Early diagnosis of Alzheimer’s Disease: Graph theoretical analysis of cerebellar network features based on 18F-AV45 PET

Ruyi Li; Shaoping Jiang; Zhaoke Pi; Guisu Chen

doi:10.1371/journal.pone.0342738

Abstract

Pathological and neuroimaging changes in the cerebellum of Alzheimer’s disease (AD) patients have been well documented. However, the changes in cerebellar amyloid plaque deposition connectivity networks during AD progression based on positron emission tomography (PET) imaging remain unclear. We selected ¹⁸F-florbetapir PET (¹⁸F-AV45 PET) imaging data from the Alzheimer’s disease neuroimaging initiative (ADNI) dataset (n = 612) and employed graph theoretical analysis to examine amyloid plaque deposition connectivity, comparing the connectivity differences across cognitively normal (CN), early mild cognitive impairment (EMCI), late mild cognitive impairment (LMCI), and AD groups. In addition, we combined graph theoretical features with the standardized uptake value ratio (SUVR) of regions of interest and applied them to machine learning models for the early diagnosis of AD. As cognitive decline progressed, significant changes in cerebellar network connectivity were observed across groups. Regarding local connectivity, changes in betweenness centrality were evident in multiple cerebellar regions at different cognitive stages. Cerebellar amyloid networks revealed early changes in amyloid plaque deposition connectivity. The machine learning model achieved an area under the curve (AUC) of 0.950 for distinguishing AD from CN, 0.995 for CN vs. EMCI, 0.964 for EMCI vs. LMCI and 0.632 for LMCI vs. AD. These findings provide new insights into the cerebellar pathological features of AD and highlight the potential of this approach for early identification and prediction of AD progression.

Citation: Li R, Jiang S, Pi Z, Chen G (2026) Early diagnosis of Alzheimer’s Disease: Graph theoretical analysis of cerebellar network features based on ¹⁸F-AV45 PET. PLoS One 21(2): e0342738. https://doi.org/10.1371/journal.pone.0342738

Editor: Yi Su, Banner Alzheimer's Institute, UNITED STATES OF AMERICA

Received: July 8, 2025; Accepted: January 28, 2026; Published: February 17, 2026

Copyright: © 2026 Li et al. This is an open access article distributed under the terms of the Creative Commons Attribution License, which permits unrestricted use, distribution, and reproduction in any medium, provided the original author and source are credited.

Data Availability: All imaging files are available from the ADNI database (http://adni.loni.usc.edu).

Funding: The author(s) received no specific funding for this work.

Competing interests: The authors have declared that no competing interests exist.

Introduction

Alzheimer’s disease (AD) represents the most prevalent neurodegenerative condition and is the primary cause of dementia. It is marked by the gradual buildup of amyloid-beta (Aβ) plaques and tau neurofibrillary tangles, processes that begin decades before the appearance of dementia symptoms and progressively deteriorate over time. The progression of the disease may extend over a span of up to 30 years [1]. Identifying neurodegenerative changes at different stages, including cognitively normal (CN), early mild cognitive impairment (EMCI), late mild cognitive impairment (LMCI), and AD, is crucial for enabling timely therapeutic intervention.

Previous studies have consistently demonstrated that AD pathology predominantly involves selective neocortical layers and specific subcortical structures, particularly the parietal, temporal, frontal cortices, and hippocampus [2,3]. However, the role of the cerebellum in neurodegenerative diseases has often been underestimated [4]. The cerebellum plays a role in cognitive processes and emotional regulation, interacting with the cerebral cortex through a distinctive communication pathway [5]. Recent research has highlighted the crucial role of the cerebellum in cognitive functions, showing that damage to this area can result in cognitive impairments in patients with AD [6]. In addition, the cerebellum plays a key role in the onset and progression of AD [7]. In AD patients, the cerebellum exhibits pathological changes similar to those found in the cortical layers and hippocampus, such as Aβ deposition, p-Tau, and neuronal loss [8–10]. Consequently, the cerebellum has become a focal point in studying AD and other neurodegenerative disorders [11].

Graph theory is a mathematical tool for analyzing and quantifying brain networks [12]. By providing a simple model of the brain’s connectome, graph theory considers the entire network structure, represented by nodes and edges [13]. Electrophysiological data and graph theory analyses have revealed synaptic disruption and changes in functional connectivity in AD [14,15]. Topological metrics of AD patients hippocampal and medial temporal lobe structural networks have been uncovered [16]. Recent magnetic resonance imaging (MRI) studies have demonstrated the predictive value of both cerebellar radiological signatures and network characteristics in AD [17]. Graph theory allows for quantifying of brain structural network organization by representing these networks as graphs [18]. Despite extensive research based on graph theory to analyze AD’s pathological features and topological properties, studies on cerebellar topological networks based on positron emission tomography (PET) imaging remain relatively limited. PET can quantify pathological burdens in AD sensitive regions by measuring the standardized uptake value ratio (SUVR) of Aβ [19]. PET is the only technique capable of spatially resolving AD pathology in the living human brain. While structural MRI is a cost effective alternative for measuring neurodegeneration, it has lower specificity for AD pathology compared to PET [20]. The relationship between amyloid pathology as measured by ¹⁸F-florbetapir PET (¹⁸F-AV45 PET) and various cognitive functions has been explored, with the occipital and temporoparietal regions mediating these relationships [21].

Graph theory analysis provides a better understanding of the deposition or metabolic connectivity patterns between different brain regions. The use of graph theory to quantify the ¹⁸F-FDG PET brain covariance has been shown to be particularly relevant to AD, demonstrating a correlation with pathological progression [22]. Clinical assessments, cerebrospinal fluid analysis, and ¹⁸F-FDG PET imaging indicate that, particularly in the early stages of AD, hypo-connectivity and hyper-connectivity play distinct roles, with metabolism in hyper-connected regions acting as a mediator in the neurodegeneration of core pathological areas [23]. Graph theory analysis has shifted from traditional population-based network studies to individual-based approaches [24,25]. For each group, partial correlation coefficients of brain regions are calculated, controlling for age and gender, and an undirected weighted metabolic connectivity matrix is constructed based on brain nodes for population-level graph theory analysis [24]. Individual-specific metabolic networks are constructed by adding each patient to the reference group, calculating the differences between disturbed networks and the reference network, and applying Z-score transformation to the individual difference networks, followed by graph theory analysis to assess network features [25]. Individual-level studies contribute to a deeper understanding of patient heterogeneity. The transition from group-based studies to individualized graph theory research not only enhances diagnostic precision but also serves as a complement to the traditional group-based research.

In PET studies, the amyloid PET derivative ¹⁸F-AV45 PET can detect Aβ plaque deposition in the brain [26]. Previous PET studies have indicated the presence of amyloid plaque deposition in the cerebellum of CN carriers, and the pathological changes typically manifest approximately a decade prior to the clinical onset of symptoms in autosomal dominant Alzheimer’s disease [27]. This makes it valuable for early AD diagnosis, especially when patients exhibit mild cognitive impairment or remain asymptomatic, as it can detect abnormal Aβ deposition, thus providing a basis for early intervention.

The primary objective of this investigation is to analyze changes in cerebellar amyloid plaque deposition connectivity across different cognitive states and apply these findings to early AD diagnostic classification (Fig 1). Specifically, we will compare differences between the CN and EMCI groups, EMCI and LMCI groups, and LMCI and AD groups. Using ¹⁸F-AV45 PET imaging data, we will employ graph theory analysis to construct a network model of the cerebellum and explore the impact of amyloid plaque deposition on the deposition connectivity of various cerebellar regions. The study will analyze changes in cerebellar network connectivity between groups, including metrics such as average degree, clustering coefficient (CC), and centrality, to reveal the characteristics of cerebellar connectivity changes across different cognitive states and their relationship to cognitive function. Furthermore, we hypothesize that features extracted from cerebellar graph theory analysis can provide insights into the potential pathology of AD. Therefore, our goal is to develop and evaluate machine learning (ML) models using cerebellar graph theory analysis features extracted from PET for the early diagnosis of AD.

Download:

Fig 1. Workflow for Data Analysis Pipeline:

(A) Data preprocessing steps. (B) Extraction of the SUVR for each cerebellar region, construction of connectivity matrices, and subsequent graph theory analysis. (C) Statistical analysis based on graph theory features and visualization of differential nodes. (D) Identification of the most relevant features through least absolute shrinkage and selection operator regression, using graph theory analysis features, with performance assessed via 5-fold cross-validation. The potential of cerebellar graph theory features for early prediction of AD was evaluated.

https://doi.org/10.1371/journal.pone.0342738.g001

Materials and methods

Subjects

The information utilized in this research was sourced from the Alzheimer’s disease neuroimaging initiative (ADNI) dataset (http://adni.loni.usc.edu). We accessed the ADNI database in June 2024. A total of 612 participants with corresponding MRI and PET imaging data were initially included. The data were obtained from three phases of the ADNI study: ADNI GO, ADNI 2, and ADNI 3. These participants comprised 172 CN individuals, 159 with EMCI, 141 with LMCI, and 140 with AD. All data were fully de-identified, and the authors had no access to any personally identifiable information during or after data collection. Detailed image acquisition is provided in the Supplementary Material S1 File. This analysis incorporated ¹⁸F-AV45 PET scan data and clinical variables, including demographic variables (age, sex) and cognitive assessment results (MMSE scores).

Ethical approval for ADNI was granted by the review boards of the involved institutions, and all participants provided informed consent before enrolling in the study.

Data preprocessing

After acquiring the MRI and ¹⁸F-AV45 PET data, preprocessing of each subject’s brain images was performed using the SPM12 toolbox implemented in MATLAB. The procedures were as follows: first, all MRI and PET data were converted from DICOM to NIFTI format. PET images were then corrected for head motion to reduce artifacts caused by subject movement. Subsequently, the origins of the MRI and motion-corrected PET images were realigned to the anterior commissure posterior commissure line. Each PET image was then coregistered to its corresponding MRI image to achieve intermodal spatial alignment. The coregistered MRI images were segmented into gray matter, white matter, and cerebrospinal fluid, the spatial normalization was performed. Specifically, the coregistered MRI images were nonlinearly registered to the Montreal Neurological Institute (MNI) template, and the resulting deformation fields were applied to the corresponding PET images. The PET images were resampled to a voxel size of 2 × 2 × 2 mm³. Finally, partial volume effect correction (PVEc) of the spatially normalized PET images was performed using the geometric transfer matrix (GTM) method implemented in the PETPVE12 toolbox [28].

Then, intensity normalization of the PET brain images for all subjects was performed under both non-corrected and PVEc-GTM corrected conditions using the SNBPI Toolbox [29]. Specifically, Based on the AAL116 template, cerebellar and typical cortical brain region templates were created, with the typical cortical brain regions including areas of the brain with higher amyloid burden in AD, such as the frontal, temporal, parietal cortices, and precuneus [30]. The cerebellum was subdivided into 26 regions of interest, with the bilateral cerebellar lobules including Crus I, Crus II, III, IV, V, VI, VIIB, VIII, IX, XI, and the Vermis I, II, III, IV, V, VI, VII, VIII, IX, and X. Meanwhile, the frontal, temporal, parietal cortices, and precuneus were subdivided into 72 regions of interest (Supplementary Material Table S1 in S1 File). Using the updated_pons_vermis intensity normalization template provided in the SNBPI Toolbox, the pons was manually segmented in 3D Slicer (the version 5.8.1. link: https://slicer.Org. Supplementary material S1 File). The pons was used as the reference region to calculate the SUVRs for the 26 cerebellar subregions and 72 cortical regions under both non-corrected and PVEc-GTM corrected conditions [31]. The mean SUVRs for the 26 cerebellar regions in the CN, EMCI, LMCI, and AD groups were calculated under both PVEc and non-PVEc conditions (S2 and S3 Tables in S1 File). The formula for calculating the SUVR is as follows:

(1)

The cerebellar and cortical SUVR_pons were calculated using this formula (1).

Brain network construction

According to different data processing methods, We constructed individual-specific cerebellar and cortical deposition networks both with and without PVEc. In our parallel control study, a recently developed methodology was employed to construct individual-specific networks, thereby to obtain the individual-specific deposition network [25,32]. The construction process of individual-specific cerebellar deposition network without PVEc as follows: First, for CN subjects, a group-level cerebellar deposition network was constructed using the covariance method, serving as the reference network (). Specifically, a deposition network was constructed using all CN subjects. The nodes in the network were defined as 26 regions of the cerebellum, and the edges between each pair of nodes represented the correlation coefficient of the SUVRs between brain regions. Then, we added patients (EMCI, LMCI, AD) to the reference group, and a new cerebellar deposition network () was constructed using subjects ( reference subjects and 1 patient). Next, the difference between the perturbed network () and the reference network () was calculated as . Since was shown to follow a symmetric distribution similar to a normal distribution, referred to as the “volcanic distribution,” a Z-score transformation was applied to as follows [33]:

(2)

where is the number of subjects in the reference group. In the individual-specific cerebellar deposition network of each patient, the weight of each edge was the Z-score obtained from the above Equation (2). Additionally, Fisher’s Z transformation was performed to ensure normality. Similarly, CN subjects were added one by one to the reference group to obtain the individual-specific cerebellar deposition network for healthy controls.

To further analyze, the cerebellar deposition network matrix was converted into a binary matrix, considering only positive correlations. A sparsity threshold ranging from 0.05 (the minimum sparsity threshold estimated using the ‘gretna_get_rmax.m’ code in Gretna software) to 0.35 with a step size of 0.01, was applied to exclude spurious connections and ensure network integrity [34]. Subsequently, the Gretna toolbox was used to calculate ten network properties for the cerebellar networks [35]. These measures encompassed global characteristics such as small-world metrics (sigma, lambda, and gamma), CC, characteristic path length, global efficiency, and local efficiency. Additionally, nodal characteristics were considered, including degree centrality, betweenness centrality (BC), nodal CC, nodal efficiency, and nodal local efficiency. While global measures reflect the overall structure and functionality of the network, nodal measures pertain to specific cerebellar lobules within the network. Notably, BC refers to the number of shortest paths that pass through a given node in the network, serving as an indicator of the node’s significance and influence within the overall network structure.

Similarly, individual-specific cerebellar and cortical deposition networks with PVEc correction were constructed according to the above method (S4 and S6 Table in S1 File).

Statistical analyses

Statistical processing was implemented through SPSS 27.0 (IBM Corp) and MATLAB R2021a. Chi-square testing was employed to examine intergroup variations in categorical measures. The normality of continuous demographic and graph-theoretic variables was verified using Shapiro-Wilk tests. The selection between independent t-tests and Wilcoxon rank-sum tests for intergroup analysis was determined by normality test results. The Bonferroni correction was applied for multiple comparisons across regions of interest (ROIs). The number of comparisons was determined separately for each brain area: 26 cerebellar ROIs and 72 cortical ROIs. Therefore, the corrected significance thresholds were set to p < 0.05/ 26 for the cerebellar analyses and p < 0.05/ 72 for the cortical analyses. Network characteristics exhibiting non-normal distributions underwent log-transformation to meet normality assumptions for comparative analyses. Demographic variables (age and gender) were controlled as covariates in all analyses, with intergroup differences evaluated using two-tailed Student’s t-tests. Statistical significance was defined as achieving a p-value below the 0.05 threshold.

Model development, evaluation and interpretation

The classification features were extracted from individual data, including the region of interest SUVR_pons and individual-specific deposition network graph theory features, which encompassed both global and local attributes. To identify statistically significant features, t-tests and Mann- Whitney U tests were performed. Subsequently, the Least absolute shrinkage and selection operator (LASSO) was employed to identify the key graph theory features across the different groups. LASSO enhances model accuracy by shrinking the coefficients of less relevant features to zero, thereby improving feature selection. The optimal λ value was determined through 5-fold cross-validation by minimizing the validation error.

In this study, stratified random sampling was used to divide the dataset into two groups: a training set (containing 70% of the total data) and a test set (containing the remaining 30%). To further assess the model’s performance, six machine learning classifiers were employed: Support Vector Machine (SVM), Random Forest (RF), Logistic Regression (LR), Multilayer Perceptron (MLP), eXtreme Gradient Boosting (XGBoost), and k-Nearest Neighbors (k-NN). To identify the optimal parameters for these six models, a 5-fold cross-validation was performed for hyperparameter tuning. To ensure the validity of evaluation, the test set was not used during the model tuning phase. The specific parameters for each model are provided in S8 Table in S1 File. For a comprehensive evaluation of classification performance, several metrics were employed, including accuracy, specificity, sensitivity, Area Under the Curve (AUC), and Receiver Operating Characteristic (ROC) curves, and their corresponding 95% confidence intervals (CI). Only the model with the best classification results across different groups is presented in this study, while the performance of other models can be found in S9 Table in S1 File. The source code for this method will be freely available at https://github.com/LYRLJL/cerebellum_machine-learning.

Model interpretability with shapley additive explan

The Shapley additive explanations (SHAP) framework was used to quantify the feature-specific contributions of cerebellar network components and the SUVR of different cerebellar regions to model prediction accuracy. This approach, based on Shapley values derived from cooperative game theory, allows us to quantify the contribution of each feature to the predictions of the most effective model. Shapley values provide a measure of the importance of each feature in enhancing classification performance, and they are applicable to various feature types. The SHAP feature importance plot ranks attributes along the vertical axis by their contribution magnitude, with a chromatic scale (yellow denoting maximal values, purple indicating minimal values) encoding feature-value relationships.

Results

Demographic and clinical comparisons

Table 1 provides an overview of the demographic characteristics of all participants. The study sample included 172 individuals with CN status (98 males, 74 females; mean age 75.89 ± 7.97 years), 159 patients diagnosed with EMCI (78 males, 81 females; mean age 72.99 ± 7.04 years), 141 patients with LMCI (67 males, 74 females; mean age 72.75 ± 7.37 years), and 140 patients with AD (82 males, 58 females; mean age 74.41 ± 8.40 years). MMSE scores showed statistically significant variations among the study groups (p < 0.05). Significant differences in the mean values of PVEc, non-PVEc cerebellum, and mean cortical SUVR were observed across all group comparisons (CN vs. AD, CN vs. EMCI, EMCI vs. LMCI, LMCI vs. AD), except between the EMCI and LMCI groups.

Download:

Table 1. Presents the baseline demographic and clinical features of study participants.

https://doi.org/10.1371/journal.pone.0342738.t001

Changes in amyloid plaque deposition connectivity in CN to AD

Compared to the CN group, the AD group exhibited significant alterations in cerebellar connectivity (Table 2, Fig 3), specifically characterized by an increase in both average degree (1.463 vs. 1.489, p < 0.001; Table 2, Fig 3E) and global network efficiency (0.086 vs. 0.098, p < 0.001; Table 2, Fig 3G). Topological analysis revealed a significant increase in the cerebellar BC values in several regions when comparing AD patients with CN individuals. These regions included the left Crus I, Crus II, III, VIIB, VIII, IX, right Crus I, Crus II, VI, VIIB, VIII, X, as well as Vermis III, Vermis VI, and Vermis VII. (p < 0.05, Bonferroni corrected; Fig 2A). The cortical network metrics exhibited significant changes, with the small-world properties Sigma, Lambda, and Gamma values significantly decreased, while global network efficiency, local efficiency, average degree, and average CC significantly increased (S5 Table in S1 File). Moreover, significant changes in BC values were observed in multiple cortical regions (S1 Fig in S1 File).

Download:

Table 2. Comparison of amyloid accumulation connectivity alterations between CN, AD, EMCI, and LMCI.

https://doi.org/10.1371/journal.pone.0342738.t002

Download:

Fig 2. Cerebellar Graph Theory Analysis of CN, EMCI, LMCI, and AD Individuals.

Cerebellar graph theory analysis reveals significant differences in amyloid plaque deposition connectivity between the CN, EMCI, LMCI, and AD groups: (A) AD vs. CN group; (B) CN vs. EMCI group; (C) EMCI vs. LMCI group; (D) LMCI vs. AD group in terms of significant differences in betweenness centrality (p < 0.05, Bonferroni correction). The nodes represent 26 cerebellar lobules, with red and blue indicating regions with significantly increased and decreased intermediary centrality. L,left; R, right.

https://doi.org/10.1371/journal.pone.0342738.g002

Changes in amyloid plaque deposition connectivity in CN to EMCI

Statistical analyses revealed a significant increase in the small world property lambda in EMCI compared to CN (0.361 vs. 0.380, p = 0.011; Table 2, Fig 3B). Regarding local connectivity, region-specific alterations in cerebellar BC were identified in EMCI individuals relative to CN controls. Specifically, the BC values in the Vermis I and II lobules were significantly elevated, while the BC values in the right Crus I, right X, and left VI lobules were significantly decreased (p < 0.05, Bonferroni corrected; Fig 2B). The cortical network metrics showed significant increases in local efficiency, global efficiency, average CC values (S5 Table in S1 File). Moreover, significant changes in BC values were observed in multiple cortical regions (S2 Fig in S1 File).

Download:

Fig 3. Inter-group comparison of cerebellar amyloid plaque deposition connectivity in the study groups.

The data illustrate the distribution of changes in cerebellar network connectivity, including small-world properties, average degree, clustering coefficient, and centrality. P-values were obtained after correcting for multiple comparisons across the 26 cerebellar regions using false discovery rate methods (* p < 0.05; ** p < 0.01; *** p < 0.001). CN: cognitively normal, EMCI: early mild cognitive impairment, LMCI: late mild cognitive impairment, AD: Alzheimer’s disease, CC: clustering coefficient, BC: betweenness centrality.

https://doi.org/10.1371/journal.pone.0342738.g003

Changes in amyloid plaque deposition connectivity in EMCI to LMCI

Compared to the EMCI group, there were no significant changes in the overall cerebellar deposition connectivity in the LMCI group. Additionally, in terms of local connectivity, the LMCI group exhibited BC changes in only a few regions. Specifically, the cerebellar BC value in the right VIIB lobule was significantly decreased, while the BC values in the Vermis VI and Vermis VII lobules were significantly increased. (p < 0.05, Bonferroni corrected; Fig 2C). The cortical network metrics showed significant decreases in the small world property parameters (Lambda and Gamma values), while global efficiency values exhibited a significant increase (S5 Table in S1 File). Moreover, significant changes in BC values were observed in multiple cortical regions (S3 Fig in S1 File).

Changes in amyloid plaque deposition connectivity in LMCI to AD

Compared to the LMCI group, the AD group exhibited significant changes in cerebellar connectivity (Table 2, Fig 3). Specifically, the small world properties were significantly altered, with a decrease in sigma (0.330 vs. 0.311, p = 0.014; Fig 3A), lambda (0.369 vs. 0.348, p = 0.008; Fig 3B), gamma (0.418 vs. 0.371, p = 0.009; Fig 3C), local efficiency (0.162 vs. 0.156, p = 0.020; Fig 3D), and average CC (0.147 vs. 0.139, p = 0.006; Fig 3F). Additionally, the average degree (1.471 vs. 1.489, p = 0.022; Fig 3E) and global efficiency (0.088 vs. 0.098, p < 0.001; Fig 3G) were significantly increased. In terms of local connectivity, the AD group demonstrated significant differences in BC across multiple cerebellar regions compared to the LMCI group. Specifically, BC was significantly decreased in the left Crus I, Crus II, III, VIIB, VIII, IX, the right Crus II, III, VIIB, and bilateral Vermis III, Vermis VI, Vermis VII, and Vermis VIII (p < 0.05, Bonferroni corrected; Fig 2D). For cortical network metrics, in the small world properties Sigma, Lambda, and Gamma showed significant decreases, while local efficiency, global efficiency, and average CC values exhibited significant increases (S5 Table in S1 File). Moreover, significant changes in BC values were observed in multiple cortical regions (S4 Fig in S1 File).

Distinguishing performance of cerebellar network models and cortical brain models

In addition to the cerebellum model, a cortical brain model was construted by extracting cortical network feature and cortical SUVR features, Then the classification performance between the cerebellum and cortical models was compared, with their performance metrics presented in Table 3 and illustrated in Fig 4. The cerebellum model effectively distinguishes between AD and CN individuals, with XGBoost showing superior performance (S9 Table in S1 File). In the test group, the AUC was 0.950, accuracy was 0.862, sensitivity was 0.887, and specificity was 0.829. Similar results were observed in the training set (Table 3, Fig 4A). Furthermore, the cortical brain model outperformed the cerebellum model in classifying AD and CN, achieving an AUC of 0.989, accuracy of 0.926, sensitivity of 0.925, and specificity of 0.927 in the test set, with comparable performance in the training set (Table 3, Fig 4E). In the classification of CN vs. EMCI, EMCI vs. LMCI, and LMCI vs. AD, the LR classifier demonstrated stronger robustness in the cerebellar model. In the CN vs. EMCI test set, the AUC was 0.995, accuracy was 0.963, sensitivity was 0.936, and specificity was 0.981 (Table 3, Fig 4B). The cerebellum model performed better than the cortical model in distinguishing CN from EMCI (Table 3). In the EMCI vs. LMCI classification, the cortical brain model outperformed the cerebellum model. For the cerebellum model in the EMCI vs. LMCI classification, the AUC in the test set was 0.964, accuracy was 0.911, sensitivity was 0.913, and specificity was 0.909 (Table 3, Fig 4C). The cortical brain model, achieved an AUC of 0.997, accuracy of 0.956, sensitivity of 0.952, and specificity of 0.958 (Table 3, Fig 4G). In the LMCI vs. AD classification, the cerebellum model showed an AUC of 0.632, accuracy of 0.612, sensitivity of 0.698, and specificity of 0.523 in the test set. The cortical brain model showed similar performance to the cerebellum model (Table 3, Fig 4D, Fig 4H). These results suggest that the cerebellum model may have a comparative advantage in identifying MCI when compared with the cortical brain model.

Download:

Table 3. Classification performance of cerebellar graph-theoretical features for CN vs.AD, CN vs. EMCI, EMCI vs. LMCI, and LMCI vs. AD in the training and testing sets.

https://doi.org/10.1371/journal.pone.0342738.t003

Download:

Fig 4. Performance of the cerebellum model and cortical brain model on the training and testing sets for binary classifications: CN vs. AD, CN vs. EMCI, EMCI vs. LMCI, and LMCI vs. AD.

https://doi.org/10.1371/journal.pone.0342738.g004

SHAP analysis was employed to evaluate, the contribution of individual features to the discriminative power of the cerebellar model. The results revealed that the SUVR values of cerebellar regions play a crucial role in distinguishing between different cognitive states. We found that several cerebellar region SUVR features emerged as key contributors to the statistical differentiation of AD, CN, EMCI, and LMCI (Fig 5). For example, SUVR of CRBLIX.L, CRBLVII.R, and CRBLCruslIL were important contributors in multiple classifications task. Specifically, in the AD vs CN comparison, cerebellar-specific deposition network features including NE_CRBLVI.R, NCC_CRBLVI.R, NE_CRBLIX.R, NLE_CRBLIX.R, and NLE_CRBLIV.V.R played a significant role in the classification process (Fig 5A). These finding indicate that the SUVR features of cerebellar regions play a vital role in identifying and differentiating between the various stages of cognitive impairment.

Download:

Fig 5. SHAP summary dot plots of the top 10 features of the best-performing ML models in different groups.

https://doi.org/10.1371/journal.pone.0342738.g005

Discussion

This study analyzed changes in cerebellar amyloid plaque deposition connectivity networks during the progression of AD using ¹⁸F-AV45 PET imaging. Notably, this is the first study to investigate cerebellar amyloid plaque deposition connectivity changes based on PET images. The results showed significant differences in cerebellar amyloid SUVR values across the progression of cognitive decline. Graph theory analysis was used to investigate, changes in cerebellar amyloid plaques across different cognitive stages. Compared with the CN group, AD patients showed a significant increase in average degree and global efficiency (1.463 vs. 1.489, 0.086 vs. 0.098). For the CN and EMCI groups, the small-world property index Lambda was also significantly higher in the EMCI group than in the CN group (0.361 vs. 0.380). In the comparison between the LMCI and AD groups, all graph-theoretical metrics showed significant differences. These findings reveal significant alteration in both global and local cerebellar network connectivity in AD patients. To improve early AD classification, we combined graph theory analysis features of cerebellar amyloid plaque deposition connectivity networks with SUVR values of various cerebellar regions. As shown in Table 3, we integrated both local and global features, further enhanced the diagnostic performance of early AD. These results also provide new evidence for the role of the cerebellum in early identification of AD during its preclinical stage.

Amyloid PET imaging faces challenges in detecting brainstem and cerebellar involvement due to previous studies that used the cerebellum or pons as reference regions, excluding their own involvement from detection [36]. However, increasing evidence suggests amyloid plaque deposition in the cerebellum during AD progression. While this finding provides background for our study, further investigation is needed. Recent studies have confirmed the presence of amyloid plaques in the cerebellum of AD patients, with amyloid deposition in the cerebellum being a common feature of early-onset AD [5]. Our findings align with these studies, by using the pons as a reference region, we observed significant differences in cerebellar SUVR between the EMCI and CN groups, suggesting that amyloid plaque deposition begins early in AD, consistent with AD’s pathological features. Consistent with the pathological features of the disease. Additionally, higher baseline SUVR values, typically based on amyloid plaque load using mean cortical SUVR, have been associated with greater cognitive decline [37]. Furthermore, amyloid pathology correlates negatively with various cognitive functions, with the occipital and temporal-parietal regions mediating these relationships [21]. Our research on cerebellar amyloid SUVR aligns with studies in the brain, where amyloid SUVR is elevated in the cerebellum, corresponding to cognitive decline, albeit not significantly in the EMCI to LMCI stages. The unique nature of the cerebellum suggests that it appears to be unaffected by the amyloid-related neuropathological burden in the medial temporal lobe; however, it may still be susceptible to neurodegenerative changes driven by inflammation, metabolism, or other cellular processes. Given these considerations, caution is advised when selecting the cerebellum as a “reference” region or when including it in imaging or analysis in neuroimaging and histopathological studies of dementia [38].

Aβ deposition is closely related to disease progression and the severity of cognitive symptoms and can be used to monitor disease progression [39]. However, the spatial variability of Aβ deposition limits quantitative analysis using SUVR-based PET imaging. Topological analysis of Aβ deposition patterns may more accurately reflect the progression of AD. In this study, we investigated the ¹⁸F-AV45 PET images of different stages of AD from a topological perspective at both individual and group levels. By measuring Aβ accumulation across various cerebellar regions and applying graph theory analysis, a comprehensive assessment of both global and local features was conducted. During the procession of AD, compared to CN, the mean values of degree centrality and global efficiency were significantly increased in AD. However, no significant changes in degree centrality, CC, local efficiency, and global efficiency were observed in the comparisons between CN and EMCI, as well as EMCI and LMCI. BC reflects the importance and influence of a node within the entire network, and these significant changes were particularly evident in the following cerebellar lobules: left Crus I, Crus II, III, VIIB, VIII, IX, right Crus II, III, VIIB, and bilateral Vermis III, Vermis VI, Vermis VII, and Vermis VIII. These observations confirm recent hypotheses regarding the involvement of different functional areas within the cerebellum, particularly highlighting the role of cerebellar regions in cognitive processes [11]. The Cerebellar cognitive-emotional syndrome arises following injury to the regions of the cerebellum responsible for cognition and emotion. It primarily affects the vestibular cerebellum located in lobule X, along with the adjacent regions of lobule IX, and involves lobules Crus I, Crus II, VI, VIIA, VIIB, and vermis VII [40]. Emerging evidence highlights the critical contribution of vermal cerebellar subregions to AD pathogenesis. Beyond its well-established roles in maintaining motor coordination and postural stability, the cerebellar vermis distinctly participates in affective modulation via its extensive neural connections with limbic circuitry. The vermis, particularly the lobules associated with emotional processing, is crucial for emotion-related memory, and disruption of vermal regions impairs global brain connectivity [41]. The anterior vermis could be a component of the sensorimotor representation system. Meanwhile, the posterior vermis is typically associated with emotional and cognitive functions [42]. Individuals with damage to the posterior vermis frequently display impairments in working memory, in addition to experiencing various cognitive and emotional disturbances [43]. Decreased volume or activity in the vermis is often seen in individuals with neuropsychiatric disorders, including major depressive disorder in adolescents, social memory deficits, and future gait impairments in Parkinson’s disease [41,44,45]. Our findings suggest that the cerebellar vermis may also play a significant role in the progression of AD. We observed early changes in cerebellar amyloid plaque deposition connectivity, particularly in regions I and II of the vermis, with significant changes consistent with MRI results [17]. Our findings suggest that cerebellar graph theoretical features may provide valuable insights into the unique characteristics of cerebellar involvement in MCI and AD patients.

The predictive value of the cerebellum for mild cognitive impairment may be higher than the changes observed in the cortical regions of the brain. Compared to other advanced methods (Table 4), our model, based on cerebellar amyloid plaque deposition connectivity networks, graph theory analysis features, and SUVR of cerebellar regions, demonstrated exceptional performance in diagnosing MCI using machine learning methods. Specifically, in early-stage classification, the AUC for CN vs. EMCI reached 0.995, with an accuracy of 0.963, showcasing performance that surpasses most existing algorithms. For the EMCI vs. LMCI classification, the accuracy reached 0.911, with an AUC of 0.964, outperforming current leading methods (Table 4). The single-brain region analysis in our study significantly reduced unstable variables associated with multi-brain region analysis, simplified the training process, and enhanced early-stage classification performance. However, the model’s performance in the CN vs. AD and LMCI vs. AD classifications still needs further improvement. We further utilized the SHAP framework to assess the contribution of individual features to the interpretability of the AD diagnostic model. Overall, during the diagnostic process, graph theory features, particularly cerebellar SUVR_pons, indirectly demonstrated that cerebellar features are more pronounced in staging diagnosis than visual assessments. Additionally, the extracted graph theory features hold significant value in predicting individual treatment outcomes.

Download:

Table 4. Comparative experiment results.

https://doi.org/10.1371/journal.pone.0342738.t004

These results align with previous observations, indicating that the cerebellum is involved in cognitive deficits related to AD [4]. AD is typically characterized by cognitive impairment, and extensive research suggests that the cerebellum plays a critical role in generating and maintaining both cognitive and non-cognitive functions [46]. Almost the entire cerebellar region is involved in cognitive functions, and abnormalities in cerebellar structure and function can lead to impairments in both cognitive and non-cognitive functions. Cerebellar pathology in AD can be reliably detected and quantified using various imaging biomarkers, including MRI, PET, and electrophysiological measures. MRI studies have revealed that AD may alter both the structure and function of the cerebellum, with findings such as gray matter atrophy, volumetric changes, radiological features, and topological network disruptions observed across different stages of disease severity [17,46].

Prior studies have shown that electrophysiological changes in the cerebellum of AD-related APP/PS1 mice precede any pathological changes [53]. ¹⁸F-FDG PET studies have identified cerebellar metabolic dysfunction in AD, observing an increase in cerebellar metabolism during the transition from MCI to AD, with hypermetabolism detected specifically in the AD cerebellar regions [54]. The above studies underscore the critical role of the cerebellum in cognitive impairments associated with AD. These AD biomarkers vary in terms of accessibility, cost, invasiveness, and comprehensiveness. While they may represent the most reliable and information-rich methods, ¹⁸F-AV45 PET uniquely detects Aβ pathology within the cerebellum, providing crucial insights into the underlying disease mechanisms [55,56]. Pathological changes associated with AD can be detected even during the preclinical or EMCI stages, with graph theory-based analysis showing strong predictive performance. A more detailed quantitative assessment of connectivity changes in amyloid plaque deposition within individual cerebellar lobules is essential for understanding AD progression.

This study has several limitations. First, we used the pons as the reference region, while alternative reference regions may yield different results. Future studies should perform sensitivity analyses using results from different reference regions to assess robustness. Second, since ADNI is a multi-center dataset, site-related differences may influence graph-theoretical analyses and machine learning classification. Therefore, future work should validate the findings across multiple centers to ensure consistency and reliability. Third, future research will need to conduct additional validation analyses using external datasets. Fourth, the differences in the timing between MRI and PET scans may impact the study results. Future research will further consider this potential influence.

Conclusion

In conclusion, using the ¹⁸F-AV45 PET imaging dataset, our findings highlight the presence of abnormal cerebellar amyloid plaque deposition connectivity networks in the early stages of AD. As cognitive impairment deepens and progresses into AD, alterations in the cerebellar amyloid plaque deposition connectivity network become increasingly severe, affecting both global and local cerebellar connectivity. These changes appear early on, which may aid in the early detection of AD.

Supporting information

S1 File. Supplementary Methods for Early Diagnosis of Alzheimer’s Disease: Graph Theoretical Analysis of Cerebellar Network Features Based on 18F-AV45 PET.

https://doi.org/10.1371/journal.pone.0342738.s001

(DOCX)

Acknowledgments

Data used in preparation of this article were obtained from the Alzheimer’s Disease Neuroimaging Initiative (ADNI) database (adni.loni. usc.edu). As such, the investigators within the ADNI contributed to the design and implementation of ADNI and/or provided data but did not participate in analysis or writing of this report. A complete listing of ADNI investigators can be found at: http://adni.loni.usc.edu/wp-content/uploads/how_to_apply/ADNI_Acknowledgement_List.pdf

References

1. Jack CR, Andrews SJ, Beach TG, Buracchio T, Dunn B, Graf A, et al. Revised criteria for the diagnosis and staging of Alzheimer’s disease. Nat Med. 2024;30(8):2121–4. pmid:38942991
- View Article
- PubMed/NCBI
- Google Scholar
2. Chen P, Zhao K, Zhang H, Wei Y, Wang P, Wang D. Altered global signal topography in Alzheimer’s disease. 2023.
3. Jacobs HIL, Van Boxtel MPJ, Jolles J, Verhey FRJ, Uylings HBM. Parietal cortex matters in Alzheimer’s disease: an overview of structural, functional and metabolic findings. Neurosci Biobehav Rev. 2012;36(1):297–309. pmid:21741401
- View Article
- PubMed/NCBI
- Google Scholar
4. Liu G, Yang C, Wang X, Chen X, Cai H, Le W. Cerebellum in neurodegenerative diseases: Advances, challenges, and prospects. iScience. 2024;27(11):111194. pmid:39555407
- View Article
- PubMed/NCBI
- Google Scholar
5. Jacobs HIL, Hopkins DA, Mayrhofer HC, Bruner E, van Leeuwen FW, Raaijmakers W, et al. The cerebellum in Alzheimer’s disease: Evaluating its role in cognitive decline. Brain. 2018;141(1):37–47. pmid:29053771
- View Article
- PubMed/NCBI
- Google Scholar
6. Bernard JA. Don’t forget the little brain: A framework for incorporating the cerebellum into the understanding of cognitive aging. Neurosci Biobehav Rev. 2022;137:104639. pmid:35346747
- View Article
- PubMed/NCBI
- Google Scholar
7. Um YH, Wang S-M, Kang DW, Kim S, Kim S, Kim D, et al. Cerebellar subregional structural changes across the Alzheimer’s disease continuum: A longitudinal analysis of cognitive and behavioural correlates. Brain Commun. 2025;8(1):fcaf500. pmid:41523195
- View Article
- PubMed/NCBI
- Google Scholar
8. Lopez G, Magaki SD, Williams CK, Paganini-Hill A, Vinters HV. Characterization of cerebellar amyloid-β deposits in Alzheimer disease. J Neuropathol Exp Neurol. 2024;83(2):72–8. pmid:38114098
- View Article
- PubMed/NCBI
- Google Scholar
9. Aragão Gomes L, Uytterhoeven V, Lopez-Sanmartin D, Tomé SO, Tousseyn T, Vandenberghe R, et al. Maturation of neuronal AD-tau pathology involves site-specific phosphorylation of cytoplasmic and synaptic tau preceding conformational change and fibril formation. Acta Neuropathol. 2021;141(2):173–92. pmid:33427938
- View Article
- PubMed/NCBI
- Google Scholar
10. Andersen K, Andersen BB, Pakkenberg B. Stereological quantification of the cerebellum in patients with Alzheimer’s disease. Neurobiol Aging. 2012;33(1):197.e11-20. pmid:20728248
- View Article
- PubMed/NCBI
- Google Scholar
11. Yang C, Liu G, Chen X, Le W. Cerebellum in Alzheimer’s disease and other neurodegenerative diseases: An emerging research frontier. MedComm (2020). 2024;5(7):e638. pmid:39006764
- View Article
- PubMed/NCBI
- Google Scholar
12. Baghirova N, Vũ DT, Can DC, Diaz CS, Bodlet J, Blanc G. Explainable graph-theoretical machine learning: with application to Alzheimer’s disease prediction. 2025.
13. Tijms BM, Wink AM, de Haan W, van der Flier WM, Stam CJ, Scheltens P, et al. Alzheimer’s disease: connecting findings from graph theoretical studies of brain networks. Neurobiol Aging. 2013;34(8):2023–36. pmid:23541878
- View Article
- PubMed/NCBI
- Google Scholar
14. Miraglia F, Vecchio F, Pappalettera C, Nucci L, Cotelli M, Judica E, et al. Brain Connectivity And Graph Theory Analysis in Alzheimer’s and Parkinson’s Disease: The contribution of electrophysiological techniques. Brain Sci. 2022;12(3):402. pmid:35326358
- View Article
- PubMed/NCBI
- Google Scholar
15. Vecchio F, Miraglia F, Maria Rossini P. Connectome: Graph theory application in functional brain network architecture. Clin Neurophysiol Pract. 2017;2:206–13. pmid:30214997
- View Article
- PubMed/NCBI
- Google Scholar
16. Zhao K, Zheng Q, Dyrba M, Rittman T, Li A, Che T, et al. Regional radiomics similarity networks reveal distinct subtypes and abnormality patterns in mild cognitive impairment. Adv Sci (Weinh). 2022;9(12):e2104538. pmid:35098696
- View Article
- PubMed/NCBI
- Google Scholar
17. Chen Y, Qi Y, Hu Y, Qiu X, Qiu T, Li S, et al. Integrated cerebellar radiomic-network model for predicting mild cognitive impairment in Alzheimer’s disease. Alzheimers Dement. 2025;21(1):e14361. pmid:39535490
- View Article
- PubMed/NCBI
- Google Scholar
18. Bullmore E, Sporns O. Complex brain networks: Graph theoretical analysis of structural and functional systems. Nat Rev Neurosci. 2009;10(3):186–98. pmid:19190637
- View Article
- PubMed/NCBI
- Google Scholar
19. Rabinovici GD, Gatsonis C, Apgar C, Chaudhary K, Gareen I, Hanna L, et al. Association of amyloid positron emission tomography with subsequent change in clinical management among medicare beneficiaries with mild cognitive impairment or dementia. JAMA. 2019;321(13):1286–94. pmid:30938796
- View Article
- PubMed/NCBI
- Google Scholar
20. Johnson KA, Fox NC, Sperling RA, Klunk WE. Brain imaging in Alzheimer disease. Cold Spring Harb Perspect Med. 2012;2(4):a006213. pmid:22474610
- View Article
- PubMed/NCBI
- Google Scholar
21. Chen H-J, Zhang M, Wei M, Yu X, Wang Y, Yang J, et al. Amyloid pathology related to aberrant structure-function coupling of brain networks in Alzheimer’s disease: Insights from [18F]-florbetapir PET imaging. Eur J Nucl Med Mol Imaging. 2025;52(8):2929–40. pmid:40053120
- View Article
- PubMed/NCBI
- Google Scholar
22. Veronese M, Moro L, Arcolin M, Dipasquale O, Rizzo G, Expert P, et al. Covariance statistics and network analysis of brain PET imaging studies. Sci Rep. 2019;9(1):2496. pmid:30792460
- View Article
- PubMed/NCBI
- Google Scholar
23. Galli A, Inglese M, Presotto L, Malito R, Di X, Toschi N, et al. Glucose metabolism in hyper-connected regions predicts neurodegeneration and speed of conversion in Alzheimer’s disease. Eur J Nucl Med Mol Imaging. 2025;52(12):4639–51. pmid:40471318
- View Article
- PubMed/NCBI
- Google Scholar
24. Shim H-K, Lee H-J, Kim SE, Lee BI, Park S, Park KM. Alterations in the metabolic networks of temporal lobe epilepsy patients: A graph theoretical analysis using FDG-PET. Neuroimage Clin. 2020;27:102349. pmid:32702626
- View Article
- PubMed/NCBI
- Google Scholar
25. Lu W, Song T, Li J, Zhang Y, Lu J. Individual-specific metabolic network based on 18F-FDG PET revealing multi-level aberrant metabolisms in Parkinson’s disease. Hum Brain Mapp. 2024;45(14):e70026. pmid:39300894
- View Article
- PubMed/NCBI
- Google Scholar
26. Ruan D, Sun L. Amyloid-β PET in Alzheimer’s disease: A systematic review and Bayesian meta-analysis. Brain Behav. 2023;13(1):e2850. pmid:36573329
- View Article
- PubMed/NCBI
- Google Scholar
27. Ghisays V, Lopera F, Goradia DD, Protas HD, Malek-Ahmadi MH, Chen Y, et al. PET evidence of preclinical cerebellar amyloid plaque deposition in autosomal dominant Alzheimer’s disease-causing Presenilin-1 E280A mutation carriers. Neuroimage Clin. 2021;31:102749. pmid:34252876
- View Article
- PubMed/NCBI
- Google Scholar
28. Gonzalez-Escamilla G, Lange C, Teipel S, Buchert R, Grothe MJ. Alzheimer’s Disease Neuroimaging Initiative. PETPVE12: an SPM toolbox for Partial Volume Effects correction in brain PET - Application to amyloid imaging with AV45-PET. Neuroimage. 2017;147:669–77. pmid:28039094
- View Article
- PubMed/NCBI
- Google Scholar
29. Zhang T, Nie B, Liu H, Shan B. Alzheimer’s Disease Neuroimaging Initiative. Unified spatial normalization method of brain PET images using adaptive probabilistic brain atlas. Eur J Nucl Med Mol Imaging. 2022;49(9):3073–85. pmid:35258689
- View Article
- PubMed/NCBI
- Google Scholar
30. Klunk WE, Koeppe RA, Price JC, Benzinger TL, Devous MD, Jagust WJ, et al. The Centiloid Project: standardizing quantitative amyloid plaque estimation by PET. Alzheimers Dement. 2015;11(1):1-15.e1-4. pmid:25443857
- View Article
- PubMed/NCBI
- Google Scholar
31. Li Y, Ng YL, Paranjpe MD, Ge Q, Gu F, Li P, et al. Tracer-specific reference tissues selection improves detection of 18 F-FDG, 18 F-florbetapir, and 18 F-flortaucipir PET SUVR changes in Alzheimer’s disease. Hum Brain Mapp. 2022;43(7):2121–33. pmid:35165964
- View Article
- PubMed/NCBI
- Google Scholar
32. Liu Z, Palaniyappan L, Wu X, Zhang K, Du J, Zhao Q, et al. Resolving heterogeneity in schizophrenia through a novel systems approach to brain structure: Individualized structural covariance network analysis. Mol Psychiatry. 2021;26(12):7719–31. pmid:34316005
- View Article
- PubMed/NCBI
- Google Scholar
33. Liu X, Wang Y, Ji H, Aihara K, Chen L. Personalized characterization of diseases using sample-specific networks. Nucleic Acids Res. 2016;44(22):e164. pmid:27596597
- View Article
- PubMed/NCBI
- Google Scholar
34. Cheng S, Zeng F, Zhou J, Dong X, Yang W, Yin T, et al. Altered static and dynamic functional brain network in knee osteoarthritis: A resting-state functional magnetic resonance imaging study: Static and dynamic FNC in KOA. Neuroimage. 2024;292:120599. pmid:38608799
- View Article
- PubMed/NCBI
- Google Scholar
35. Wang J, Wang X, Xia M, Liao X, Evans A, He Y. GRETNA: A graph theoretical network analysis toolbox for imaging connectomics. Front Hum Neurosci. 2015;9:386. pmid:26175682
- View Article
- PubMed/NCBI
- Google Scholar
36. Thal DR, Vandenberghe R. Monitoring the progression of Alzheimer’s disease with τ-PET. Brain. 2016;139(Pt 5):1318–20. pmid:27189579
- View Article
- PubMed/NCBI
- Google Scholar
37. Farrell ME, Kennedy KM, Rodrigue KM, Wig G, Bischof GN, Rieck JR, et al. Association of longitudinal cognitive decline with amyloid burden in middle-aged and older adults: Evidence for a dose-response relationship. JAMA Neurol. 2017;74(7):830–8. pmid:28558099
- View Article
- PubMed/NCBI
- Google Scholar
38. Samstag CL, Chapman NH, Gibbons LE, Geller J, Loeb N, Dharap S, et al. Neuropathological correlates of vulnerability and resilience in the cerebellum in Alzheimer’s disease. Alzheimers Dement. 2025;21(2):e14428. pmid:39713867
- View Article
- PubMed/NCBI
- Google Scholar
39. Jagust WJ, Teunissen CE, DeCarli C. The complex pathway between amyloid β and cognition: Implications for therapy. Lancet Neurol. 2023;22(9):847–57. pmid:37454670
- View Article
- PubMed/NCBI
- Google Scholar
40. Schmahmann JD. Cerebellum in Alzheimer’s disease and frontotemporal dementia: Not a silent bystander. Brain. 2016;139(Pt 5):1314–8. pmid:27189578
- View Article
- PubMed/NCBI
- Google Scholar
41. Chao OY, Pathak SS, Zhang H, Augustine GJ, Christie JM, Kikuchi C, et al. Social memory deficit caused by dysregulation of the cerebellar vermis. Nat Commun. 2023;14(1):6007. pmid:37752149
- View Article
- PubMed/NCBI
- Google Scholar
42. Krienen FM, Buckner RL. Segregated fronto-cerebellar circuits revealed by intrinsic functional connectivity. Cereb Cortex. 2009;19(10):2485–97. pmid:19592571
- View Article
- PubMed/NCBI
- Google Scholar
43. Schmahmann JD, Sherman JC. The cerebellar cognitive affective syndrome. Brain. 1998;121 ( Pt 4):561–79. pmid:9577385
- View Article
- PubMed/NCBI
- Google Scholar
44. Maiti B, Rawson KS, Tanenbaum AB, Koller JM, Snyder AZ, Campbell MC, et al. Functional connectivity of vermis correlates with future gait impairments in Parkinson’s Disease. Mov Disord. 2021;36(11):2559–68. pmid:34109682
- View Article
- PubMed/NCBI
- Google Scholar
45. Wang L, Zhao P, Zhang J, Zhang R, Liu J, Duan J, et al. Functional connectivity between the cerebellar vermis and cerebrum distinguishes early treatment response for major depressive episodes in adolescents. J Affect Disord. 2023;339:256–63. pmid:37437740
- View Article
- PubMed/NCBI
- Google Scholar
46. Lin Y, Kuo S-H. The emerging role of the cerebellum in neurodegeneration linked to cognitive impairment. J Alzheimers Dis. 2024;102(1):30–2. pmid:39497298
- View Article
- PubMed/NCBI
- Google Scholar
47. de Mendonça LJC, Ferrari RJ. Alzheimer’s disease classification based on graph kernel SVMs constructed with 3D texture features extracted from MR images. Expert Systems with Applications. 2023;211:118633.
- View Article
- Google Scholar
48. Lei B, Li Y, Fu W, Yang P, Chen S, Wang T, et al. Alzheimer’s disease diagnosis from multi-modal data via feature inductive learning and dual multilevel graph neural network. Med Image Anal. 2024;97:103213. pmid:38850625
- View Article
- PubMed/NCBI
- Google Scholar
49. Chen Z, Bi S, Shan Y, Cui B, Yang H, Qi Z, et al. Multiparametric hippocampal signatures for early diagnosis of Alzheimer’s disease using 18F-FDG PET/MRI Radiomics. CNS Neurosci Ther. 2024;30(4):e14539. pmid:38031997
- View Article
- PubMed/NCBI
- Google Scholar
50. Lin A, Chen Y, Chen Y, Ye Z, Luo W, Chen Y, et al. MRI radiomics combined with machine learning for diagnosing mild cognitive impairment: A focus on the cerebellar gray and white matter. Front Aging Neurosci. 2024;16:1460293. pmid:39430972
- View Article
- PubMed/NCBI
- Google Scholar
51. Feng J, Jiang M, Zhang H, Yin L. Integrating imaging and genetic data via wavelet transform-based CNN for Alzheimer ’s disease classification. Biomedical Signal Processing and Control. 2025;104:107583.
- View Article
- Google Scholar
52. Abdelaziz M, Wang T, Anwaar W, Elazab A. Multi-scale multimodal deep learning framework for Alzheimer’s disease diagnosis. Comput Biol Med. 2025;184:109438. pmid:39579666
- View Article
- PubMed/NCBI
- Google Scholar
53. Yu H, Wang M, Yang Q, Xu X, Zhang R, Chen X, et al. The electrophysiological and neuropathological profiles of cerebellum in APPswe /PS1ΔE9 mice: A hypothesis on the role of cerebellum in Alzheimer’s disease. Alzheimers Dement. 2023;19(6):2365–75. pmid:36469008
- View Article
- PubMed/NCBI
- Google Scholar
54. Gupta V, Booth S, Ko JH. Hypermetabolic cerebellar connectome in Alzheimer’s Disease. Brain Connect. 2023;13(6):356–66. pmid:34155909
- View Article
- PubMed/NCBI
- Google Scholar
55. Garland EF, Dennett O, Lau LC, Chatelet DS, Bottlaender M, Nicoll JAR, et al. The mitochondrial protein TSPO in Alzheimer’s disease: relation to the severity of AD pathology and the neuroinflammatory environment. J Neuroinflammation. 2023;20(1):186. pmid:37580767
- View Article
- PubMed/NCBI
- Google Scholar
56. Shimada H, Minatani S, Takeuchi J, Takeda A, Kawabe J, Wada Y, et al. Heavy tau burden with subtle amyloid β accumulation in the cerebral cortex and cerebellum in a case of familial Alzheimer’s disease with APP Osaka Mutation. Int J Mol Sci. 2020;21(12):4443. pmid:32580499
- View Article
- PubMed/NCBI
- Google Scholar

[ref1] 1. Jack CR, Andrews SJ, Beach TG, Buracchio T, Dunn B, Graf A, et al. Revised criteria for the diagnosis and staging of Alzheimer’s disease. Nat Med. 2024;30(8):2121–4. pmid:38942991
View Article
PubMed/NCBI
Google Scholar

[2] View Article

[3] PubMed/NCBI

[4] Google Scholar

[ref2] 2. Chen P, Zhao K, Zhang H, Wei Y, Wang P, Wang D. Altered global signal topography in Alzheimer’s disease. 2023.

[ref3] 3. Jacobs HIL, Van Boxtel MPJ, Jolles J, Verhey FRJ, Uylings HBM. Parietal cortex matters in Alzheimer’s disease: an overview of structural, functional and metabolic findings. Neurosci Biobehav Rev. 2012;36(1):297–309. pmid:21741401
View Article
PubMed/NCBI
Google Scholar

[7] View Article

[8] PubMed/NCBI

[9] Google Scholar

[ref4] 4. Liu G, Yang C, Wang X, Chen X, Cai H, Le W. Cerebellum in neurodegenerative diseases: Advances, challenges, and prospects. iScience. 2024;27(11):111194. pmid:39555407
View Article
PubMed/NCBI
Google Scholar

[11] View Article

[12] PubMed/NCBI

[13] Google Scholar

[ref5] 5. Jacobs HIL, Hopkins DA, Mayrhofer HC, Bruner E, van Leeuwen FW, Raaijmakers W, et al. The cerebellum in Alzheimer’s disease: Evaluating its role in cognitive decline. Brain. 2018;141(1):37–47. pmid:29053771
View Article
PubMed/NCBI
Google Scholar

[15] View Article

[16] PubMed/NCBI

[17] Google Scholar

[ref6] 6. Bernard JA. Don’t forget the little brain: A framework for incorporating the cerebellum into the understanding of cognitive aging. Neurosci Biobehav Rev. 2022;137:104639. pmid:35346747
View Article
PubMed/NCBI
Google Scholar

[19] View Article

[20] PubMed/NCBI

[21] Google Scholar

[ref7] 7. Um YH, Wang S-M, Kang DW, Kim S, Kim S, Kim D, et al. Cerebellar subregional structural changes across the Alzheimer’s disease continuum: A longitudinal analysis of cognitive and behavioural correlates. Brain Commun. 2025;8(1):fcaf500. pmid:41523195
View Article
PubMed/NCBI
Google Scholar

[23] View Article

[24] PubMed/NCBI

[25] Google Scholar

[ref8] 8. Lopez G, Magaki SD, Williams CK, Paganini-Hill A, Vinters HV. Characterization of cerebellar amyloid-β deposits in Alzheimer disease. J Neuropathol Exp Neurol. 2024;83(2):72–8. pmid:38114098
View Article
PubMed/NCBI
Google Scholar

[27] View Article

[28] PubMed/NCBI

[29] Google Scholar

[ref9] 9. Aragão Gomes L, Uytterhoeven V, Lopez-Sanmartin D, Tomé SO, Tousseyn T, Vandenberghe R, et al. Maturation of neuronal AD-tau pathology involves site-specific phosphorylation of cytoplasmic and synaptic tau preceding conformational change and fibril formation. Acta Neuropathol. 2021;141(2):173–92. pmid:33427938
View Article
PubMed/NCBI
Google Scholar

[31] View Article

[32] PubMed/NCBI

[33] Google Scholar

[ref10] 10. Andersen K, Andersen BB, Pakkenberg B. Stereological quantification of the cerebellum in patients with Alzheimer’s disease. Neurobiol Aging. 2012;33(1):197.e11-20. pmid:20728248
View Article
PubMed/NCBI
Google Scholar

[35] View Article

[36] PubMed/NCBI

[37] Google Scholar

[ref11] 11. Yang C, Liu G, Chen X, Le W. Cerebellum in Alzheimer’s disease and other neurodegenerative diseases: An emerging research frontier. MedComm (2020). 2024;5(7):e638. pmid:39006764
View Article
PubMed/NCBI
Google Scholar

[39] View Article

[40] PubMed/NCBI

[41] Google Scholar

[ref12] 12. Baghirova N, Vũ DT, Can DC, Diaz CS, Bodlet J, Blanc G. Explainable graph-theoretical machine learning: with application to Alzheimer’s disease prediction. 2025.

[ref13] 13. Tijms BM, Wink AM, de Haan W, van der Flier WM, Stam CJ, Scheltens P, et al. Alzheimer’s disease: connecting findings from graph theoretical studies of brain networks. Neurobiol Aging. 2013;34(8):2023–36. pmid:23541878
View Article
PubMed/NCBI
Google Scholar

[44] View Article

[45] PubMed/NCBI

[46] Google Scholar

[ref14] 14. Miraglia F, Vecchio F, Pappalettera C, Nucci L, Cotelli M, Judica E, et al. Brain Connectivity And Graph Theory Analysis in Alzheimer’s and Parkinson’s Disease: The contribution of electrophysiological techniques. Brain Sci. 2022;12(3):402. pmid:35326358
View Article
PubMed/NCBI
Google Scholar

[48] View Article

[49] PubMed/NCBI

[50] Google Scholar

[ref15] 15. Vecchio F, Miraglia F, Maria Rossini P. Connectome: Graph theory application in functional brain network architecture. Clin Neurophysiol Pract. 2017;2:206–13. pmid:30214997
View Article
PubMed/NCBI
Google Scholar

[52] View Article

[53] PubMed/NCBI

[54] Google Scholar

[ref16] 16. Zhao K, Zheng Q, Dyrba M, Rittman T, Li A, Che T, et al. Regional radiomics similarity networks reveal distinct subtypes and abnormality patterns in mild cognitive impairment. Adv Sci (Weinh). 2022;9(12):e2104538. pmid:35098696
View Article
PubMed/NCBI
Google Scholar

[56] View Article

[57] PubMed/NCBI

[58] Google Scholar

[ref17] 17. Chen Y, Qi Y, Hu Y, Qiu X, Qiu T, Li S, et al. Integrated cerebellar radiomic-network model for predicting mild cognitive impairment in Alzheimer’s disease. Alzheimers Dement. 2025;21(1):e14361. pmid:39535490
View Article
PubMed/NCBI
Google Scholar

[60] View Article

[61] PubMed/NCBI

[62] Google Scholar

[ref18] 18. Bullmore E, Sporns O. Complex brain networks: Graph theoretical analysis of structural and functional systems. Nat Rev Neurosci. 2009;10(3):186–98. pmid:19190637
View Article
PubMed/NCBI
Google Scholar

[64] View Article

[65] PubMed/NCBI

[66] Google Scholar

[ref19] 19. Rabinovici GD, Gatsonis C, Apgar C, Chaudhary K, Gareen I, Hanna L, et al. Association of amyloid positron emission tomography with subsequent change in clinical management among medicare beneficiaries with mild cognitive impairment or dementia. JAMA. 2019;321(13):1286–94. pmid:30938796
View Article
PubMed/NCBI
Google Scholar

[68] View Article

[69] PubMed/NCBI

[70] Google Scholar

[ref20] 20. Johnson KA, Fox NC, Sperling RA, Klunk WE. Brain imaging in Alzheimer disease. Cold Spring Harb Perspect Med. 2012;2(4):a006213. pmid:22474610
View Article
PubMed/NCBI
Google Scholar

[72] View Article

[73] PubMed/NCBI

[74] Google Scholar

[ref21] 21. Chen H-J, Zhang M, Wei M, Yu X, Wang Y, Yang J, et al. Amyloid pathology related to aberrant structure-function coupling of brain networks in Alzheimer’s disease: Insights from [18F]-florbetapir PET imaging. Eur J Nucl Med Mol Imaging. 2025;52(8):2929–40. pmid:40053120
View Article
PubMed/NCBI
Google Scholar

[76] View Article

[77] PubMed/NCBI

[78] Google Scholar

[ref22] 22. Veronese M, Moro L, Arcolin M, Dipasquale O, Rizzo G, Expert P, et al. Covariance statistics and network analysis of brain PET imaging studies. Sci Rep. 2019;9(1):2496. pmid:30792460
View Article
PubMed/NCBI
Google Scholar

[80] View Article

[81] PubMed/NCBI

[82] Google Scholar

[ref23] 23. Galli A, Inglese M, Presotto L, Malito R, Di X, Toschi N, et al. Glucose metabolism in hyper-connected regions predicts neurodegeneration and speed of conversion in Alzheimer’s disease. Eur J Nucl Med Mol Imaging. 2025;52(12):4639–51. pmid:40471318
View Article
PubMed/NCBI
Google Scholar

[84] View Article

[85] PubMed/NCBI

[86] Google Scholar

[ref24] 24. Shim H-K, Lee H-J, Kim SE, Lee BI, Park S, Park KM. Alterations in the metabolic networks of temporal lobe epilepsy patients: A graph theoretical analysis using FDG-PET. Neuroimage Clin. 2020;27:102349. pmid:32702626
View Article
PubMed/NCBI
Google Scholar

[88] View Article

[89] PubMed/NCBI

[90] Google Scholar

[ref25] 25. Lu W, Song T, Li J, Zhang Y, Lu J. Individual-specific metabolic network based on 18F-FDG PET revealing multi-level aberrant metabolisms in Parkinson’s disease. Hum Brain Mapp. 2024;45(14):e70026. pmid:39300894
View Article
PubMed/NCBI
Google Scholar

[92] View Article

[93] PubMed/NCBI

[94] Google Scholar

[ref26] 26. Ruan D, Sun L. Amyloid-β PET in Alzheimer’s disease: A systematic review and Bayesian meta-analysis. Brain Behav. 2023;13(1):e2850. pmid:36573329
View Article
PubMed/NCBI
Google Scholar

[96] View Article

[97] PubMed/NCBI

[98] Google Scholar

[ref27] 27. Ghisays V, Lopera F, Goradia DD, Protas HD, Malek-Ahmadi MH, Chen Y, et al. PET evidence of preclinical cerebellar amyloid plaque deposition in autosomal dominant Alzheimer’s disease-causing Presenilin-1 E280A mutation carriers. Neuroimage Clin. 2021;31:102749. pmid:34252876
View Article
PubMed/NCBI
Google Scholar

[100] View Article

[101] PubMed/NCBI

[102] Google Scholar

[ref28] 28. Gonzalez-Escamilla G, Lange C, Teipel S, Buchert R, Grothe MJ. Alzheimer’s Disease Neuroimaging Initiative. PETPVE12: an SPM toolbox for Partial Volume Effects correction in brain PET - Application to amyloid imaging with AV45-PET. Neuroimage. 2017;147:669–77. pmid:28039094
View Article
PubMed/NCBI
Google Scholar

[104] View Article

[105] PubMed/NCBI

[106] Google Scholar

[ref29] 29. Zhang T, Nie B, Liu H, Shan B. Alzheimer’s Disease Neuroimaging Initiative. Unified spatial normalization method of brain PET images using adaptive probabilistic brain atlas. Eur J Nucl Med Mol Imaging. 2022;49(9):3073–85. pmid:35258689
View Article
PubMed/NCBI
Google Scholar

[108] View Article

[109] PubMed/NCBI

[110] Google Scholar

[ref30] 30. Klunk WE, Koeppe RA, Price JC, Benzinger TL, Devous MD, Jagust WJ, et al. The Centiloid Project: standardizing quantitative amyloid plaque estimation by PET. Alzheimers Dement. 2015;11(1):1-15.e1-4. pmid:25443857
View Article
PubMed/NCBI
Google Scholar

[112] View Article

[113] PubMed/NCBI

[114] Google Scholar

[ref31] 31. Li Y, Ng YL, Paranjpe MD, Ge Q, Gu F, Li P, et al. Tracer-specific reference tissues selection improves detection of 18 F-FDG, 18 F-florbetapir, and 18 F-flortaucipir PET SUVR changes in Alzheimer’s disease. Hum Brain Mapp. 2022;43(7):2121–33. pmid:35165964
View Article
PubMed/NCBI
Google Scholar

[116] View Article

[117] PubMed/NCBI

[118] Google Scholar

[ref32] 32. Liu Z, Palaniyappan L, Wu X, Zhang K, Du J, Zhao Q, et al. Resolving heterogeneity in schizophrenia through a novel systems approach to brain structure: Individualized structural covariance network analysis. Mol Psychiatry. 2021;26(12):7719–31. pmid:34316005
View Article
PubMed/NCBI
Google Scholar

[120] View Article

[121] PubMed/NCBI

[122] Google Scholar

[ref33] 33. Liu X, Wang Y, Ji H, Aihara K, Chen L. Personalized characterization of diseases using sample-specific networks. Nucleic Acids Res. 2016;44(22):e164. pmid:27596597
View Article
PubMed/NCBI
Google Scholar

[124] View Article

[125] PubMed/NCBI

[126] Google Scholar

[ref34] 34. Cheng S, Zeng F, Zhou J, Dong X, Yang W, Yin T, et al. Altered static and dynamic functional brain network in knee osteoarthritis: A resting-state functional magnetic resonance imaging study: Static and dynamic FNC in KOA. Neuroimage. 2024;292:120599. pmid:38608799
View Article
PubMed/NCBI
Google Scholar

[128] View Article

[129] PubMed/NCBI

[130] Google Scholar

[ref35] 35. Wang J, Wang X, Xia M, Liao X, Evans A, He Y. GRETNA: A graph theoretical network analysis toolbox for imaging connectomics. Front Hum Neurosci. 2015;9:386. pmid:26175682
View Article
PubMed/NCBI
Google Scholar

[132] View Article

[133] PubMed/NCBI

[134] Google Scholar

[ref36] 36. Thal DR, Vandenberghe R. Monitoring the progression of Alzheimer’s disease with τ-PET. Brain. 2016;139(Pt 5):1318–20. pmid:27189579
View Article
PubMed/NCBI
Google Scholar

[136] View Article

[137] PubMed/NCBI

[138] Google Scholar

[ref37] 37. Farrell ME, Kennedy KM, Rodrigue KM, Wig G, Bischof GN, Rieck JR, et al. Association of longitudinal cognitive decline with amyloid burden in middle-aged and older adults: Evidence for a dose-response relationship. JAMA Neurol. 2017;74(7):830–8. pmid:28558099
View Article
PubMed/NCBI
Google Scholar

[140] View Article

[141] PubMed/NCBI

[142] Google Scholar

[ref38] 38. Samstag CL, Chapman NH, Gibbons LE, Geller J, Loeb N, Dharap S, et al. Neuropathological correlates of vulnerability and resilience in the cerebellum in Alzheimer’s disease. Alzheimers Dement. 2025;21(2):e14428. pmid:39713867
View Article
PubMed/NCBI
Google Scholar

[144] View Article

[145] PubMed/NCBI

[146] Google Scholar

[ref39] 39. Jagust WJ, Teunissen CE, DeCarli C. The complex pathway between amyloid β and cognition: Implications for therapy. Lancet Neurol. 2023;22(9):847–57. pmid:37454670
View Article
PubMed/NCBI
Google Scholar

[148] View Article

[149] PubMed/NCBI

[150] Google Scholar

[ref40] 40. Schmahmann JD. Cerebellum in Alzheimer’s disease and frontotemporal dementia: Not a silent bystander. Brain. 2016;139(Pt 5):1314–8. pmid:27189578
View Article
PubMed/NCBI
Google Scholar

[152] View Article

[153] PubMed/NCBI

[154] Google Scholar

[ref41] 41. Chao OY, Pathak SS, Zhang H, Augustine GJ, Christie JM, Kikuchi C, et al. Social memory deficit caused by dysregulation of the cerebellar vermis. Nat Commun. 2023;14(1):6007. pmid:37752149
View Article
PubMed/NCBI
Google Scholar

[156] View Article

[157] PubMed/NCBI

[158] Google Scholar

[ref42] 42. Krienen FM, Buckner RL. Segregated fronto-cerebellar circuits revealed by intrinsic functional connectivity. Cereb Cortex. 2009;19(10):2485–97. pmid:19592571
View Article
PubMed/NCBI
Google Scholar

[160] View Article

[161] PubMed/NCBI

[162] Google Scholar

[ref43] 43. Schmahmann JD, Sherman JC. The cerebellar cognitive affective syndrome. Brain. 1998;121 ( Pt 4):561–79. pmid:9577385
View Article
PubMed/NCBI
Google Scholar

[164] View Article

[165] PubMed/NCBI

[166] Google Scholar

[ref44] 44. Maiti B, Rawson KS, Tanenbaum AB, Koller JM, Snyder AZ, Campbell MC, et al. Functional connectivity of vermis correlates with future gait impairments in Parkinson’s Disease. Mov Disord. 2021;36(11):2559–68. pmid:34109682
View Article
PubMed/NCBI
Google Scholar

[168] View Article

[169] PubMed/NCBI

[170] Google Scholar

[ref45] 45. Wang L, Zhao P, Zhang J, Zhang R, Liu J, Duan J, et al. Functional connectivity between the cerebellar vermis and cerebrum distinguishes early treatment response for major depressive episodes in adolescents. J Affect Disord. 2023;339:256–63. pmid:37437740
View Article
PubMed/NCBI
Google Scholar

[172] View Article

[173] PubMed/NCBI

[174] Google Scholar

[ref46] 46. Lin Y, Kuo S-H. The emerging role of the cerebellum in neurodegeneration linked to cognitive impairment. J Alzheimers Dis. 2024;102(1):30–2. pmid:39497298
View Article
PubMed/NCBI
Google Scholar

[176] View Article

[177] PubMed/NCBI

[178] Google Scholar

[ref47] 47. de Mendonça LJC, Ferrari RJ. Alzheimer’s disease classification based on graph kernel SVMs constructed with 3D texture features extracted from MR images. Expert Systems with Applications. 2023;211:118633.
View Article
Google Scholar

[180] View Article

[181] Google Scholar

[ref48] 48. Lei B, Li Y, Fu W, Yang P, Chen S, Wang T, et al. Alzheimer’s disease diagnosis from multi-modal data via feature inductive learning and dual multilevel graph neural network. Med Image Anal. 2024;97:103213. pmid:38850625
View Article
PubMed/NCBI
Google Scholar

[183] View Article

[184] PubMed/NCBI

[185] Google Scholar

[ref49] 49. Chen Z, Bi S, Shan Y, Cui B, Yang H, Qi Z, et al. Multiparametric hippocampal signatures for early diagnosis of Alzheimer’s disease using 18F-FDG PET/MRI Radiomics. CNS Neurosci Ther. 2024;30(4):e14539. pmid:38031997
View Article
PubMed/NCBI
Google Scholar

[187] View Article

[188] PubMed/NCBI

[189] Google Scholar

[ref50] 50. Lin A, Chen Y, Chen Y, Ye Z, Luo W, Chen Y, et al. MRI radiomics combined with machine learning for diagnosing mild cognitive impairment: A focus on the cerebellar gray and white matter. Front Aging Neurosci. 2024;16:1460293. pmid:39430972
View Article
PubMed/NCBI
Google Scholar

[191] View Article

[192] PubMed/NCBI

[193] Google Scholar

[ref51] 51. Feng J, Jiang M, Zhang H, Yin L. Integrating imaging and genetic data via wavelet transform-based CNN for Alzheimer ’s disease classification. Biomedical Signal Processing and Control. 2025;104:107583.
View Article
Google Scholar

[195] View Article

[196] Google Scholar

[ref52] 52. Abdelaziz M, Wang T, Anwaar W, Elazab A. Multi-scale multimodal deep learning framework for Alzheimer’s disease diagnosis. Comput Biol Med. 2025;184:109438. pmid:39579666
View Article
PubMed/NCBI
Google Scholar

[198] View Article

[199] PubMed/NCBI

[200] Google Scholar

[ref53] 53. Yu H, Wang M, Yang Q, Xu X, Zhang R, Chen X, et al. The electrophysiological and neuropathological profiles of cerebellum in APPswe /PS1ΔE9 mice: A hypothesis on the role of cerebellum in Alzheimer’s disease. Alzheimers Dement. 2023;19(6):2365–75. pmid:36469008
View Article
PubMed/NCBI
Google Scholar

[202] View Article

[203] PubMed/NCBI

[204] Google Scholar

[ref54] 54. Gupta V, Booth S, Ko JH. Hypermetabolic cerebellar connectome in Alzheimer’s Disease. Brain Connect. 2023;13(6):356–66. pmid:34155909
View Article
PubMed/NCBI
Google Scholar

[206] View Article

[207] PubMed/NCBI

[208] Google Scholar

[ref55] 55. Garland EF, Dennett O, Lau LC, Chatelet DS, Bottlaender M, Nicoll JAR, et al. The mitochondrial protein TSPO in Alzheimer’s disease: relation to the severity of AD pathology and the neuroinflammatory environment. J Neuroinflammation. 2023;20(1):186. pmid:37580767
View Article
PubMed/NCBI
Google Scholar

[210] View Article

[211] PubMed/NCBI

[212] Google Scholar

[ref56] 56. Shimada H, Minatani S, Takeuchi J, Takeda A, Kawabe J, Wada Y, et al. Heavy tau burden with subtle amyloid β accumulation in the cerebral cortex and cerebellum in a case of familial Alzheimer’s disease with APP Osaka Mutation. Int J Mol Sci. 2020;21(12):4443. pmid:32580499
View Article
PubMed/NCBI
Google Scholar

[214] View Article

[215] PubMed/NCBI

[216] Google Scholar

Figures

Abstract

Introduction

Materials and methods

Subjects

Data preprocessing

Brain network construction

Statistical analyses

Model development, evaluation and interpretation

Model interpretability with shapley additive explan

Results

Demographic and clinical comparisons

Changes in amyloid plaque deposition connectivity in CN to AD

Changes in amyloid plaque deposition connectivity in CN to EMCI

Changes in amyloid plaque deposition connectivity in EMCI to LMCI

Changes in amyloid plaque deposition connectivity in LMCI to AD

Distinguishing performance of cerebellar network models and cortical brain models

Discussion

Conclusion

Supporting information

S1 File. Supplementary Methods for Early Diagnosis of Alzheimer’s Disease: Graph Theoretical Analysis of Cerebellar Network Features Based on 18F-AV45 PET.

Acknowledgments

References