Alzheimer’s Disease Detection Using Deep Learning on MRI: A Review

Abstract

Alzheimer’s disease (AD) affects millions globally, making early detection crucial for effective treatment. This review paper analyzes deep learning approaches for AD detection using MRI scans, focusing on patient-level data splitting, class imbalance handling, and model efficiency. The study reviews an end-to-end pipeline using the OASIS dataset, highlighting challenges such as severe data imbalance and data leakage in multi-slice patient datasets. A comparative analysis between a baseline CNN and EfficientNet-B0 is presented using techniques like focal loss and minority-focused data augmentation. Results show that accuracy alone is misleading in imbalanced medical datasets, and metrics like Macro-F1 provide a more reliable evaluation. The study also emphasizes explainability using Grad-CAM and demonstrates a lightweight deployment via a Streamlit interface. This review highlights the importance of proper validation strategies, reproducibility, and ethical considerations in medical AI systems, making it suitable for real-world screening and research applications.