Magnetic resonance imaging-based Alzheimer's disease detection using an EfficientNet-CMSACCN framework.
Authors
Affiliations (4)
Affiliations (4)
- Department of Computer Science and Engineering, M.Kumarasamy College of Engineering, Karur, Tamil Nadu 639113 India.
- Department of Computer Science and Engineering, Jaishriram Engineering College, Avinashipalayam, Tamil Nadu 638660 India.
- Department of Computational Intelligence, Saveetha School of Engineering, SIMATS University, Chennai, Tamil Nadu 602105 India.
- Department of Computer Science and Engineering, Saveetha School of Engineering, SIMATS University, Chennai, Tamil Nadu 602105 India.
Abstract
Alzheimer's Disease (AD) is a degenerative neurological condition characterized by memory loss, cognitive deterioration, and brain tissue shrinkage. Detecting it at an early stage is difficult due to variations in disease progression and the restricted scope of single-modality neuroimaging methods. Magnetic Resonance Imaging (MRI)-based Alzheimer's diagnosis, such as magnetic resonance imaging, offers complementary structural and functional insights, but existing deep learning methods often struggle with data imbalance, high computational complexity, and limited generalization. To fill these research gaps, design an MRI-based EfficientNet feature extraction framework for Alzheimer's stage classification. EfficientNet, equipped with compound scaling, depthwise-separable layers, and squeeze-and-excitation components, enables precise characterization of cortical structures and whole-brain variations while preserving computational efficiency. Extracted features are classified using a Compression-based Multi-Scale Attention Convolutional Network (C-MSACCN), which integrates attention mechanisms and compression strategies to enhance accuracy and reduce model complexity. Furthermore, the Improved Cellular Neighbours Optimiser (ICNO) fine-tunes hyperparameters, striking a balance between exploration and exploitation for optimal convergence and robustness. With 99.9% accuracy, precision, recall, and F1-score on datasets, the model outperforms prior work. Validation confirms consistency, and visualisation methods highlight disease-relevant regions to provide clinical insight.