Optimized hybrid deep learning architecture for robust Alzheimer’s disease diagnosis using SMOTE-based data augmentation

2026-06-09

PRE

Chakraborty Sudeepta Timir

Achyut Shankar *

Sanchali Das

Wattana Viriyasitavat

DOI：10.1016/j.imavis.2026.106060

Original

Original request for help

Save

Abstract

En 中文

• Proposes a hybrid deep learning framework combining Vision Transformer (ViT) and Swin Transformer for robust Alzheimer’s disease diagnosis. • Integrates SMOTE-based data augmentation to effectively address class imbalance in MRI datasets. • Employs Bayesian Optimization for automated hyperparameter tuning, improving model performance and generalization. • Validated on multiple heterogeneous datasets (ADNI, OASIS, Kaggle) to ensure cross-dataset robustness. • Achieves superior performance with 97.6% accuracy, 96.1% sensitivity, and 97.0% F1-score, outperforming baseline models by 6–9%. • Demonstrates strong potential for early-stage Alzheimer’s detection and clinical decision support systems.

Optimized hybrid deep learning architecture for robust Alzheimer’s disease diagnosis using SMOTE-based data augmentation

Abstract

Journal

Organization

Cited Papers

Citing Papers