A 2-Level Stacking Ensemble Framework for Multi-Class Alzheimer’s Disease Diagnosis Using CNN Feature Extractors and SVM-XGBoost Classifiers

Topic: Brain imaging, connectivity and assessment

Alzheimer's disease classification by means of Magnetic Resonance Imaging demands strong models that are able to generalise to different stages of dementia. This study suggests and tests a two-level stacking ensemble architecture for four-class classification “Mildly Demented, Moderately Demented, Non-Demented, and Very Mildly Demented”. This framework included three feature extractors: ResNet152, DenseNet121, and a custom convolutional network (Shashi-CNN). The extracted features were combined and standardised with Standard Scaler, then reduced using Principal Component Analysis by keeping 95% variance, and classified at Level-1 with linear Support Vector Machine with probability output. To prevent data leakage, an XGBoost meta-learning classifier was trained on out-of-fold predictions, and level-2 meta-learning was performed. The number of MRI images was 6400. The Full Ensemble has a mean “accuracy of 0.9997 ± 0.0004, precision of 0.9997, recall of 0.9997, F1-score of 0.9997, macro sensitivity of 0.9998, macro specificity of 0.9998, Cohen's kappa of 0.9995, Matthews Correlation Coefficient of 0.9995, and micro-average ROC AUC of 1.0000” across five folds. Synergistic benefit was shown in the single- and dual-model configurations in an ablation study. The improvement was found to be statistically significant using McNemar's test (χ²=98.0865, p < 0.0001) comparing Full Ensemble and Shashi Model. The focus was anatomically relevant as confirmed by Grad-CAM visualisations. The proposed ensemble achieved better results than the “state-of-the-art baseline results reported for the ADNI and OASIS datasets”.

Key findings

• The improvement was found to be statistically significant using McNemar's test (χ²=98.0865, p < 0.0001) comparing Full Ensemble and Shashi Model.
• The proposed ensemble achieved better results than the “state-of-the-art baseline results reported for the ADNI and OASIS datasets”.

Study details

• Evidence strengthLow

Read the original

• DOI: 10.21203/rs.3.rs-10843182/v1
• Original source →

Free copy unknown: no free-copy check has an answer for this record yet. That does not mean it is paywalled.

@article{mourya2026-e7a2d6b7,
  author  = {Mourya, S. K. and Yadav, A. K. S.},
  title   = {{A 2-Level Stacking Ensemble Framework for Multi-Class Alzheimer’s Disease Diagnosis Using CNN Feature Extractors and SVM-XGBoost Classifiers}},
  journal = {Research Square},
  year    = {2026},
  doi     = {10.21203/rs.3.rs-10843182/v1},
  url     = {https://doi.org/10.21203/rs.3.rs-10843182/v1}
}

Related studies in Long-term cohort studies

• Multi-modal latent space inducing ensemble SVM classifier for early dementia diagnosis with neuroimaging data Medical image analysis · 2019-12-28 · usefulness score 47/100
• A Deep Learning Approach for Automated Diagnosis and Multi-Class Classification of Alzheimer's Disease Stages Using Resting-State fMRI and Residual Neural Networks Journal of medical systems · 2019-12-18 · usefulness score 64/100
• Effective feature learning and fusion of multimodality data using stage-wise deep neural network for dementia diagnosis Human brain mapping · 2018-11-01 · usefulness score 50/100
• Biomarker fidelity score: A quantitative framework for individual-level validation of explainability methods in 3D Alzheimer’s disease MRI classification bioRxiv · 2026-08-20 · usefulness score 16/100
• Meta-analysis based SVM classification enables accurate detection of Alzheimer's disease across different clinical centers using FDG-PET and MRI Psychiatry research · 2012-11-10 · usefulness score 65/100
• 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 Therapeutic advances in neurological disorders · 2019-03-29 · usefulness score 52/100

Similar studies

• Early Detection of Alzheimer's Disease Using Magnetic Resonance Imaging: A Novel Approach Combining Convolutional Neural Networks and Ensemble Learning 2020 · Digital health tools
• Brain MRI analysis for Alzheimer's disease diagnosis using an ensemble system of deep convolutional neural networks 2018 · Digital health tools
• DEMNET: A Deep Learning Model for Early Diagnosis of Alzheimer Diseases and Dementia From MR Images 2021 · Long-term cohort studies
• Ensembles of Deep Learning Architectures for the Early Diagnosis of the Alzheimer's Disease 2016 · Digital health tools
• Automated classification of Alzheimer's disease and mild cognitive impairment using a single MRI and deep neural networks 2019 · Digital health tools
• Multimodal neuroimaging feature learning for multiclass diagnosis of Alzheimer's disease 2015 · Digital health tools

More on Long-term cohort studies

• A review that covers this ground ¹⁸F-FDG PET for the early diagnosis of Alzheimer's disease dementia and other dementias in people with mild cognitive impairment (MCI) The Cochrane database of systematic reviews · 2015-01-28 · usefulness score 69/100
• Newest in this subject Ultra-sensitive Mapping of Cell-free DNA-derived Methylomes Reveals Brain-specific Epigenetic Alterations Implicated in Alzheimer’s Disease medRxiv · 2026-10-26 · Preprint — not yet peer-reviewed · usefulness score 16/100
• Newest in this subject Interpretable Phenotype Discovery Using Temporal Risk-Aware Clustering medRxiv · 2026-10-20 · Preprint — not yet peer-reviewed · usefulness score 16/100

See all studies in Long-term cohort studies →