Abstract
Alzheimer’s disease (AD) is a progressive neurodegenerative disorder that affects millions of individuals worldwide, leading to memory loss, cognitive decline, and functional impairments. Early and accurate detection of AD is critical for effective management and treatment planning. This paper presents an efficient approach for Alzheimer’s disease classification using a deep learning model based on the EfficientNetV2S architecture, leveraging transfer learning to enhance performance. EfficientNetV2S, an evolution of the EfficientNet model, is designed to balance speed and accuracy by combining fused-MBConv and MBConv layers, making it highly suitable for tasks requiring both high performance and computational efficiency. In this study, we fine-tune a model initialized with ImageNet-pretrained weights on a domain-specific Alzheimer’s dataset. Furthermore, we rigorously validate the model’s performance using k-fold cross-validation, confirming its reliability and generalizability across diverse data subsets. The proposed model achieved an accuracy of 98.1%, a precision of 98.9%, recall of 98.3%, and F1-score of 98.6%. These results demonstrate significant improvements in performance, outperforming other state-of-the-art models. Transfer learning allows the model to adapt pretrained features to the Alzheimer’s domain, speeding up training and improving generalization. Our findings highlight the potential of EfficientNetV2S for high-performance applications in medical image classification, where both computational efficiency and accuracy are crucial.
Keywords
Alzheimer’s disease, Machine learning, Deep learning, Image preprocessing, Magnetic resonance imaging (MRI)