gender, APOE status)

^a Visual assessment in the internal validation subset was adjudicated by the third reader

Abbreviations: SD Standard deviation; IQR Interquartile range; MMSE Mini-Mental State Examination; MoCA Montreal Cognitive Assessment; APOE Apolipoprotein E; SUVR Standardized uptake value ratio; WM White matter; Aβ Amyloid-beta; CSF Cerebrospinal fluid; T-tau Total tau; P-tau181 Phosphorylated tau 181

Cerebellar cortical Aβ deposition after lecanemab treatment

Two patients with cerebellar cortical Aβ deposition received lecanemab treatment and underwent longitudinal Aβ PET follow-up at 7 and 11 months, respectively.

Aβ PET images revealed distinct patterns of Aβ deposition changes following treatment. Visual assessment of raw PET images (Fig. 6a and b) showed that cerebellar cortical Aβ deposition remained positive in both patients at follow-up. Notably, in patient 1, cortical Aβ deposition did not visibly decrease; in patient 2, cortical Aβ deposition showed a visible reduction. In patient 1 (Fig. 6a, c and e), SUVR_WM remained essentially stable, changing only from 0.55 to 0.56, while cerebellar SUVR_WM increased from 0.56 to 0.61, both exceeding the established threshold for cerebellar cortical Aβ positivity (0.428). In contrast, CL values decreased from 8.9 to 3.6. In patient 2 (Fig. 6b, d and e), SUVR_WM decreased from 0.68 to 0.61, while cerebellar SUVR_WM increased from 0.44 to 0.55, remaining above the 0.428 threshold. CL values decreased from 70.9 to 22.4.

Fig. 6.

Longitudinal Aβ PET images and quantitative measures in two patients with cerebellar cortical Aβ deposition receiving lecanemab treatment. Baseline (upper row) and follow-up (lower row) raw PET images of patient 1 (a) and patient 2 (b), and SUVR_WM images of patient 1 (c) and patient 2 (d). Visual assessment of cerebellar cortical Aβ deposition remained positive in both patients at follow-up, cerebral cortical Aβ deposition did not visibly decrease in patient 1, whereas patient 2 showed a visible reduction (a, b). (e) Quantitative measures for both patients. Values in bold indicate cerebellar SUVR_WM above the threshold of 0.428. After lecanemab treatment, distinct changes in quantitative Aβ metrics were observed in both patients. In patient 1, cerebral cortex SUVR_WM remained essentially stable, changing only from 0.55 to 0.56, cerebellar SUVR_WM increased from 0.56 to 0.61, and CL values decreased from 8.9 to 3.6. In patient 2, cerebral cortex SUVR_WM decreased from 0.68 to 0.61, cerebellar SUVR_WM increased from 0.44 to 0.55, and CL values decreased from 70.9 to 22.4. Abbreviations: MMSE Mini-Mental State Examination; MoCA Montreal Cognitive Assessment; APOE Apolipoprotein E; Aβ Amyloid-beta; PET Positron emission tomography; SUVR Standardized uptake value ratio; WM White matter; CL Centiloid

Discussion

This study systematically investigated the impact of cerebellar cortical Aβ deposition on CL values, regional SUVs and multi-reference SUVRs. The primary objectives were to compare regional variations in these metrics, identify a stable reference region, and establish a diagnostic threshold for cerebellar cortical Aβ positivity. In relation to the research questions posed, our results indicate that cerebellar cortical Aβ deposition significantly reduced CL values, suggesting that conventional quantification methods for Aβ PET are biased when cerebellar pathology is present. We validated WM as the only stable reference region among those tested and established a cerebellar SUVR_WM threshold of 0.428 for identifying cerebellar cortical Aβ burden with high diagnostic accuracy (AUC = 0.931). However, the PPV was notably lower (51.7%), which may be due to the limited sample size of the cerebellar cortical Aβ + group (n = 52). In contrast, the excellent NPV (97.4%) supports its utility as a screening tool to rule out cerebellar cortical Aβ involvement. Together, these findings suggest that cerebellar cortical Aβ deposition systematically biases conventional Aβ PET quantification, challenging the validity of WC as an unaffected reference region in Aβ PET CL analyses.

The significantly reduced CL values observed in individuals with cerebellar cortical Aβ deposition represent the most clinically impactful finding of this study. Mechanistically, this bias arises from dual opposing effects: cerebellar cortical Aβ accumulation elevates SUV_WC while SUV_cortex is conversely reduced. This finding is consistent with previous studies that have shown cerebellar pathology affects quantification [5, 9, 23], but contrasts with a previous report that cerebellar cortical Aβ plaques do not significantly affect SUVRs when using the cerebellum as reference [41]. One possible explanation for this difference is that their analysis compared individuals with varying degrees of cerebellar pathology, whereas our study directly contrasted groups with and without cerebellar cortical deposition, suggesting that the bias emerges once plaques appear rather than increasing with plaque burden. The reduction in SUV_cortex in individuals with cerebellar cortical Aβ deposition may reflect the natural progression of Aβ pathology in advanced AD, rather than a direct causal relationship. Direct experimental evidence demonstrates that continuous clearance of soluble Aβ from the CSF leads to a significant reduction in cortical Aβ plaques in animal models, establishing a causal link between CSF soluble Aβ levels and cortical Aβ pathology [42]. In the natural course of AD, CSF soluble Aβ levels progressively decline in advanced stages [43]. Given that cerebellar cortical Aβ deposition is a feature of advanced disease, such CSF changes may contribute to the observed SUV_cortex reduction. A more direct explanation lies in concurrent synaptic loss. In vivo PET studies using synaptic vesicle glycoprotein 2A (SV2A) tracers have consistently demonstrated significant cerebral synaptic loss in Aβ + patients [44–46]. These synaptic deficits are closely linked to reduced cerebral blood flow and neuronal activity, which directly impact Aβ PET tracer uptake [47, 48]. Thus, the lower SUV_cortex in these individuals may reflects underlying synaptic dysfunction rather than a decrease in total Aβ burden.

To identify a stable reference region unaffected by cerebellar pathology, we examined SUV stability across tested regions. Notably, SUV_WM exhibited significant intergroup differences, whereas SUV_pons did not. In line with the existing literature, the findings confirm that SUV_pons remains stable across groups [49]. However, while SUV_pons remained stable, SUVR_pons was lower in the cerebellar cortical Aβ + group, consistent with the reduction in SUV_cortex. This indicates that the lower SUVR_pons is probably driven by SUV_cortex reduction rather than pons instability or external cerebellar influences. In contrast, SUVR_WM showed no significant between-group difference, identifying WM as a robust reference region at the SUVR level, unaffected by both cerebellar and cerebral pathology. Together with previous studies demonstrating its reliability in longitudinal analyses [10–13, 50], WM may be suited for evaluating Aβ-modifying treatments in patients with cerebellar pathology. Although the CL value is a well-established metric, its interpretation requires careful consideration in patients with cerebellar cortical Aβ deposition.

A key theoretical contribution of this study lies in its clarification of how cerebellar pathology affects the validity of established PET quantification methods. The findings extend existing theory by demonstrating that the assumption that the cerebellum is unaffected by Aβ pathology does not hold uniformly across disease stages and fails in advanced AD with cerebellar cortical involvement. By highlighting the role of cerebellar cortical Aβ deposition as a previously underappreciated source of measurement bias, the case illustrations further support this conclusion. For patient 1, the pre-treatment CL value was 8.9, which was already lowered by cerebellar cortical Aβ deposition despite visual positivity in the cerebral cortex. After treatment, CL decreased to 3.6, while cerebellar cortical Aβ remained positive and cerebral cortical Aβ remained visually positive without conversion to negativity. SUVR_WM remained essentially stable (0.55 to 0.56). For patient 2, the baseline CL value was 70.9, with cerebellar cortical Aβ also positive and cerebral cortical Aβ burden substantially higher than that in the cerebellum. After treatment, CL decreased to 22.4, falling below the recommended Aβ positivity threshold of 30 CL [9]. Visual assessment showed an appreciable reduction in cerebral cortical Aβ deposition; however, both cerebral and cerebellar cortical Aβ remained positive. Collectively, these two cases demonstrate that when cerebellar cortical Aβ is present, CL values are systematically underestimated and should not be used alone to determine Aβ status conversion or treatment response. Accordingly, exclusive reliance on CL values may lead to erroneous assessment of Aβ clearance and misguide clinical decisions. Notably, cerebellar SUVR_WM increased in both patients after treatment, indicating persistent cerebellar cortical Aβ pathology despite lecanemab therapy. Unlike the cerebral cortex, where compact cored plaques predominate, the cerebellum is characterized by diffuse Aβ plaques [51, 52]. This distinction may have therapeutic implications, as evidenced by autopsy findings from the AN-1792 trial, where cerebellar diffuse plaques resisted clearance despite neocortical plaque removal [53]. Since active immunization shares microglial pathways with passive antibodies such as lecanemab, including fragment crystallizable-gamma (Fcγ) receptor-mediated phagocytosis and upregulation of triggering receptor expressed on myeloid cells 2 (TREM2) and APOE that correlate with Aβ clearance [54], this observation may extend to lecanemab therapy. In this context, the limited response of cerebellar diffuse deposits in our patients is consistent with the notion that diffuse plaques are cleared via pathways less dependent on Fcγ receptor-mediated phagocytosis than compact cored plaques [55]. Thus, while lecanemab may engage clearance pathways for cerebral cortical plaques, the same mechanisms appear less operative for cerebellar diffuse deposits, accounting for the persistent cerebellar SUVR_WM we observed despite treatment. These findings have important implications for quantitative diagnosis and therapeutic trials. Clinicians should be aware that cerebellar cortical Aβ deposition artificially lowers CL values, potentially leading to misclassification. The WM reference region offers a reliable alternative, and the cerebellar SUVR_WM threshold provides a practical tool for screening cerebellar cortical Aβ positivity. For longitudinal monitoring, SUVR_WM may provide a more robust metric than CL-based quantification, which is susceptible to bias from cerebellar cortical deposition.

Several limitations should be acknowledged. At the same time, these implications should be interpreted with caution. First, the study included only two patients with longitudinal follow-up, limiting the generalizability of the longitudinal findings. Second, the optimal cerebellar SUVR_WM threshold was derived from a single-center cohort and required independent validation in larger, multicenter cohorts. Third, while our primary analyses were performed using ¹⁸F-AV45, direct replication using other tracers in independent cohorts remains warranted. Future studies should further investigate the distinct patterns of cerebral and cerebellar cortical Aβ dynamics, particularly the persistence of cerebellar cortical deposition despite treatment. Larger, multicenter cohorts are needed to validate the proposed cerebellar SUVR_WM threshold and to assess the utility of WM-referenced SUVR for monitoring treatment response in patients with cerebellar pathology.

In conclusion, this study challenges the widely used WC reference region in Aβ PET quantification by revealing that cerebellar cortical Aβ deposition systematically confounds the estimation of cerebral Aβ burden when using WC as reference. WM proves a robust reference, even in cases with suspected cerebellar pathology. The proposed cerebellar SUVR_WM threshold provides a standardized, readily implementable tool for detecting cerebellar cortical Aβ positivity. Clinically, adopting WM reference reduces false-negative cerebral Aβ assessments, improves anti‑Aβ therapy stratification, and enhances diagnostic accuracy in atypical or early‑stage AD. Scientifically, these findings redefine reference region validity criteria, highlight the need to account for cerebellar Aβ pathology in longitudinal and multi‑site studies, and support integrating WM‑based normalization into next‑generation quantitative PET pipelines.

Author contributions

Jie Lu, Shaozhen Yan and Zhilian Zhao contributed to the conception and design of the study. Meixi Wang, Sheng Bi, Hanxiao Xue, Le Guan, Yan Wang, Xuemei Zhang, Xiaocao Liu, Xiaoyin Xu contributed to the literature search and data extraction. Meixi Wang, Sheng Bi, Chun Zhang, Zhigang Qi and Zhilian Zhao contributed to data analysis and interpretation. Meixi Wang and Sheng Bi drafted the initial manuscript. Zhilian Zhao, Shaozhen Yan and Jie Lu revised the manuscript. Zhilian Zhao, Shaozhen Yan and Jie Lu were responsible for the decision to submit the manuscript. All authors contributed to the critical revision of the manuscript. All authors approved the manuscript.

Funding

This research was supported by National Natural Science Foundation of China (Grant No. 82394434), Beijing High level Innovative and Entrepreneurial Youth Top notch Talent Program (Grant No. G202522130) and Xuanwu Hospital Talent Convergence Program (Grant No. HZ2025PYDTR006).

Data availability

The datasets generated during and/or analyzed during the current study are available from the corresponding author on reasonable request.

Declarations

Ethics approval and consent to participate

This study was performed in line with the principles of the Declaration of Helsinki. Approval was granted by the Ethics Committee of Capital Medical University (No. [2024] 131-001).

Consent for publication

Written informed consent was obtained from the parents.

Competing interests

The authors have no relevant financial or non-financial interests to disclose.