A Deep Learning-Powered Framework for the Early Prediction of Alzheimer's Disease through Advanced Magnetic Resonance Imaging and Analysis.
Authors
Affiliations (3)
Affiliations (3)
- Department of Electronics and Communication Engineering, Cambridge Institute of Technology, Bengaluru, Karnataka, India. Electronic address: [email protected].
- Department of Electronics and Communication Engineering, Cambridge Institute of Technology, Bengaluru, Karnataka, India.
- Department of Electronics and Communication Engineering, Sreenidhi Institute of Science and Technology, Hyderabad, Telangana, India.
Abstract
Alzheimer's disease (AD) is a progressive neurodegenerative disorder that severely impairs cognitive and memory functions, highlighting the importance of early and accurate diagnoses. Although deep learning (DL)-based automated diagnostic systems have demonstrated promising results, many existing methods remain limited by inadequate feature representation, feature redundancy, and insufficient modeling of local and global brain patterns in magnetic resonance imaging (MRI) data. This study proposes an advanced DL-powered framework for the early prediction of AD using MRI data, aiming to improve diagnostic accuracy and interpretability. MRI images from the Augmented Alzheimer MRI Dataset, comprising non-demented, very mild demented, mild demented, and moderate demented classes, were used. Data pre-processing included resizing, normalization, noise reduction, and data augmentation to enhance image quality. High-level features are extracted using a pre-trained ResNet-50 network. Optimal feature selection is achieved using the Levy-Echo Search Optimizer (LESO), a hybrid of Cuckoo Search and Bat Algorithm. Classification was performed using the novel hybrid architecture DenEffNet, which integrates DenseNet-121 and EfficientNet-B3 to capture complementary local and multi-scale features. The eXplainable Artificial Intelligence (XAI) techniques, including Grad-CAM, LIME, and SHAP, were incorporated to enhance model transparency. The proposed framework achieved an accuracy of 97.8%, a precision of 98.0%, a recall of 98.6%, a specificity of 98.5%, an F1 score of 98.3%, and a Matthews correlation coefficient of 0.97, outperforming existing methods. The proposed framework demonstrates high accuracy, robustness, and interpretability, supporting its potential for reliable early AD diagnosis in clinical settings.