Table 1. Comprehensive summary of machine learning and artificial intelligence applications in Alzheimer's disease diagnosis and progression prediction.

CNN: convolutional neural network, MRI: magnetic resonance imaging, MCI: mild cognitive impairment, AD: Alzheimer's disease, APOE: apolipoprotein E, ML: machine learning, EEG: electroencephalogram, cfDNA: cell-free deoxyribonucleic acid, PET: positron emission tomography, FDG: fludeoxyglucose-18, OASIS: Open Access Series of Imaging Studies, KNN: K-nearest neighbor, AUC: area under the curve, DAT: dementia of AD type, VEGF-A: vascular endothelial growth factor A, LSR: lesion-to-spinal cord signal intensity ratio, AUROC: area under the ROC curve, rs-fMRI: resting state task-based fMRI, Smile-GAN: semi-supervised clustering GAN, SCD: subjective cognitive decline, SVM: support vector machine, PCA: principal component analysis, LASSO: least absolute shrinkage and selection operator, NLP: natural language processing, SbRNS: sandpiper-based recurrent neural system

Table credits: Gurnoor Gill

Author(s)	Key findings	Unique characteristics
GA Aguayo et al., 2023 [1]	Machine learning predicts neurodegenerative diseases in older populations.	TabTransformer model applied in a cohort study.
D AlSaeed et al., 2022 [2]	CNN-based MRI feature extraction achieved high AD classification accuracy.	Automated CNN applied to neuroimaging datasets.
Rye A et al., 2022 [3]	Predictive modeling for MCI to AD conversion using biomarkers.	Biomarkers validated via APOE and hippocampal volume.
N J Herzog et al., 2021 [4]	Brain asymmetry features identified for AD diagnosis.	Neuroanatomical asymmetry is used in ML classification.
H Yu, 2021 [5]	Platelet biomarkers linked to cognitive decline using proteomics.	Proteomic markers revealed novel pathways for AD.
J Sheng et al., 2022 [6]	Genetic and imaging data achieved 98% classification accuracy.	Multimodal diagnostic approach for AD.
T W Rowe et al., 2021 [8]	Life-time risk prediction models for AD were reviewed systematically.	Focused on ML models’ predictive performance.
N Chedid et al., 2022 [9]	EEG-based ML pipeline achieved 81% accuracy for AD.	Artifact-free EEG data processing validated.
CY Cheung et al., 2022 [10]	Retinal biomarkers linked with AD detection achieved 83.6% accuracy.	Retinal imaging is proposed as a non-invasive diagnostic tool.
RO Bahado-Singh et al., 2022 [11]	cfDNA methylation patterns identified for early AD detection.	Highlighted epigenetic mechanisms as biomarkers.
PR Millar et al.,2023 [12]	Brain age biomarkers linked to cognitive decline and AD stages.	Multimodal MRI imaging combined with functional biomarkers.
M Odusami et al., 2022 [13]	ML framework achieved high accuracy for early AD detection.	DenseNet and ResNet integrated with MRI.
L Chiricosta et al., 2022 [14]	Blood transcriptome data linked oxidative stress to AD pathology.	Transcriptomics-based machine learning predictive models developed.
A A et al., 2022 [15]	Automated PET neuroimaging enhanced AD early-stage detection.	CNN frameworks applied to PET imaging.
I Beheshti et al., 2022 [16]	FDG-PET imaging tracked MCI progression to AD.	PET scans were validated for metabolic activity metrics.
X Wang et al., 2022 [17]	Retinal biomarkers correlated with cognitive decline.	Eye-based imaging is proposed as a scalable diagnostic method.
A Taylor 15 al., 2022 [18]	Longitudinal imaging patterns predicted AD-related neurodegeneration.	Combined brain age biomarkers with explainable AI models.
C Kavitha et al., 2022 [19]	Gradient boosting models improved AD early-stage prediction.	ML classifiers applied to OASIS datasets.
JB Toledo et al., 2022 [20]	SPARE-Tau indices predicted cognitive decline stages in AD.	Imaging biomarkers validated for clinical utility.
Y M Elgammal et al., 2022 [21]	Multifractal KNN achieved superior classification accuracy in AD MRI imaging.	Geometry-based classification methods validated for AD.
Diogo et al., 2022 [22]	Multi-diagnostic ML approach for AD using MRI, achieving 90.6% accuracy in distinguishing healthy controls and AD.	Hippocampal features as major contributors; generalizable across datasets and MRI protocols.
Gao et al., 2023 [23]	PRS and EHR data predicted AD with an AUC of 0.88 in a large UK Biobank cohort.	PRSs for age-at-onset were more predictive than age; novel feature importance patterns were identified.
Mirabnahrazam et al., 2022 [24]	Combined MRI and genetic data for DAT prediction showed improved performance in distinguishing DAT progression.	Novel multimodal stratification method for DAT; detailed analysis of imaging and genetic contributions.
Sekaran et al., 2023 [25]	Identified the ORAI2 gene as a blood-based biomarker for AD using explainable AI.	AI-identified STIM1 and TRPC3 as interacting genes linked to AD progression.
Petrelis et al., 2022 [26]	VEGF-A-related genetic variants were found to protect against AD progression.	A model with epistatic interactions between VEGF-A, APOE, and LSR demonstrated 72% accuracy.
Feng et al., 2023 [27]	Blood-based metabolic pathway signatures developed for non-invasive AD diagnosis.	Distinct AD subgroups identified with varying metabolic and immune profiles; achieved AUC of 0.99 in validation.
Vik et al., 2023 [28]	Subtle changes in daily functioning predicted AD progression from MCI with 70% accuracy.	Informant-reported functional activity levels and verbal memory were highlighted as key predictors.
Kobayashi et al., 2022 [29]	Multi-task drawing tests improved early AD detection to 75.2% accuracy.	Combined drawing data captured multiple cognitive impairments; automated analysis scalable to non-specialist settings.
Zhang et al., 2021 [30]	3D CNN and ensemble learning achieved 95.2% accuracy in AD vs. normal controls.	Innovative data denoising module; high performance in MRI-based AD diagnosis.
Feng et al., 2022 [31]	Deep learning outperformed other biomarkers for prodromal AD detection (AUROC = 0.788).	Model localized hippocampal activation; reduced patient burden and cost through non-invasive MRI.
Alorf and Khan, 2022 [32]	Multi-label classification of six AD stages from rs-fMRI using graph convolutional networks achieved 84.03% accuracy.	Identified key brain regions (e.g., frontal gyrus) for differentiating AD stages.
Yang et al., 2021 [33]	Smile-GAN identified subtypes of neurodegeneration and predicted AD progression pathways.	Semi-supervised clustering via GAN offered precision diagnostics and informed clinical trials.
Guan et al., 2023 [34]	Attention-guided autoencoder accurately predicted SCD progression to MCI or AD.	Leveraged domain transfer learning for small datasets and localized disease-specific brain regions.
Lai et al., 2022 [35]	Identified immune subtypes and genes linked to AD pathology using explainable machine learning.	CXCR4, PPP3R1, HSP90AB1, CXCL10, and S100A12 genes are linked to distinct immune subtypes of AD.
Prasad VK et al., 2024 [36]	CNN achieved 99.29% accuracy for Alzheimer’s diagnosis using MRI data.	Proposed a cloud-based framework integrating preprocessing, execution, and diagnostic layers, emphasizing scalability and data security.
Mofrad et al., 2021 [37]	Combined cognitive trajectories and MRI features to predict AD conversion from MCI with improved accuracy.	Longitudinal mixed-effects modeling for both cognitive and structural brain measures.
Tian et al., 2021 [38]	Retinal vasculature features provided an alternative biomarker for AD diagnosis with 82.44% accuracy.	Highlighted the utility of retinal imaging as non-invasive and cost-effective.
Bron et al., 2021 [39]	Validated cross-cohort generalizability of SVM and CNN for AD classification, showing comparable performance.	Demonstrated robustness in external validation with similar AUC scores for both models.
Muñoz-Castro et al., 2022 [40]	Machine learning models identified distinct astrocyte and microglia states in AD and normal aging.	Multiplex fluorescent immunohistochemistry captured spatial relationships with AD pathology.
Kim et al., 2022 [41]	Deep learning (VUNO Med-DeepBrain) achieved 87.1% accuracy in diagnosing AD with 2D brain MRI.	Improved sensitivity (93.3%) and specificity (85.5%) compared to medical experts; scalable for non-specialist settings.
Li et al., 2022 [42]	Proposed a classification framework for complex, imbalanced data using functional PCA and group LASSO for early AD.	Achieved high sensitivity for longitudinal and high-dimensional data; focused on early detection challenges in AD.
Liu et al., 2022 [43]	Developed a transfer learning model using speech and NLP for early AD detection, achieving 88% accuracy.	Combined pre-trained DistilBERT and logistic regression for robust language feature extraction and binary classification.
Swarnalatha, 2023 [44]	Proposed SbRNS model for EEG-based AD severity prediction with precision and recall above conventional methods.	Addressed noise in EEG signals with advanced pre-processing; classified severity into low, medium, and high categories.
Bogdanovic et al., 2022 [45]	XGBoost model analyzed 12,000+ subjects, yielding 0.84 F1 score and explainable insights into AD diagnosis.	Used Shapley values for feature importance; focused on gender, APOE4, and age as key diagnostic predictors.