Explainable Deep Learning for Alzheimers Diagnosis: A Comprehensive Review of Models, Methods, and Clinical Relevance

Abstract

Alzheimer’s Disease (AD) is a progressive neu-rodegenerative condition mostly affecting the elderly, causing cognitive decline. It is an active field of research as currently there is no approved cure. Current breakthroughs in deep learn-ing have been promising for AD detection from neuroimaging and clinical information. The ”black-box” limitation of such models is a concern regarding trust and clinical uptake. This research investigates the potential of integrating deep learning with Explainable AI (XAI) to resolve these issues. It outlines a systematic review of the literature from 2015 to 2024, organizing studies into four categories: general reviews, models that lack explainability, models with the integration of partial or complete XAI, and explainability methods. The article compares methods, datasets (ADNI, NACC, AIBL), and main results, emphasizing advancesinvisualinterpretabilityyetpointingtowardspersisting issues like a deficiency of standard metrics in evaluation, sparse clinical validation, and limited generalizability. Several studies lackrule-basedexplanationsorfailtoquantifythereliabil-ityofmodeldecisionsandthereisanecessitytointegrate more longitudinal, multimodal, and demographically diversified datainADresearch.Thisreviewhighlightstheimportanceof standardised, clinically validated explainability frameworksto connect research with real-world usage. It seeks to direct future efforts in the creation of responsible, interpretable AItoolsforhealthcare,providingbeneficialinsightsforresearchers, clinicians, and policymakers.

Introduction

Alzheimer’s Disease (AD) is a progressive neurological disorder and one of the leading causes of dementia worldwide. It mainly affects older adults and causes gradual decline in memory, thinking ability, and daily functioning. Brain changes associated with Alzheimer’s begin years before visible symptoms appear, making early detection extremely important. Current diagnosis methods involve clinical evaluation, cognitive tests, neurological examinations, MRI, and PET imaging. However, there is no permanent cure for AD, so early identification can help in better treatment planning and disease management.

Recent advancements in Artificial Intelligence (AI), Machine Learning (ML), and Deep Learning (DL) have improved Alzheimer’s diagnosis by detecting hidden patterns in brain images, clinical records, and biological data. Deep learning models such as CNNs, RNNs, transformers, and graph neural networks have achieved high diagnostic accuracy using MRI, PET, EEG, and electronic health record (EHR) data. However, many AI models function as black-box systems, meaning their decision-making process is unclear, which reduces trust and limits clinical adoption.

To solve this problem, Explainable Artificial Intelligence (XAI) techniques are being integrated into healthcare AI systems. XAI provides understandable explanations for model predictions, helping doctors identify important biomarkers and validate AI decisions. Common XAI methods include:

• SHAP (Shapley Additive Explanations): Explains feature importance and contribution to predictions.
• LIME (Local Interpretable Model-Agnostic Explanations): Provides local explanations for individual predictions.
• Grad-CAM: Generates visual heatmaps showing important brain regions influencing decisions.
• Integrated Gradients, DeepLIFT, and Layer-wise Relevance Propagation (LRP): Provide deeper insights into neural network decisions.

Objectives of the Review:

The study reviews research from 2015–2024 to:

• Analyze AI-based approaches for early Alzheimer’s detection.
• Study XAI techniques used for clinical interpretation.
• Compare different explainability methods for imaging and clinical data.
• Identify research challenges and future directions.

Literature Review Findings:

1. AI/ML Models Without XAI:
Deep learning models such as CNNs, ensemble networks, transfer learning models, and lightweight architectures have shown high accuracy in AD diagnosis. MRI-based models using datasets like ADNI and OASIS have successfully detected AD and Mild Cognitive Impairment (MCI). However, these models lack transparency because they cannot explain why a prediction was made.

2. AI Models with Explainability:
Researchers have combined AI models with XAI techniques to improve reliability:

• Grad-CAM and heatmaps identify disease-related brain regions.
• SHAP and LIME explain important clinical and imaging features.
• Attention-based and graph-based models improve visualization of brain relationships.
• Multimodal approaches combine MRI, PET, genetics, cognitive scores, and EHR data for better prediction.

Some advanced models such as AXIAL use attention mechanisms and voxel-level explanations to identify important brain areas including the hippocampus, parahippocampus, and amygdala, matching clinical observations.

Major Research Gaps:

Despite significant progress, several challenges remain:

• Lack of standardized methods to evaluate XAI explanations.
• Many studies rely on single datasets, limiting generalization.
• Limited clinical validation of AI predictions.
• Difficulty distinguishing stable MCI from progressive MCI.
• Small and imbalanced datasets affect reliability.
• Lack of rule-based explanations from heatmaps.
• Inconsistent reporting of model performance and explainability results.

Deep Learning Techniques in AD Diagnosis:

Commonly used approaches include:

• CNN models: Detect structural changes in MRI/PET images.
• Transfer learning: Improves performance when datasets are limited.
• Ensemble models: Combine multiple models for better accuracy.
• Transformers and attention models: Capture complex patterns in sequential medical data.
• Graph Neural Networks (GNNs): Analyze relationships between brain regions.
• Autoencoders and GANs: Extract features and generate additional data.

Applications of XAI in Alzheimer’s Detection:

XAI helps doctors by:

• Identifying important brain regions affected by AD.
• Explaining patient risk factors.
• Improving trust in AI-assisted diagnosis.
• Supporting ethical and transparent healthcare decisions.

Conclusion

ThisreviewhighlightstheincreasingoverlapbetweenAI and healthcare research, particularly in the context of Alzheimer’s Disease.Recent advancements in Alzheimer’s Disease (AD) diagnosis show deep learning models’ promise when applied to datasets such as ADNI, AIBL, and NACC. These models have been achieving outstanding precision in diagnostics. Explainability is stressed now to improve model transparency for all. This emphasis helps in building trust. Still, there are challenges that persist despite all of these developments.Sincemanyheatmap-basedoutputsdostill fail to provide actionable rule-based perceptions, the fieldnowlacksstandardizedbenchmarksforanyevaluatingof explainability methods.Smallorimbalanceddatasetsgenerate concern, plus underrepresentation of diverse patient groups and limited external validation generate concern.

References

[1] S.Bach,A.Binder,G.Montavon,F.Klauschen,K.-R.Mu¨ller,andW.Samek,“OnPixel-WiseExplanationsforNon LinearClassifierDecisionsbyLayer-WiseRelevancePropagation,”PLoSOne,vol.10,no.7,p.e0130140,Jul.2015,doi: https://doi.org/10.1371/journal.pone.013014010.1371/journal.pone.0130140. [2] B.Zhou,A.Khosla,A.Lapedriza,A.Oliva,and a. Torralba, “Learning Deep Features for DiscriminativeLocalization,” arXiv preprint arXiv:1512.04150, Dec. 2015, doi:https://doi.org/10.48550/arXiv.1512.0415010.48550/arXiv.1512.04150. [3] M.T.Ribeiro,S.Singh,andC.Guestrin,“‘WhyShouldITrustYou?’:ExplainingthePredictionsofAnyClas-sifier,” arXiv preprint arXiv:1602.04938, Aug. 2016, doi:https://doi.org/10.48550/arXiv.1602.0493810.48550/arXiv.1602.04938. [4] S.LundbergandS.-I.Lee,“AUnifiedApproachtoInterpretingModel Predictions,” arXiv preprint arXiv:1705.07874, Nov. 2017, doi:https://doi.org/10.48550/arXiv.1705.0787410.48550/arXiv.1705.07874. [5] J. Islam and Y. Zhang, “Brain MRI analysis for Alzheimer’s diseasediagnosis using an ensemble system of deep convolutional neuralnetworks,” Brain Informatics, vol. 5, no. 2, p. 2, May 2018, doi:https://doi.org/10.1186/s40708-018-0080-310.1186/s40708-018-0080-3. [6] R.Zhu,F.Dornaika,andY.Ruichek,“Learningadiscriminantgraph-based embedding with feature selection for imagecategorization,” Neural Networks, vol. 111, pp. 35–46, Mar. 2019, doi:https://doi.org/10.1016/j.neunet.2018.12.00810.1016/j.neunet.2018.12.008. [7] A.Shrikumar,P.Greenside,andA.Kundaje,“LearningImportantFeaturesThroughPropagatingActivationDiffer-ences,” arXiv preprint arXiv:1704.02685, Oct. 2019, doi:https://doi.org/10.48550/arXiv.1704.0268510.48550/arXiv.1704.02685. [8] E. Lee, J.-S. Choi, M. Kim, and H.-I. Suk, “Toward an interpretableAlzheimer’s disease diagnostic model with regional abnormality repre-sentation via deep learning,” NeuroImage, vol. 202, p. 116113, Nov.2019, doi: 10.1016/j.neuroimage.2019.116113. [9] R.R.Selvaraju,M.Cogswell,A.Das,R.Vedantam,D.Parikh,and D. Batra, “Grad-CAM: Visual Explanations from Deep Networks viaGradient-based Localization,” Int J Comput Vis, vol. 128, no. 2, pp.336–359, Feb. 2020, doi: 10.1007/s11263-019-01228-7. [10] E.Nigri,N.Ziviani,F.Cappabianco,A.Antunes,andA.Veloso, “Explainable Deep CNNs for MRI-Based Diagnosis ofAlzheimer’s Disease,” Apr. 25, 2020, arXiv: arXiv:2004.12204. doi:10.48550/arXiv.2004.12204. [11] F. Viton, M. Elbattah, J.-L. Gue´rin, and G. Dequen, “Heatmaps forVisual Explainability of CNN-Based Predictions for Multivariate TimeSeries with Application to Healthcare,” in 2020 IEEE InternationalConference on Healthcare Informatics (ICHI), Nov. 2020, pp. 1–8. doi:10.1109/ICHI48887.2020.9374393. [12] D. Dave, H. Naik, S. Singhal, and P. Patel, “Explainable AI meetsHealthcare: A Study on Heart Disease Dataset,” Nov. 06, 2020, arXiv:arXiv:2011.03195. doi: 10.48550/arXiv.2011.03195. [13] K. Yang and E. A. Mohammed, “A Review of Artificial IntelligenceTechnologies for Early Prediction of Alzheimer’s Disease,” Dec. 22,2020, arXiv: arXiv:2101.01781. doi: 10.48550/arXiv.2101.01781. [14] T.Abuhmed,S.El-Sappagh,andJ.M.Alonso,“Robusthy-brid deep learning models for Alzheimer’s progression detection,”Knowledge-Based Systems, vol. 213, p. 106688, Feb. 2021, doi:10.1016/j.knosys.2020.106688. [15] J. Liu, M. Li, Y. Luo, S. Yang, W. Li, and Y. Bi, “Alzheimer’s diseasedetection using depthwise separable convolutional neural networks,”Computer Methods and Programs in Biomedicine, vol. 203, p. 106032,May 2021, doi: 10.1016/j.cmpb.2021.106032. [16] A.Yadu,P.K.Suhas,andN.Sinha,“ClassSpecificInterpretabilityin CNN Using Causal Analysis,” in 2021 IEEE International Con-ference on Image Processing (ICIP), Sep. 2021, pp. 3702–3706. doi:10.1109/ICIP42928.2021.9506118. [17] M. Sidulova, N. Nehme, and C. H. Park, “Towards Explainable ImageAnalysis for Alzheimer’s Disease and Mild Cognitive Impairment Di-agnosis,”in2021IEEEAppliedImageryPatternRecognitionWorkshop(AIPR), Oct. 2021, pp. 1–6. doi: 10.1109/AIPR52630.2021.9762082. [18] J. V. Shanmugam, B. Duraisamy, B. C. Simon, and P. Bhaskaran,“Alzheimer’s disease classification using pre-trained deep networks,”Biomedical Signal Processing and Control, vol. 71, p. 103217, Jan.2022, doi: 10.1016/j.bspc.2021.103217. [19] A. Raza, K. P. Tran, L. Koehl, and S. Li, “Designing ECG monitoringhealthcaresystemwithfederatedtransferlearningandexplainableAI,” Knowledge-Based Systems, vol. 236, p. 107763, Jan. 2022, doi:10.1016/j.knosys.2021.107763. [20] A. Loddo, S. Buttau, and C. Di Ruberto, “Deep learning based pipelinesfor Alzheimer’s disease diagnosis: A comparative study and a noveldeep-ensemble method,” Computers in Biology and Medicine, vol. 141, p.105032,Feb.2022,doi:10.1016/j.compbiomed.2021.105032. [21] Z. Liu, H. Lu, X. Pan, M. Xu, R. Lan, and X. Luo, “Diagnosis ofAlzheimer’s disease via an attention-based multi-scale convolutionalneural network,” Knowledge-Based Systems, vol. 238, p. 107942, Feb.2022, doi: 10.1016/j.knosys.2021.107942. [22] F. Salami, A. Bozorgi-Amiri, G. M. Hassan, R. Tavakkoli-Moghaddam,and A. Datta, “Designing a clinical decision support system forAlzheimer’s diagnosis on OASIS-3 data set,” Biomedical SignalProcessing and Control, vol. 74, p. 103527, Apr. 2022, doi:10.1016/j.bspc.2022.103527. [23] B. Bogdanovic, T. Eftimov, and M. Simjanoska, “In-depth insights intoAlzheimer’s disease by using explainable machine learning approach,”SciRep,vol.12,no.1,p.6508,Apr.2022,doi:10.1038/s41598-022- 10202-2. [24] R. Han, Z. Liu, and C. L. P. Chen, “Multi-scale 3D convolution feature-based Broad Learning System for Alzheimer’s Disease diagnosis viaMRI images,” Applied Soft Computing, vol. 120, p. 108660, May 2022,doi: 10.1016/j.asoc.2022.108660. [25] L.Wangetal.,“Dementiaanalysisfromfunctionalconnectivitynetworkwith graph neural networks,” Information Processing & Management,vol. 59, no. 3, p. 102901, May 2022, doi: 10.1016/j.ipm.2022.102901. [26] F. Zhang et al., “A Single Model Deep Learning Approach forAlzheimer’s Disease Diagnosis,” Neuroscience, vol. 491, pp. 200–214,May 2022, doi: 10.1016/j.neuroscience.2022.03.026. [27] L. Zeng, H. Li, T. Xiao, F. Shen, and Z. Zhong, “Graph convolutionalnetwork with sample and feature weights for Alzheimer’s diseasediagnosis,” Information Processing & Management, vol. 59, no. 4, p.102952, Jul. 2022, doi: 10.1016/j.ipm.2022.102952. [28] M. Khojaste-Sarakhsi, S. S. Haghighi, S. M. T. F. Ghomi, and E.Marchiori,“DeeplearningforAlzheimer’sdiseasediagnosis:Asurvey,”Artificial Intelligence in Medicine, vol. 130, p. 102332, Aug. 2022, doi:10.1016/j.artmed.2022.102332. [29] Z. Y?lmaz Acar, F. Bas¸c¸iftc¸i, and A. H. Ekmekci, “Future activityprediction of multiple sclerosis with 3D MRI using 3D discrete wavelettransform,” Biomedical Signal Processing and Control, vol. 78, p.103940, Sep. 2022, doi: 10.1016/j.bspc.2022.103940. [30] L.Yu,W.Xiang,J.Fang,Y.-P.PhoebeChen,andR.Zhu,“AnovelexplainableneuralnetworkforAlzheimer’sdiseasediagno-sis,” Pattern Recognition, vol. 131, p. 108876, Nov. 2022, doi:10.1016/j.patcog.2022.108876. [31] M. Menagadevi, S. Mangai, N. Madian, and D. Thiyagarajan, “Auto-matedpredictionsystemforAlzheimerdetectionbasedondeepresidualautoencoder and support vector machine,” Optik, vol. 272, p. 170212,Feb. 2023, doi: 10.1016/j.ijleo.2022.170212. [32] H.-D. Nguyen, M. Cle´ment, B. Mansencal, and P. Coupe´, “TowardsbetterInterpretableandGeneralizableADdetectionusingCollectiveAr-tificialIntelligence,”ComputerizedMedicalImagingandGraphics,vol.104, p. 102171, Mar. 2023, doi: 10.1016/j.compmedimag.2022.102171. [33] D. Wang et al., “Deep neural network heatmaps capture Alzheimer’sdiseasepatternsreportedinalargemeta-analysisofneuroimag-ing studies,” NeuroImage, vol. 269, p. 119929, Apr. 2023, doi:10.1016/j.neuroimage.2023.119929. [34] F.Liuetal.,“MPS-FFA:Amultiplaneandmultiscalefeaturefu-sion attention network for Alzheimer’s disease prediction with struc-tural MRI,” Comput Biol Med, vol. 157, p. 106790, May 2023, doi:10.1016/j.compbiomed.2023.106790. [35] S. Shojaei, M. Saniee Abadeh, and Z. Momeni, “An evolutionaryexplainabledeeplearningapproachforAlzheimer’sMRIclassification,”Expert Systems with Applications, vol. 220, p. 119709, Jun. 2023, doi:10.1016/j.eswa.2023.119709. [36] Z. Yao et al., “Artificial intelligence-based diagnosis of Alzheimer’sdisease with brain MRI images,” Eur J Radiol, vol. 165, p. 110934,Aug. 2023, doi: 10.1016/j.ejrad.2023.110934. [37] G. Ekuma, D. B. Hier, and T. Obafemi-Ajayi, “An Explainable DeepLearning Model for Prediction of Severity of Alzheimer’s Disease,” in2023IEEEConferenceonComputationalIntelligenceinBioinformaticsand Computational Biology (CIBCB), Eindhoven, Netherlands: IEEE,Aug. 2023, pp. 1–8, doi: 10.1109/CIBCB56990.2023.10264880. [38] I. A. Fouad and F. El-Zahraa M. Labib, “Identification of Alzheimer’sdisease from central lobe EEG signals utilizing machine learning andresidual neural network,” Biomedical Signal Processing and Control,vol. 86, p. 105266, Sep. 2023, doi: 10.1016/j.bspc.2023.105266. [39] N. Y. Murad, M. H. Hasan, M. H. Azam, N. Yousuf, and J. S. Yalli,“Unraveling the Black Box: A Review of Explainable Deep LearningHealthcare Techniques,” IEEE Access, vol. 12, pp. 66556–66568, 2024,doi: 10.1109/ACCESS.2024.3398203. [40] T.Mahmud,K.Barua,S.U.Habiba,N.Sharmen,M.S.Hossain,and [41] K.Andersson,“AnExplainableAIParadigmforAlzheimer’sDiagnosisUsing Deep Transfer Learning,” Diagnostics, vol. 14, no. 3, Art. no. 3,Jan. 2024, doi: 10.3390/diagnostics14030345. [42] D.A.Arafa,H.E.-D.Moustafa,H.A.Ali,A.M.T.Ali-Eldin,andS.F.Saraya, “A deep learning framework for early diagnosis of Alzheimer’sdiseaseonMRIimages,”MultimedToolsAppl,vol.83,no.2,pp.3767–3799, Jan. 2024, doi: 10.1007/s11042-023-15738-7. [43] A. S. Alatrany, W. Khan, A. Hussain, H. Kolivand, and D. Al-Jumeily,“An explainable machine learning approach for Alzheimer’s diseaseclassification,” Sci Rep, vol. 14, no. 1, p. 2637, Feb. 2024, doi:10.1038/s41598-024-51985-w. [44] M. S. K. Inan et al., “A slice selection guided deep integrated pipelinefor Alzheimer’s prediction from Structural Brain MRI,” BiomedicalSignal Processing and Control, vol. 89, p. 105773, Mar. 2024, doi:10.1016/j.bspc.2023.105773. [45] Q.K.Nguyen,T.T.H.Nguyen,V.T.K.Nguyen,V.B.Truong, [46] T. Phan, and H. Cao, “Efficient and Concise Explanations for ObjectDetectionwithGaussian-ClassActivationMappingExplainer,”Apr.20,2024, arXiv:2404.13417. doi: 10.48550/arXiv.2404.13417. [47] W.Zhuetal.,“PredictingRiskofAlzheimer’sDiseasesandRelatedDementiaswithAIFoundationModelonElectronicHealth Records,” medRxiv, p. 2024.04.26.24306180, Apr. 2024, doi:10.1101/2024.04.26.24306180. [48] V. Dhore, A. Bhat, V. Nerlekar, K. Chavhan, and A. Umare, “Enhanc-ingExplainableAI:AHybridApproachCombiningGradCAMand LRP for CNN Interpretability,” May 20, 2024, arXiv:2405.12175. doi:10.48550/arXiv.2405.12175. [49] J.Xu,C.Yuan,X.Ma,H.Shang,X.Shi,andX.Zhu,“Inter-pretablemedicaldeepframeworkbylogits-constraintattentionguid-inggraph-basedmulti-scalefusionforAlzheimer’sdiseaseanaly-sis,” Pattern Recognition, vol. 152, p. 110450, Aug. 2024, doi:10.1016/j.patcog.2024.110450. [50] G. Lozupone, A. Bria, F. Fontanella, F. J. A. Meijer, and C. D. Stefano,“AXIAL: Attention-based eXplainability for Interpretable Alzheimer’sLocalized Diagnosis using 2D CNNs on 3D MRI brain scans,” Oct. 01,2024, arXiv:2407.02418. doi: 10.48550/arXiv.2407.02418. [51] A. R. Monteiro, D. J. Barbosa, F. Remia˜o, and R. Silva, “Alzheimer’sdisease: Insights and new prospects in disease pathophysiology,biomarkers and disease-modifying drugs,” Biochemical Pharmacology,vol. 211, p. 115522, 2023, doi: 10.1016/j.bcp.2023.115522. [52] M. V. F. Silva, C. d. M. G. Loures, L. C. V. Alves, et al., “Alzheimer’sdisease: risk factors and potentially protective measures,” Journal ofBiomedical Science, vol. 26, no. 33, 2019, doi: 10.1186/s12929-019-0524-y. [53] Z. Sadeghi, R. Alizadehsani, M. A. CIFCI, S. Kausar, R. Rehman, P.Mahanta, P. K. Bora, A. Almasri, R. S. Alkhawaldeh, S. Hussain, B.Alatas, A. Shoeibi, H. Moosaei, M. Hlad´?k, S. Nahavandi, and P. M.Pardalos,“AreviewofExplainableArtificialIntelligenceinhealthcare,”Computers and Electrical Engineering, vol. 118, Part A, 2024, Art. no.109370, doi: 10.1016/j.compeleceng.2024.109370. [54] H. W. Loh, C. P. Ooi, S. Seoni, P. D. Barua, F. Molinari, and U. R.Acharya, “Application of explainable artificial intelligence for health-care: A systematic review of the last decade (2011–2022),” ComputerMethodsandProgramsinBiomedicine,vol.226,2022,Art.no.107161,doi: 10.1016/j.cmpb.2022.107161. [55] A.B.Arrieta,N.D´?azRodr´?guez,J.DelSer,A.Bennetot,S.Tabik,A.Barbado,S.Garc´?a,S.Gil-Lo´pez,D.Molina,R.Benjamins,R.Chatila,and F. Herrera, “Explainable Artificial Intelligence (XAI): Concepts,Taxonomies, Opportunities and Challenges toward Responsible AI,”CoRR, vol. abs/1910.10045, 2019. [Online]. Available: http://arxiv.org/abs/1910.10045. [56] A. Shahroudnejad, “A Survey on Understanding, Visualizations, andExplanation of Deep Neural Networks,” CoRR, vol. abs/2102.01792,2021. [Online]. Available: https://arxiv.org/abs/2102.01792. [57] M.Sundararajan,A.Taly,andQ.Yan,“AxiomaticAttributionforDeepNetworks,”CoRR,vol.abs/1703.01365,2017.[Online].Available:http://arxiv.org/abs/1703.01365. [58] M. R. Karim, T. Islam, O. Beyan, C. Lange, M. Cochez, D. Rebholz-Schuhmann, and S. Decker, “Explainable AI for Bioinformatics: Meth-ods, Tools, and Applications,” arXiv preprint, arXiv:2212.13261, 2023.[Online]. Available: https://arxiv.org/abs/2212.13261.

Copyright

Copyright © 2026 Shrutha Prakash. This is an open access article distributed under the Creative Commons Attribution License, which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited.