This paper is in the following e-collection/theme issue:

Geroinformatics and Electronic Clinical Information/Decision Making in Geriatrics (37) Artificial Intelligence (2129) AI Language Models in Health Care (202) Reviews in Medical Informatics (112) Digital Mental Health Interventions, e-Mental Health and Cyberpsychology (2111) Neurotechnology in Clinical Studies (21)

Published on in Vol 8 (2025)

Preprints (earlier versions) of this paper are available at https://preprints.jmir.org/preprint/76981, first published .
Comparative Diagnostic Accuracy of AI-Assisted Fluorine-18 Fluorodeoxyglucose Positron Emission Tomography Versus Structural Magnetic Resonance Imaging in Alzheimer Disease: Systematic Review and Meta-Analysis

Comparative Diagnostic Accuracy of AI-Assisted Fluorine-18 Fluorodeoxyglucose Positron Emission Tomography Versus Structural Magnetic Resonance Imaging in Alzheimer Disease: Systematic Review and Meta-Analysis

Comparative Diagnostic Accuracy of AI-Assisted Fluorine-18 Fluorodeoxyglucose Positron Emission Tomography Versus Structural Magnetic Resonance Imaging in Alzheimer Disease: Systematic Review and Meta-Analysis

Authors of this article:

Bingbing Wang1 Author Orcid Image ;   Tailiang Zhao1 Author Orcid Image ;   Rongrong Ma2 Author Orcid Image ;   Xiaochuan Huo1 Author Orcid Image ;   Xiaoxiao Xiong3 Author Orcid Image ;   Minjie Wu1 Author Orcid Image ;   Yuran Wang4 Author Orcid Image ;   Liu Liu3 Author Orcid Image ;   Zhijiang Zhuang3 Author Orcid Image ;   Bin Wang4, 5 Author Orcid Image ;   Jixin Shou1 Author Orcid Image
Article Authors Cited by Tweetations Metrics

    1Department of Neurosurgery, Fifth Affiliated Hospital of Zhengzhou University, No. 3 Kangfu-qian Street, Zhengzhou, China

    2Department of Anesthesiology and Perioperative Medicine, Xuchang Central Hospital, Henan University of Science and Technology, Xuchang, China

    3Department of Neurology, Fifth Affiliated Hospital of Zhengzhou University, Zhengzhou, China

    4The Fifth Clinical Medical College, Henan Medical College, Zhengzhou University, Zhengzhou, China

    5Department of Cardiology, Beijing Tsinghua Changgung Hospital, School of Clinical Medicine, Tsinghua University, Beijing, China

    *these authors contributed equally

    Corresponding Author:

    Tailiang Zhao, PhD


    Background: Neuroimaging is crucial in the diagnosis of Alzheimer disease (AD). In recent years, artificial intelligence (AI)–based neuroimaging technology has rapidly developed, providing new methods for accurate diagnosis of AD, but its performance differences still need to be systematically evaluated.

    Objective: This study aims to conduct a systematic review and meta-analysis comparing the diagnostic performance of AI-assisted fluorine-18 fluorodeoxyglucose positron emission tomography (18F-FDG PET) and structural magnetic resonance imaging (sMRI) for AD.

    Methods: Databases including Web of Science, PubMed, and Embase were searched from inception to January 2025 to identify original studies that developed or validated AI models for AD diagnosis using 18F-FDG PET or sMRI. Methodological quality was assessed using the TRIPOD-AI (Transparent Reporting of a Multivariable Prediction Model for Individual Prognosis or Diagnosis–Artificial Intelligence) checklist. A bivariate mixed-effects model was employed to calculate pooled sensitivity, specificity, and summary receiver operating characteristic curve area (SROC-AUC).

    Results: A total of 38 studies were included, with 28 moderate-to-high-quality studies analyzed. Pooled SROC-AUC values were 0.94 (95% CI 0.92‐0.96) for sMRI and 0.96 (95% CI 0.94‐0.98) for 18F-FDG PET, demonstrating statistically significant intermodal differences (P=.02). Subgroup analyses revealed that for machine learning, pooled SROC-AUCs were 0.89 (95% CI 0.86‐0.92) for sMRI and 0.95 (95% CI 0.92‐0.96) for 18F-FDG PET, while for deep learning, these values were 0.96 (95% CI 0.94‐0.97) and 0.97 (95% CI 0.96‐0.99), respectively. Meta-regression identified heterogeneity arising from study quality stratification, algorithm types, and validation strategies.

    Conclusions: Both AI-assisted 18F-FDG PET and sMRI exhibit high diagnostic accuracy in AD, with 18F-FDG PET demonstrating superior overall diagnostic performance compared to sMRI.

    JMIR Aging 2025;8:e76981

    doi:10.2196/76981

    Keywords

    18F-FDG PET; Alzheimer disease; artificial intelligence; deep learning; machine learning; Preferred Reporting Items for Systematic Reviews and Meta-Analyses; PRISMA; sMRI; structural magnetic resonance imaging; systematic review 


    Alzheimer disease (AD) is a progressive neurodegenerative disorder characterized by insidious onset, cognitive decline, and memory impairment. In the United States, approximately 6.7 million adults aged ≥65 years are affected by AD, while China faces an even greater burden, with over 13 million cases [1,2]. The prolonged disease course and high comorbidity rates have established AD as one of the most fatal and economically burdensome conditions of the 21st century [3-5]. Accurate diagnosis of AD is critical for therapeutic decision-making and prognostic evaluation, particularly in the context of global population aging and increasing demands for precise patient stratification in clinical trials [6,7].

    Neuroimaging modalities, including fluorine-18 fluorodeoxyglucose positron emission tomography (18F-FDG PET) and structural magnetic resonance imaging (sMRI), have become cornerstone technologies in AD diagnostic frameworks (eg, National Institute on Aging-Alzheimer's Association criteria) because of their noninvasive nature and quantitative capabilities [8-10]. Recent advancements in artificial intelligence (AI) have revolutionized medical image analysis: machine learning (ML) enables data-driven predictive modeling beyond traditional rule-based programming, while deep learning (DL), an advanced ML paradigm, employs multilayer neural networks to extract high-level features from complex datasets, demonstrating transformative potential in neuroimaging [11-14]. Despite extensive research on AI-assisted sMRI or 18F-FDG PET for AD diagnosis, two critical challenges hinder evaluation of model generalizability: (1) substantial heterogeneity in algorithm designs and validation frameworks, and (2) significant variability in the quality of individual studies [15]. Furthermore, no high-quality meta-analysis has comprehensively compared the diagnostic performance of these two imaging modalities.

    This systematic review and meta-analysis addresses three critical objectives: (1) quantitative evaluation of diagnostic accuracy metrics for AI-assisted 18F-FDG PET and sMRI based on moderate-to-high-quality evidence; (2) direct comparison of diagnostic performance between modalities; and (3) investigation of confounding factors, including study quality (moderate-to-high vs low), algorithm types (ML vs DL), and validation strategies (internal vs external), through meta-regression. The findings are anticipated to inform evidence-based optimization of AD diagnostic pathways.


    Overview

    This study adhered to the PRISMA-DTA (Preferred Reporting Items for Systematic Reviews and Meta-Analyses of Diagnostic Test Accuracy) guidelines and was prospectively registered on PROSPERO (ID: CRD42023449927) to ensure transparency and minimize reporting bias [16,17]. Two reviewers independently conducted all stages of the review process, including title and abstract screening, full-text evaluation, data extraction, adherence assessment to reporting guidelines, and risk-of-bias evaluation. Discrepancies were resolved through group consensus.

    Literature Search Strategy

    Two investigators (TLZ and BW) systematically searched PubMed, Web of Science, and Embase from inception to January 2025 using a combination of Medical Subject Headings (MeSH) terms and free-text keywords. Additional searches were performed on clinical trial registries and OpenGrey to identify unpublished clinical trials and gray literature. Search terms encompassed four domains: (1) disease terminology (Alzheimer disease, AD, Alzheimer Syndrome); (2) imaging modalities (18F-FDG PET, sMRI); (3) AI methodologies (ML, DL); and (4) diagnostic metrics (sensitivity, specificity, summary receiver operating characteristic curve area [SROC-AUC]). Reference lists of included studies were manually screened to identify additional relevant publications. The specific search strategies are provided in Table S1 in Multimedia Appendix 1.

    Literature Inclusion and Exclusion Criteria

    The inclusion criteria were as follows: (1) human studies developing or validating AI models using 18F-FDG PET or sMRI to differentiate AD from normal controls; (2) AD diagnosis based on National Institute on Aging-Alzheimer's Association or International Working Group for New Research Criteria for Alzheimer’s Disease criteria; (3) availability of diagnostic performance metrics (eg, true positives, false positives, true negatives, false negatives) or explicit reporting of sensitivity and specificity; and (4) full-text availability in English. The exclusion criteria were as follows: (1) case reports, reviews, letters, or conference abstracts; (2) studies lacking sufficient diagnostic performance data; and (3) duplicate publications reporting on the same cohort without novel analyses.

    Literature Screening and Data Extraction

    Two reviewers (TZ and BW) independently performed title and abstract and full-text screening, followed by cross-verification to ensure accuracy. Two additional investigators (RM and XH) extracted data using predefined forms, including study characteristics (first author, publication year), model specifications (algorithm type, validation strategies), and diagnostic performance metrics (2×2 contingency tables, sensitivity, specificity). Extracted data were cross-checked for completeness and precision.

    Quality Assessment of Included Studies

    The methodological quality of included studies was evaluated using an adapted TRIPOD-AI (Transparent Reporting of a Multivariable Prediction Model for Individual Prognosis or Diagnosis–Artificial Intelligence) checklist (Table S2 in Multimedia Appendix 1) by two independent reviewers (RM and XH) [18]. The instrument assessed four domains: (1) data quality (sample diversity, adequacy, preprocessing standardization), (2) model development (feature extraction rationale, algorithm selection, hyperparameter optimization), (3) validation methods (cross-validation rigor, external validation), and (4) clinical applicability (interpretability, net clinical benefit). Each of the nine items was scored 0‐2 (0=unsatisfied; 1=partially satisfied; 2=fully satisfied), with total scores categorized as high quality (≥16), moderate quality (10-15), or low quality (≤9). Disagreements were resolved through discussion.

    Data Analysis

    Diagnostic performance metrics were calculated using a bivariate mixed-effects model. SROC-AUC served as the primary outcome due to its threshold independence. Diagnostic accuracy was classified per National Institutes of Health criteria as high (AUC≥0.90), moderate (0.70‐0.89), or low (AUC<0.70) [19].

    Heterogeneity Assessment and Publication Bias Evaluation

    Heterogeneity was quantified using Cochran Q test (significance defined at the .05 level) and I2 statistics (25%: low; 50%: moderate; 75%: high heterogeneity) [20]. Threshold effects were assessed using Spearman correlation between logit-transformed sensitivity and 1–specificity. Sensitivity analyses evaluated outlier influence by iteratively excluding individual studies. Univariable meta-regression analyses were conducted to assess the influence of potential confounding factors, including study quality (moderate to high vs low quality), algorithm type (ML vs DL), and validation strategy (internal vs external validation). Statistical significance of modifiers of the pooled effect size was defined at the .05 level after adjustment via the Holm-Bonferroni method. Publication bias was assessed via funnel plot asymmetry and Egger test, with significance defined at the .05 level. All analyses were conducted in Stata 16.0 using the MIDAS commands.

    As this study focused on diagnostic accuracy, we employed the bivariate random-effects model via the MIDAS command in Stata to jointly synthesize sensitivity and specificity. This approach accounts for both within- and between-study variability, as well as the inherent correlation between sensitivity and specificity. In such models, conventional heterogeneity measures like I², although still reported for completeness, may not fully capture the joint variability of test performance measures. Therefore, to better understand heterogeneity, we additionally performed meta-regression and calculated joint likelihood ratio tests to explore potential effect modifiers. These methods align with recommended practices in diagnostic test accuracy meta-analyses [21,22].


    Literature Screening Process and Results

    Initial searches identified 876 PubMed records (n=89 involving 18F-FDG PET; n=787 sMRI), 932 Embase records (n=111 18F-FDG PET; n=821 sMRI), and 2610 Web of Science records (n=230 18F-FDG PET; n=2380 sMRI), supplemented by 3 additional studies. After deduplication in EndNote 20, 2 reviewers independently screened titles and abstracts using the Population, Intervention, Comparison, Outcome, Study Design (PICOS) criteria, followed by full-text evaluation against predefined inclusion/exclusion criteria. This rigorous process yielded 38 studies for systematic review and meta-analysis [23-60]. The literature screening process and results are shown in Figure 1.

    ‎
    Figure 1. PRISMA (Preferred Reporting Items for Systematic Reviews and Meta-Analysis) flow diagram.

    Characteristics of Included Studies

    Table 1 and Table S3 in Multimedia Appendix 1 detail the characteristics of the included studies. Imaging modalities comprised sMRI (n=15, 39%), 18F-FDG PET (n=15, 39%), and combined sMRI/18F-FDG PET (n=8, 22%). Data sources predominantly relied on open-access databases (n=29, 76%), with 28 (97%) from the Alzheimer’s Disease Neuroimaging Initiative database and 1 (3%) from the Open Access Series of Imaging Studies database. Four studies (11%) used institutional data, while 5 (13%) combined Alzheimer’s Disease Neuroimaging Initiative database with local datasets. Algorithmically, 26 studies (68%) employed ML, predominantly support vector machines (17/26, 65%), whereas 12 (32%) utilized DL, primarily convolutional neural networks (9/12, 75%). Internal validation was implemented in 33 studies (87%), with 10-fold cross-validation in 16 (48%); only 5 studies (13%) incorporated external validation.

    Table 1. Characteristics of all included studies (n=38).
    StudiesModel development and internal validationModel external validation
    DefinitionModalityData sourceAlgorithm usedTraining/validation (ratio)TestingType of internal validationData sourceDefinition
    Zhang et al [23], 2011ADa: 51; NCb: 52sMRIc; PETdADNIe databaseSVMfNRgNR10-fold CVhNRNR
    Yun et al [24], 2015AD: 71; NC: 85sMRI; PETADNI databaseLDAiNRNRLOOCVjNRNR
    Westman et al [25], 2012AD: 96; NC: 111sMRIADNI databaseOPLSkNRNR7-fold CVNRNR
    Vemuri et al [26], 2008AD: 190; NC: 190sMRIMayo ClinicSVM2801004-fold CV; hold-outNRNR
    Suk et al [27], 2014AD: 93; NC: 101sMRI; PETADNI databaseDBMl; SVMNRNR10-fold CVNRNR
    Sayeed et al [28], 2002AD: 18; NC: 10PETHammersmith HospitalDFAmNRNRLOOCVNRNR
    Pan et al [29], 2019AD: 247; NC: 246PETADNI databaseSVMNRNR10-fold CVNRNR
    Padilla et al [30], 2012AD: 53; NC: 52PETADNI databaseSVMNRNRLOOCVNRNR
    Ni et al [31], 2021AD: 638; NC: 629PETADNI databaseCNNn1000267Hold-outNRNR
    Magnin et al [32], 2009AD: 16; NC: 22sMRIPitié-Salpétriere HospitalSVM75%25%Bootstrap resamplingNRNR
    Lu et al [33], 2018AD: 226; NC: 304PETADNI databaseMDNNoNRNR10-fold CVNRNR
    Liu et al [34], 2012AD: 198; NC: 229sMRIADNI databaseSRCpNRNR10-fold CVNRNR
    Liu et al [35], 2018AD: 93; NC: 100PETADNI database2D-CNNNRNR10-fold CVNRNR
    Li et al [36], 2015AD: 44; NC: 45PETADNI database; TUM databaseGMMq; SVMNRNR10-fold CVNRNR
    Lerch et al [37], 2008AD: 19; NC: 17sMRILudwig Maximilian University of MunichQDArNRNRLOOCVNRNR
    Kim et al [38], 2020AD: 141; NC: 348PETADNI databaseCNNNRNRNRSeverance datasetAD: 80; NC: 72
    Kim et al [39], 2020AD: 139; NC: 347PETADNI databaseBEGANs; SVMNRNRNRSeverance datasetAD: 73; NC: 68
    Katako et al [40], 2018AD: 94; NC: 111PETADNI databaseSVMNRNR10-fold CVNRNR
    Ismail et al [41], 2023AD: 511; NC: 535sMRI; PETADNI databaseMultiAz-Net; SVMNRNR10-fold CVNRNR
    Illán et al [42], 2011AD: 95; NC: 97PETADNI databaseSVMNRNRLOOCVNRNR
    Hinrichs et al [43], 2009AD: 89; NC: 94sMRI; PETADNI databaseLPboostingtNRNRLeave-many-out CVNRNR
    Gray et al [44], 2012AD: 50; NC: 54PETADNI databaseSVM75%25%Monte Carlo CVNRNR
    Gray et al [45], 2013AD: 37; NC: 35sMRI; PETADNI databaseRFu75%25%Monte Carlo CVNRNR
    Feng et al [46], 2019AD: 93; NC: 100sMRI; PETADNI databaseFSBi-LSTMv; 3D-CNNNRNR10-fold CVNRNR
    Cuingnet et al [47], 2011AD: 137; NC: 162sMRIADNI databaseSVM50%50%Hold-outNRNR
    Chen et al [48], 2022AD: 326; NC: 413sMRIADNI-1 database; ADNI-2 database2D-CNN; 3D-CNNNRNRNRADNI-1 database; ADNI-2 databaseAD: 326; NC: 413
    Song et al [49], 2021AD: 95; NC: 126sMRI; PETADNI database3D-CNNNRNR10-fold CVNRNR
    Li et al [50], 2019AD: 130; NC: 162PETADNI databaseSVM70%30%Monte Carlo CVHuashan databaseAD: 22; NC: 22
    Ahila et al [51], 2022AD: 220; NC: 635PETADNI database2D-CNN90%10%Hold-outNRNR
    Toussaint et al [52], 2012AD: 80; NC: 80PETADNI databaseSVMNRNRLOOCVNRNR
    Tong et al [53], 2014AD: 198; NC: 231sMRIADNI databaseSVMNRNRLOOCVNRNR
    Min et al [54], 2014AD: 97; NC: 128sMRIADNI databaseSVMNRNR10-fold CVNRNR
    Jin et al [55], 2020AD: 488; NC: 536sMRIIn-house database; ADNI database3D attention networkNRNR10-fold CV; leave center out CVADNI database; in-house databaseAD: 488; NC: 536
    Cho et al [56], 2012AD: 128; NC: 160sMRIADNI databaseLDA142146Hold-outNRNR
    Chincarini et al [57], 2011AD: 144; NC: 189sMRIADNI databaseRF; SVMNRNR20-fold CVNRNR
    Beheshti et al [58], 2017AD: 92; NC: 94sMRIADNI databaseSVMNRNR10-fold CVNRNR
    Anandh et al [59], 2016AD: 30; NC: 55sMRIOASISw databaseSVMNRNR10-fold CVNRNR
    Amoroso et al [60], 2018AD: 86; NC: 81sMRIADNI databaseRF6710010-fold CV; hold-outNRNR

    aAD: Alzheimer disease.

    bNC: normal cognitive.

    csMRI: structural magnetic resonance imaging.

    dPET: positron emission tomography.

    eADNI: Alzheimer’s Disease Neuroimaging Initiative.

    fSVM: support vector machine.

    gNR: no report.

    hCV: cross-validation.

    iLDA: linear discriminant analysis.

    jLOOCV: leave-one-out cross-validation.

    kOPLS: orthogonal partial least squares.

    lDBM: deep Boltzmann machine.

    mDFA: discriminant function analysis.

    nCNN: convolutional neural network.

    oMDNN: multiscale deep neural network.

    pSRC: sparse representation–based classifier.

    qGMM: Gaussian mixture model.

    rQDA: quadratic discriminant analysis.

    sBEGAN: boundary equilibrium generative adversarial network.

    tLP: linear program.

    uRF: random forest.

    vFSBi-LSTM: fully stacked bidirectional long short-term memory.

    wOASIS: Open Access Series of Imaging Studies.

    Quality Assessment Results of Included Studies

    Nine (24%) studies were high quality, characterized by adequate sample sizes, rigorous validation (eg, multicenter external validation), and standardized reporting. Moderate-quality studies (n=19, 50%) met basic methodological standards (eg, cross-validation) but had limitations such as single-center data or insufficient clinical correlation analyses. Ten (26%) studies were low quality due to small sample sizes (<100 cases) and inadequate validation strategies, compromising external validity. The results of quality assessments for each study are detailed in Table S4 in Multimedia Appendix 1.

    Data Analysis Results

    Overall Diagnostic Performance

    Analysis of 29 sMRI-related contingency tables demonstrated a pooled sensitivity of 88% (95% CI 86%-89%), a specificity of 92% (95% CI 90%-93%), and an SROC-AUC of 0.94 (95% CI 0.92-0.96). Subgroup analyses stratified by algorithm type revealed that ML (19 tables) achieved a sensitivity of 86% (95% CI 85%-88%), a specificity of 91% (95% CI 88%-93%), and an SROC-AUC of 0.88 (95% CI 0.85-0.90), while DL (10 tables) showed improved performance with a sensitivity of 88% (95% CI 86%-91%), a specificity of 92% (95% CI 90%-94%), and an SROC-AUC of 0.96 (95% CI 0.94-0.97). Stratification by study quality indicated that moderate-to-high-quality studies (24 tables) yielded a sensitivity of 87% (95% CI 85%-89%), a specificity of 91% (95% CI 89%-93%), and an SROC-AUC of 0.94 (95% CI 0.92-0.96), whereas low-quality studies (5 tables) reported elevated metrics with a sensitivity of 91% (95% CI 87%-94%), a specificity of 95% (95% CI 92%-97%), and an SROC-AUC of 0.98 (95% CI 0.96-0.99). Validation strategy–based analysis showed that internal validation (25 tables) achieved a sensitivity of 88% (95% CI 86%-90%), a specificity of 92% (95% CI 90%-94%), and an SROC-AUC of 0.95 (95% CI 0.92-0.96), while external validation (4 tables) demonstrated marginally lower performance with a sensitivity of 85% (95% CI 81%-89%), a specificity of 91% (95% CI 85%-94%), and an SROC-AUC of 0.93 (95% CI 0.90-0.95).

    For 18F-FDG PET (27 tables), pooled estimates demonstrated a sensitivity of 90% (95% CI 88%-92%), a specificity of 93% (95% CI 91%-94%), and an SROC-AUC of 0.96 (95% CI 0.94-0.98). Subgroup analyses stratified by algorithm type revealed that ML (15 tables) achieved a sensitivity of 89% (95% CI 86%-90%), a specificity of 91% (95% CI 88%-93%), and an SROC-AUC of 0.94 (95% CI 0.91-0.96), while DL (12 tables) exhibited superior performance with a sensitivity of 91% (95% CI 89%-93%), a specificity of 94% (95% CI 93%-96%), and an SROC-AUC of 0.98 (95% CI 0.96-0.99). Moderate-to-high-quality studies (21 tables) demonstrated a sensitivity of 90% (95% CI 88%-92%), a specificity of 93% (95% CI 91%-94%), and an SROC-AUC of 0.96 (95% CI 0.94-0.98), whereas low-quality studies (6 tables) showed comparable metrics with a sensitivity of 91% (95% CI 86%-94%), a specificity of 93% (95% CI 87%-96%), and an SROC-AUC of 0.96 (95% CI 0.94-0.98). Internal validation (24 tables) yielded a sensitivity of 90% (95% CI 89%-92%), a specificity of 93% (95% CI 91%-94%), and an SROC-AUC of 0.96 (95% CI 0.94%-0.97%), while external validation data (3 tables) were insufficient for SROC-AUC calculation but reported a sensitivity of 87% (95% CI 81%-93%) and a specificity of 95% (95% CI 91%-97%). The data analysis results are shown in Table 2.

    Table 2. Summary estimates and meta-regression of pooled performance of all studies.
    Main and subordinate directoryTables, nSensitivity (%; 95% CI)I2Specificity (%; 95% CI)I2Joint P valueAUCa (95% CI)
    sMRIb
    Overall2988 (86-89)55.3292 (90-93)74.08—j0.94 (0.92-0.96)
    Study quality.02
    Medium-to-high quality2487 (85-89)57.9991 (89-93)75.940.94 (0.92-0.96)
    Low quality591 (87-94)10.295 (92-97)00.98 (0.96-0.99)
    Algorithm type.40
    MLd1986 (85-88)21.2391 (88-93)73.820.88 (0.85-0.90)
    DLe1088 (86-91)76.5692 (90-94)76.530.96 (0.94-0.97)
    Validation strategy.33
    Internal validation2588 (86-90)46.3892 (90-94)71.650.95 (0.92-0.96)
    External validation485 (81-89)72.2691 (85-94)83.650.93 (0.90-0.95)
    SVMf1087 (85-90)17.4793 (90-95)57.650.93 (0.90-0.95)
    CNNg988 (85-91)78.1992 (90-94)77.550.96 (0.93-0.97)
    18F-FDG PETk
    Overall2790 (88-92)42.0493 (91-94)61.49—0.96 (0.94-0.98)
    Study quality.96
    Medium-to-high quality2190 (88-92)44.5893 (91-94)58.520.96 (0.94-0.98)
    Low quality691 (86-94)42.7393 (87-96)72.30.96 (0.94-0.98)
    Algorithm type.01
    ML1589 (86-90)1.6191 (88-93)45.640.94 (0.91-0.96)
    DL1291 (89-93)55.2394 (93-96)45.920.98 (0.96-0.99)
    Validation strategy.26
    Internal validation2490 (89-92)34.6193 (91-94)64.90.96 (0.94-0.97)
    External validation387 (81-93)—95 (91-99)——
    SVM1190 (87-92)092 (89-94)37.50.94 (0.92-0.96)
    CNN891 (88-94)69.7693 (92-95)34.630.97 (0.95-0.98)

    aAUC: area under the receiver operating characteristic curve.

    bsMRI: structural magnetic resonance imaging.

    cNot applicable.

    dML: machine learning.

    eDL: deep learning.

    fSVM: support vector machine.

    gCNN: convolutional neural network.

    h18F-FDG PET: fluorine-18 fluorodeoxyglucose positron emission tomograph.

    Moderate-to-High-Quality Study Subgroup Analysis

    For sMRI, pooled sensitivity, specificity, and SROC-AUC were 87% (95% CI 85%‐89%), 91% (95% CI 89%‐93%), and 0.94 (95% CI 0.92‐0.96), respectively (see Figures 2 and 3). Stratification by algorithm type indicated that ML models achieved 86% (95% CI 84%‐88%) sensitivity, 89% (95% CI 86%‐92%) specificity, and an SROC-AUC of 0.89 (95% CI 0.86‐0.92), while DL models demonstrated 88% (95% CI 86%‐91%) sensitivity, 92% (95% CI 90%‐94%) specificity, and an SROC-AUC of 0.96 (95% CI 0.94‐0.97).

    For 18F-FDG PET, pooled sensitivity, specificity, and SROC-AUC were 90% (95% CI 88%‐92%), 93% (95% CI 91%‐94%), and 0.96 (95% CI 0.94‐0.98), respectively (see Figures 4 and 5). Subgroup analyses revealed ML models achieved 89% (95% CI 86‐91%) sensitivity, 91% (95% CI 87%‐93%) specificity, and an SROC-AUC of 0.95 (95% CI 0.92‐0.96), whereas DL models outperformed with 91% (95% CI 89%‐93%) sensitivity, 94% (95% CI 93%‐96%) specificity, and SROC-AUC 0.97 (95% CI 0.96‐0.99). The data analysis results are also shown in Table 3.

    ‎
    Figure 2. The pooled sensitivities and specificities based on moderate-to-high-quality studies for structural magnetic resonance imaging [23,24,25,26,27,34,41,43,45,46,47,48,49,53,55,57,60].
    ‎
    Figure 3. Summary receiver operating characteristic (SROC) based on moderate-to-high-quality studies curves for structural magnetic resonance imaging. AUC: area under the receiver operating characteristic curve area; SENS: sensitivity; SPEC: specificity.
    ‎
    Figure 4. The pooled sensitivities and specificities based on moderate-to-high-quality studies for fluorine-18 fluorodeoxyglucose positron emission tomography (18F-FDG PET) [23,24,27,29,31,33,35,38,39,41,42,43,44,45,45,46,49,50].
    ‎
    Figure 5. Summary receiver operating characteristic (SROC) based on moderate-to-high-quality studies curves for fluorine-18 fluorodeoxyglucose positron emission tomography (18F-FDG PET). AUC: area under the receiver operating characteristic curve area; SENS: sensitivity; SPEC: specificity.
    Table 3. Summary estimates and meta-regression of pooled performance of moderate-to-high-quality studies.
    Main and subordinate directoryTables, nSensitivity (%; 95% CI)I2Specificity (%; 95% CI)I2Joint P valueAUCa (95% CI)
    Overall.02
    sMRIb2487 (85-97)57.9991 (89-93)75.940.94 (0.92-0.96)
    18F-FDG PETc2190 (88-92)44.5893 (91-94)58.520.96 (0.94-0.98)
    MLd
    1486 (84-88)3.7289 (86-92)75.580.89 (0.86-0.92)
    18F-FDG PET1089 (86-91)16.5191 (87-93)53.950.95 (0.92-0.96)
    DLe
    1088 (86-91)76.5692 (90-94)76.530.96 (0.94-0.97)
    18F-FDG PET1191 (89-93)54.5994 (93-95)36.230.97 (0.96-0.99)

    aAUC: area under the receiver operating characteristic curve.

    bsMRI: structural magnetic resonance imaging.

    c18F-FDG PET: fluorine-18 fluorodeoxyglucose positron emission tomography.

    dML: machine learning.

    eDL: deep learning.

    Exploration of Statistical Heterogeneity Sources

    Full Study Cohort

    For sMRI contingency tables, sensitivity exhibited moderate heterogeneity (I²=55.32%), while specificity showed high heterogeneity (I²=74.08%). Threshold effect testing revealed no significant threshold effect (r=−0.101, P=.60). Subgroup analyses identified study quality stratification (P=.02) as a source of heterogeneity, with moderate-to-high-quality studies demonstrating lower sensitivity (87% vs 91%) and specificity (91% vs 95%) compared to low-quality studies. Algorithm type (P=.40) and validation strategy (P=.33) were not significant contributors.

    For 18F-FDG PET analyses, sensitivity and specificity displayed moderate (I²=42.04%) and high heterogeneity (I²=61.49%), respectively, with no threshold effect (r=−0.087, P=.67). Algorithm type (P=.01) significantly influenced heterogeneity, as ML models demonstrated lower sensitivity (89% vs 91%) and specificity (91% vs 94%) than DL. Study quality (P=.96) and validation strategy (P=.26) showed no significant impact. The data analysis results are also shown in Table 2.

    Moderate-to-High-Quality Study Subgroups

    For sMRI-based models, ML implementations demonstrated minimal sensitivity heterogeneity (I²=3.72%) but high specificity heterogeneity (I²=75.58%). Exclusion of studies using traditional ensemble algorithms (random forest, boosting) reduced specificity heterogeneity to moderate (I²=45.47%), revealing that ensemble methods achieved higher sensitivity (89% vs 87%) but significantly lower specificity (79% vs 93%) compared to nonensemble ML models (P<.001). In contrast, DL-based sMRI models exhibited high heterogeneity for both sensitivity (I²=76.56%) and specificity (I²=76.53%). Removing external validation data mitigated heterogeneity to moderate (I²=54.36%) and low levels (I²=38.60%), respectively, with external validation studies showing significantly reduced sensitivity (85% vs 90%) and specificity (91% vs 93%) compared to internal validation (P<.001).

    For 18F-FDG PET-based models, ML implementations showed low sensitivity heterogeneity (I²=16.51%) and moderate specificity heterogeneity (I²=53.95%). DL models exhibited moderate sensitivity heterogeneity (I²=54.59%) and low specificity heterogeneity (I²=36.23%). The data analysis results also are shown in Table 3 and Table S5 in Multimedia Appendix 1.

    Publication Bias Test

    Funnel plot analysis of the full study cohort revealed no significant publication bias for sMRI (P=.69), whereas the 18F-FDG PET subgroup exhibited significant bias (P<.001). In the moderate-to-high-quality cohort, both sMRI (P=.03) and 18F-FDG PET (P=.01) demonstrated publication bias. However, subgroup analyses showed no significant bias for sMRI+ML (P=.06), sMRI+DL (P=.89), 18F-FDG PET+ML (P=.08), or 18F-FDG PET+DL (P=.28).


    Principal Findings

    This meta-analysis confirms that AI-enhanced sMRI and 18F-FDG PET achieve high diagnostic accuracy for AD, with pooled SROC-AUCs of 0.94 and 0.96, respectively, outperforming conventional visual assessments [61,62]. 18F-FDG PET demonstrated superior overall performance (P=.02), likely attributable to its sensitivity to AD-specific metabolic abnormalities. Notably, ML amplified PET’s diagnostic advantage (SROC-AUC: 0.95 vs 0.89), while DL narrowed the gap (SROC-AUC: 0.97 vs 0.96), highlighting DL’s capacity to extract complex metabolic features and mitigate structural imaging limitations [63].

    Subgroup analyses identified three key determinants of heterogeneity and performance variation. First, methodological quality significantly influenced sMRI models: moderate-to-high-quality studies reported lower sensitivity (87% vs 91%) and specificity (91% vs 95%) compared to low-quality studies (P=.02), suggesting inflated performance in the latter due to small sample sizes, single-center data, or nonstandardized preprocessing. Adherence to TRIPOD-AI guidelines and transparent reporting of preprocessing workflows are critical for future studies [18]. Second, algorithm type drove technical divergence in PET models: DL outperformed ML (sensitivity: 91% vs 89%; specificity: 94% vs 91%; P=.01) through end-to-end feature learning. However, ML’s interpretability better aligns with clinical demands for transparency, whereas DL’s reliance on large annotated datasets and computing resources limits its deployment in resource-constrained regions [64]. Third, validation strategies exposed generalizability limitations: external validation of sMRI+DL models showed reduced sensitivity (85% vs 90%) and specificity (91% vs 93%) compared to internal validation (P<.001).

    The reduced diagnostic accuracy observed in externally validated models may stem from (1) data distribution shift—differences in feature distributions, class balance, and temporal trends between training and external datasets; (2) overfitting—models may have overfitted to training-specific noise or spurious patterns, limiting generalization; and (3) implementation and annotation inconsistencies—variations in data preprocessing, feature scaling, and labeling protocols across datasets [65].

    Future Directions and Study Limitations

    This study has two methodological constraints: potential selection bias from English-only inclusion and possible overestimation by prioritizing optimal contingency tables. Future research should focus on enhancing evidence robustness through multinational, multiethnic cohorts; improving transparency via open-source preprocessing codes, findable, accessible, interoperable, and reusable-compliant data sharing, and independent validation; and balancing performance and interpretability via explainable DL frameworks to meet clinical ethical standards [66].

    We also acknowledge the risk of publication bias, as studies with positive results are more likely to be published, particularly in AI-related research where rapid progress and selective reporting may skew the literature. Additionally, language and geographic biases may exist since only English-language articles were included and most studies originated from a limited number of regions (eg, China, the United States, and Europe). This may limit the generalizability of our findings to underrepresented regions.

    Despite 18F-FDG PET’s higher accuracy, its radiation exposure and cost hinder widespread screening [67]. Conversely, sMRI’s cost-effectiveness and noninvasiveness position it as a first-line screening tool, with PET reserved for confirmatory testing in complex cases—a tiered diagnostic strategy. The 2023 Responsible AI for Social and Ethical Healthcare consensus implementation priorities, including transparent cost-benefit frameworks cross-modal standardization, and dynamic performance monitoring [68].

    Comparison With Existing Reviews

    Although previous reviews have explored the diagnostic accuracy of AI in AD, studies involving meta-analyses remain scarce [69-71]. Conducting such meta-analyses faces significant challenges, particularly due to substantial methodological heterogeneity across studies. This heterogeneity manifests in multiple dimensions, including variations in neuroimaging modalities, disparities in model validation strategies, and differences in algorithm types—all of which influence AD diagnostic performance and complicate the synthesis of evidence.

    Borchert et al [70] conducted a comprehensive systematic review of 255 neuroimaging studies utilizing AI for dementia diagnosis and prognosis. Their findings demonstrated that discriminative models, particularly DL approaches, outperformed algorithmic classifiers in distinguishing AD patients from healthy controls. However, they emphasized critical methodological limitations, with conclusions primarily relying on qualitative synthesis rather than quantitative evidence.

    In a systematic review and meta-analysis by Sun et al [71], the diagnostic accuracy of DL models based on 18F-FDG PET for AD was investigated. While the study reported excellent diagnostic performance, notable heterogeneity was observed during meta-analysis, raising concerns about the reliability of the findings. Furthermore, the study focused exclusively on DL, overlooking the widespread application of traditional ML methods in current clinical research for AD diagnostic modeling.

    In contrast, our meta-analysis incorporates a broader range of studies, rigorously controls methodological heterogeneity through stringent quality assessment and detailed subgroup analyses, and systematically evaluates the diagnostic accuracy of both ML and DL in AD. By emphasizing methodological rigor and the importance of external validation in AI-assisted neuroimaging for AD diagnosis, this study addresses critical gaps in the existing literature.

    Although our subgroup comparisons between ML and DL models provide a useful overview of broad methodological trends, it must be noted that algorithm complexity, training data size, and model optimization procedures vary considerably within each group. The observed performance differences may, therefore, reflect not only model class but also differences in dataset size, feature representation, and implementation quality. Future studies should aim to compare individual algorithms under standardized conditions.

    Conclusions

    In conclusion, AI can effectively support the diagnosis of AD using sMRI and 18F-FDG PET imaging. Among these approaches, combining PET imaging with DL techniques yields the highest diagnostic accuracy. These findings suggest that a future direction lies in integrating precision neuroimaging with AI tools. To bring such systems into routine clinical use—helping doctors detect AD earlier, personalize treatment, and improve patient outcomes—future studies should focus on repeated validation with high-quality clinical datasets and the development of standardized implementation protocols.

    Acknowledgments

    This research was funded by the Henan Province Medical Science and Technology Public Relations Plan Joint Construction Project (LHGJ20230402).

    Conflicts of Interest

    None declared.

    Multimedia Appendix 1

    Detailed database search process, characteristics of included studies, quality assessment results, and subgroup analysis findings.

    DOCX File, 59 KB
    Checklist 1

    PRISMA 2020 checklist.

    PDF File, 101 KB
    1. 2023 Alzheimer’s disease facts and figures. Alzheimers Dement. Apr 2023;19(4):1598-1695. [CrossRef]
    2. Ren R, Qi J, Lin S, et al. The China Alzheimer Report 2022. Gen Psychiatr. 2022;35(1):e100751. [CrossRef] [Medline]
    3. Jia J, Wei C, Chen S, et al. The cost of Alzheimer’s disease in China and re-estimation of costs worldwide. Alzheimers Dement. Apr 2018;14(4):483-491. [CrossRef] [Medline]
    4. Scheltens P, Blennow K, Breteler MMB, et al. Alzheimer’s disease. Lancet. Jul 30, 2016;388(10043):505-517. [CrossRef] [Medline]
    5. Pickett AC, Valdez D, White LA, et al. The CareVirtue digital journal for family and friend caregivers of people living with Alzheimer disease and related dementias: exploratory topic modeling and user engagement study. JMIR Aging. Dec 24, 2024;7:e67992. [CrossRef] [Medline]
    6. Cahill S. WHO’s global action plan on the public health response to dementia: some challenges and opportunities. Aging Ment Health. Feb 2020;24(2):197-199. [CrossRef] [Medline]
    7. Wu Y, Fu L, Li Q, et al. Recent advancements in the early diagnosis and treatment of Alzheimer’s disease. Adv Therap. Nov 2023;6(11). URL: https://onlinelibrary.wiley.com/toc/23663987/6/11 [CrossRef] [Medline]
    8. Jack CR, Bennett DA, Blennow K, et al. NIA‐AA Research Framework: toward a biological definition of Alzheimer’s disease. Alzheimer’s & Dementia. Apr 2018;14(4):535-562. [CrossRef]
    9. Jack CR, Andrews JS, Beach TG, et al. Revised criteria for diagnosis and staging of Alzheimer’s disease: Alzheimer’s Association Workgroup. Alzheimer’s & Dementia. Aug 2024;20(8):5143-5169. [CrossRef]
    10. By S, Kahl A, Cogswell PM. Alzheimer’s disease clinical trials: what have we learned from magnetic resonance imaging. J Magn Reson Imaging. Feb 2025;61(2):579-594. [CrossRef] [Medline]
    11. Kuo RYL, Harrison C, Curran TA, et al. Artificial intelligence in fracture detection: a systematic review and meta-analysis. Radiology. Jul 2022;304(1):50-62. [CrossRef] [Medline]
    12. Wang J, Xue L, Jiang J, et al. Diagnostic performance of artificial intelligence-assisted PET imaging for Parkinson’s disease: a systematic review and meta-analysis. NPJ Digit Med. Jan 22, 2024;7(1):17. [CrossRef] [Medline]
    13. Cheng Y, Malekar M, He Y, et al. High-throughput phenotyping of the symptoms of Alzheimer disease and related dementias using large language models: cross-sectional study. JMIR AI. Jun 3, 2025;4:e66926. [CrossRef] [Medline]
    14. Bosco C, Shojaei F, Theisz AA, et al. Testing 3 modalities (voice assistant, chatbot, and mobile app) to assist older African American and Black adults in seeking information on Alzheimer disease and related dementias: wizard of Oz usability study. JMIR Form Res. Dec 9, 2024;8:e60650. [CrossRef] [Medline]
    15. Rudroff T, Rainio O, Klén R. AI for the prediction of early stages of Alzheimer’s disease from neuroimaging biomarkers—a narrative review of a growing field. Neurol Sci. Nov 2024;45(11):5117-5127. [CrossRef] [Medline]
    16. Jackson D, Turner R. Power analysis for random-effects meta-analysis. Res Synth Methods. Sep 2017;8(3):290-302. [CrossRef] [Medline]
    17. McInnes MDF, Moher D, Thombs BD, et al. Preferred reporting items for a systematic review and meta-analysis of diagnostic test accuracy studies: the PRISMA-DTA statement. JAMA. Jan 23, 2018;319(4):388-396. [CrossRef] [Medline]
    18. TRIPOD+AI statement: updated guidance for reporting clinical prediction models that use regression or machine learning methods. BMJ. Apr 2024:q902. [CrossRef]
    19. Singh S, Chang SM, Matchar DB, Bass EB. Chapter 7: grading a body of evidence on diagnostic tests. J Gen Intern Med. Jun 2012;27 Suppl 1(Suppl 1):S47-S55. [CrossRef] [Medline]
    20. Ruppar T. Meta-analysis: how to quantify and explain heterogeneity? Eur J Cardiovasc Nurs. Oct 2020;19(7):646-652. [CrossRef] [Medline]
    21. Rutter CM, Gatsonis CA. A hierarchical regression approach to meta-analysis of diagnostic test accuracy evaluations. Stat Med. Oct 15, 2001;20(19):2865-2884. [CrossRef] [Medline]
    22. Reitsma JB, Glas AS, Rutjes AWS, Scholten RJPM, Bossuyt PM, Zwinderman AH. Bivariate analysis of sensitivity and specificity produces informative summary measures in diagnostic reviews. J Clin Epidemiol. Oct 2005;58(10):982-990. [CrossRef] [Medline]
    23. Zhang D, Wang Y, Zhou L, Yuan H, Shen D. Multimodal classification of Alzheimer’s disease and mild cognitive impairment. Neuroimage. Apr 1, 2011;55(3):856-867. [CrossRef] [Medline]
    24. Yun HJ, Kwak K, Lee JM, Alzheimer’s Disease Neuroimaging Initiative. Multimodal discrimination of Alzheimer’s disease based on regional cortical atrophy and hypometabolism. PLoS ONE. 2015;10(6):e0129250. [CrossRef] [Medline]
    25. Westman E, Muehlboeck JS, Simmons A. Combining MRI and CSF measures for classification of Alzheimer’s disease and prediction of mild cognitive impairment conversion. Neuroimage. Aug 1, 2012;62(1):229-238. [CrossRef] [Medline]
    26. Vemuri P, Gunter JL, Senjem ML, et al. Alzheimer’s disease diagnosis in individual subjects using structural MR images: validation studies. Neuroimage. Feb 1, 2008;39(3):1186-1197. [CrossRef] [Medline]
    27. Suk HI, Lee SW, Shen D. Hierarchical feature representation and multimodal fusion with deep learning for AD/MCI diagnosis. Neuroimage. Nov 1, 2014;101:569-582. [CrossRef] [Medline]
    28. Sayeed A, Petrou M, Spyrou N, Kadyrov A, Spinks T. Diagnostic features of Alzheimer’s disease extracted from PET sinograms. Phys Med Biol. Jan 7, 2002;47(1):137-148. [CrossRef] [Medline]
    29. Pan X, Adel M, Fossati C, et al. Multiscale spatial gradient features for 18F-FDG PET image-guided diagnosis of Alzheimer’s disease. Comput Methods Programs Biomed. Oct 2019;180:105027. [CrossRef] [Medline]
    30. Padilla P, López M, Górriz JM, et al. NMF-SVM based CAD tool applied to functional brain images for the diagnosis of Alzheimer’s disease. IEEE Trans Med Imaging. Feb 2012;31(2):207-216. [CrossRef] [Medline]
    31. Ni YC, Tseng FP, Pai MC, et al. Detection of Alzheimer’s disease using ECD SPECT images by transfer learning from FDG PET. Ann Nucl Med. Aug 2021;35(8):889-899. [CrossRef] [Medline]
    32. Magnin B, Mesrob L, Kinkingnéhun S, et al. Support vector machine-based classification of Alzheimer’s disease from whole-brain anatomical MRI. Neuroradiology. Feb 2009;51(2):73-83. [CrossRef] [Medline]
    33. Lu DH, Popuri K, Ding GW, Balachandar R, Beg MF. Multiscale deep neural network based analysis of FDG-PET images for the early diagnosis of Alzheimer’s disease. Med Image Anal. May 2018;46:26-34. [CrossRef]
    34. Liu MH, Zhang DQ, Shen DG. Ensemble sparse classification of Alzheimer’s disease. Neuroimage. Apr 2012;60(2):1106-1116. [CrossRef]
    35. Liu M, Cheng D, Yan W, Alzheimer’s Disease Neuroimaging Initiative. Classification of Alzheimer’s disease by combination of convolutional and recurrent neural networks using FDG-PET images. Front Neuroinform. 2018;12:35. [CrossRef] [Medline]
    36. Li R, Perneczky R, Yakushev I, et al. Gaussian mixture models and model selection for [18F] fluorodeoxyglucose positron emission tomography classification in Alzheimer’s disease. PLoS ONE. Apr 2015;10(4):e0122731. [CrossRef]
    37. Lerch JP, Pruessner J, Zijdenbos AP, et al. Automated cortical thickness measurements from MRI can accurately separate Alzheimer’s patients from normal elderly controls. Neurobiol Aging. Jan 2008;29(1):23-30. [CrossRef] [Medline]
    38. Kim HW, Lee HE, Oh K, Lee S, Yun M, Yoo SK. Multi-slice representational learning of convolutional neural network for Alzheimer’s disease classification using positron emission tomography. Biomed Eng Online. Sep 7, 2020;19(1):70. [CrossRef] [Medline]
    39. Kim HW, Lee HE, Lee S, Oh KT, Yun M, Yoo SK. Slice-selective learning for Alzheimer’s disease classification using a generative adversarial network: a feasibility study of external validation. Eur J Nucl Med Mol Imaging. Aug 2020;47(9):2197-2206. [CrossRef] [Medline]
    40. Katako A, Shelton P, Goertzen AL, et al. Machine learning identified an Alzheimer’s disease-related FDG-PET pattern which is also expressed in Lewy body dementia and Parkinson’s disease dementia. Sci Rep. Sep 5, 2018;8(1):13236. [CrossRef] [Medline]
    41. Ismail WN, P. P. FR, Ali MAS. A meta-heuristic multi-objective optimization method for Alzheimer’s disease detection based on multi-modal data. Mathematics. Feb 2023;11(4):957. [CrossRef]
    42. Illán IA, Górriz JM, Ramírez J, et al. 18F-FDG PET imaging analysis for computer aided Alzheimer’s diagnosis. Inf Sci (Ny). Feb 2011;181(4):903-916. [CrossRef]
    43. Hinrichs C, Singh V, Mukherjee L, et al. Spatially augmented LPboosting for AD classification with evaluations on the ADNI dataset. Neuroimage. Oct 15, 2009;48(1):138-149. [CrossRef] [Medline]
    44. Gray KR, Wolz R, Heckemann RA, et al. Multi-region analysis of longitudinal FDG-PET for the classification of Alzheimer’s disease. Neuroimage. Mar 2012;60(1):221-229. [CrossRef] [Medline]
    45. Gray KR, Aljabar P, Heckemann RA, Hammers A, Rueckert D. Random forest-based similarity measures for multi-modal classification of Alzheimer’s disease. Neuroimage. Jan 15, 2013;65:167-175. [CrossRef] [Medline]
    46. Feng C, Elazab A, Yang P, et al. Deep learning framework for Alzheimer’s disease diagnosis via 3D-CNN and FSBi-LSTM. IEEE Access. 2019;7:63605-63618. [CrossRef]
    47. Cuingnet R, Gerardin E, Tessieras J, et al. Automatic classification of patients with Alzheimer’s disease from structural MRI: a comparison of ten methods using the ADNI database. Neuroimage. May 15, 2011;56(2):766-781. [CrossRef] [Medline]
    48. Chen L, Qiao HZ, Zhu F. Alzheimer’s disease diagnosis with brain structural MRI using multiview-slice attention and 3D convolution neural network. Front Aging Neurosci. 2022;14:871706. [CrossRef] [Medline]
    49. Song J, Zheng J, Li P, Lu X, Zhu G, Shen P. An effective multimodal image fusion method using MRI and PET for Alzheimer’s disease diagnosis. Front Digit Health. 2021;3:637386. [CrossRef] [Medline]
    50. Li Y, Jiang J, Lu J, Jiang J, Zhang H, Zuo C. Radiomics: a novel feature extraction method for brain neuron degeneration disease using 18F-FDG PET imaging and its implementation for Alzheimer’s disease and mild cognitive impairment. Ther Adv Neurol Disord. 2019;12:1756286419838682. [CrossRef] [Medline]
    51. A A, M P, Hamdi M, Bourouis S, Rastislav K, Mohmed F. Evaluation of neuro images for the diagnosis of Alzheimer’s disease using deep learning neural network. Front Public Health. 2022;10:834032. [CrossRef] [Medline]
    52. Toussaint PJ, Perlbarg V, Bellec P, et al. Resting state FDG-PET functional connectivity as an early biomarker of Alzheimer’s disease using conjoint univariate and independent component analyses. Neuroimage. Nov 1, 2012;63(2):936-946. [CrossRef] [Medline]
    53. Tong T, Wolz R, Gao Q, et al. Multiple instance learning for classification of dementia in brain MRI. Med Image Anal. Jul 2014;18(5):808-818. [CrossRef] [Medline]
    54. Min R, Wu G, Cheng J, Wang Q, Shen D. Multi-atlas based representations for Alzheimer’s disease diagnosis. Hum Brain Mapp. Oct 2014;35(10):5052-5070. [CrossRef] [Medline]
    55. Jin D, Zhou B, Han Y, et al. Generalizable, reproducible, and neuroscientifically interpretable imaging biomarkers for Alzheimer’s disease. Adv Sci (Weinh). Jul 2020;7(14):2000675. [CrossRef] [Medline]
    56. Cho Y, Seong JK, Jeong Y, Shin SY. Individual subject classification for Alzheimer’s disease based on incremental learning using a spatial frequency representation of cortical thickness data. Neuroimage. Feb 1, 2012;59(3):2217-2230. [CrossRef] [Medline]
    57. Chincarini A, Bosco P, Calvini P, et al. Local MRI analysis approach in the diagnosis of early and prodromal Alzheimer’s disease. Neuroimage. Sep 15, 2011;58(2):469-480. [CrossRef] [Medline]
    58. Beheshti I, Demirel H, Matsuda H. Classification of Alzheimer’s disease and prediction of mild cognitive impairment-to-Alzheimer’s conversion from structural magnetic resource imaging using feature ranking and a genetic algorithm. Comput Biol Med. Apr 1, 2017;83(109-19):109-119. [CrossRef] [Medline]
    59. Anandh KR, Sujatha CM, Ramakrishnan S. A method to differentiate mild cognitive impairment and Alzheimer in MR images using eigen value descriptors. J Med Syst. Jan 2016;40(1):26547845. [CrossRef] [Medline]
    60. Amoroso N, La Rocca M, Bruno S, et al. Multiplex networks for early diagnosis of Alzheimer’s disease. Front Aging Neurosci. 2018;10:365. [CrossRef] [Medline]
    61. Yamane T, Ikari Y, Nishio T, et al. Visual-statistical interpretation of (18)F-FDG-PET images for characteristic Alzheimer patterns in a multicenter study: inter-rater concordance and relationship to automated quantitative evaluation. AJNR Am J Neuroradiol. Feb 2014;35(2):244-249. [CrossRef] [Medline]
    62. Harper L, Fumagalli GG, Barkhof F, et al. MRI visual rating scales in the diagnosis of dementia: evaluation in 184 post-mortem confirmed cases. Brain (Bacau). Apr 2016;139(Pt 4):1211-1225. [CrossRef] [Medline]
    63. Loddo A, Buttau S, Di Ruberto C. Deep learning based pipelines for Alzheimer’s disease diagnosis: a comparative study and a novel deep-ensemble method. Comput Biol Med. Feb 2022;141:105032. [CrossRef] [Medline]
    64. Pawar U, O’Shea D, Rea S, O’Reilly R. Explainable AI in healthcare. Presented at: 2020 International Conference on Cyber Situational Awareness, Data Analytics and Assessment (CyberSA); Jun 15-19, 2020; Dublin, Ireland. [CrossRef]
    65. Birkenbihl C, Emon MA, Vrooman H, et al. Differences in cohort study data affect external validation of artificial intelligence models for predictive diagnostics of dementia - lessons for translation into clinical practice. EPMA J. Sep 2020;11(3):367-376. [CrossRef] [Medline]
    66. Barredo Arrieta A, Díaz-Rodríguez N, Del Ser J, et al. Explainable Artificial Intelligence (XAI): Concepts, taxonomies, opportunities and challenges toward responsible AI. Information Fusion. Jun 2020;58:82-115. [CrossRef]
    67. Mayerhoefer ME, Prosch H, Beer L, et al. PET/MRI versus PET/CT in oncology: a prospective single-center study of 330 examinations focusing on implications for patient management and cost considerations. Eur J Nucl Med Mol Imaging. Jan 2020;47(1):51-60. [CrossRef] [Medline]
    68. Goldberg CB, Adams L, Blumenthal D, et al. To do no harm - and the most good - with AI in health care. Nat Med. Mar 2024;30(3):623-627. [CrossRef] [Medline]
    69. Frizzell TO, Glashutter M, Liu CC, et al. Artificial intelligence in brain MRI analysis of Alzheimer’s disease over the past 12 years: A systematic review. Ageing Res Rev. May 2022;77:101614. [CrossRef] [Medline]
    70. Borchert RJ, Azevedo T, Badhwar A, et al. Artificial intelligence for diagnostic and prognostic neuroimaging in dementia: A systematic review. Alzheimers Dement. Dec 2023;19(12):5885-5904. [CrossRef] [Medline]
    71. Sun Y, Chen Y, Dong L, et al. Diagnostic performance of deep learning-assisted [18F]FDG PET imaging for Alzheimer’s disease: a systematic review and meta-analysis. Eur J Nucl Med Mol Imaging. Aug 2025;52(10):3600-3612. [CrossRef] [Medline]

    ‎
    18FDG-PET: fluorine-18 fluorodeoxyglucose positron emission tomography
    AD: Alzheimer disease
    AI: artificial intelligence
    DL: deep learning
    MeSH: Medical Subject Headings
    ML: machine learning
    PICOS: Population, Intervention, Comparison, Outcome, Study Design
    PRISMA-DTA: Preferred Reporting Items for Systematic Reviews and Meta-Analyses of Diagnostic Test Accuracy
    sMRI: structural magnetic resonance imaging
    SROC-AUC: summary receiver operating characteristic curve area
    TRIPOD-AI: Transparent Reporting of a Multivariable Prediction Model for Individual Prognosis or Diagnosis–Artificial Intelligence

    Edited by Megan O'Connell; submitted 05.May.2025; peer-reviewed by Ravi Teja Potla, Sadhasivam Mohanadas, Seyed Ali Mirshahvalad; final revised version received 03.Jul.2025; accepted 20.Jul.2025; published 08.Oct.2025.

    Copyright

    © Bingbing Wang, Tailiang Zhao, Rongrong Ma, Xiaochuan Huo, Xiaoxiao Xiong, Minjie Wu, Yuran Wang, Liu Liu, Zhijiang Zhuang, Bin Wang, Jixin Shou. Originally published in JMIR Aging (https://aging.jmir.org), 8.Oct.2025.

    This is an open-access article distributed under the terms of the Creative Commons Attribution License (https://creativecommons.org/licenses/by/4.0/), which permits unrestricted use, distribution, and reproduction in any medium, provided the original work, first published in JMIR Aging, is properly cited. The complete bibliographic information, a link to the original publication on https://aging.jmir.org, as well as this copyright and license information must be included.