Optimized Deep Feature Fusion For Alzheimer’s Disease Detection Using AGWO And Pre-Trained Cnns

Abstract

Timely intervention and therapy of Alzheimer's disease (AD) depend on an early and precise diagnosis. The sensitivity and specificity of traditional diagnostic methods are frequently compromised, particularly when used to various phases of cognitive decline. AD is a gradually advancing neurodegenerative condition that results in memory decline, cognitive difficulties, and shifts in behavior. Detecting AD at an early stage is crucial for slowing its progression and improving patients' quality of life. To address these challenges, deep learning and optimization-based methods offer significant potential in improving classification performance. To extract features, this study suggests a hybrid diagnostic framework that incorporates two pre-trained CNN models: VGG16 and InceptionV3, with Adaptive Grey Wolf Optimizer (AGWO) for optimal feature selection. The pipeline begins with data preprocessing techniques including augmentation, normalization, filtering, and scaling. Features extracted from both pre-trained models are fused and refined using AGWO to enhance discriminative power. The optimized features are then classified using an improved Multilayer Perceptron (MLP) to categorize subjects into Alzheimer's Disease stages.The proposed VGG16+InceptionV3+AGWO hybrid framework attained a high level of categorization accuracy of 98.62% on the test dataset, demonstrating its efficiency in identifying different stages of Alzheimer's Disease with remarkable precision.The integration of multiple pre-trained models with AGWO-based feature selection significantly enhances the performance of Alzheimer's Disease classification. The results validate the potential of the proposed framework as a robust tool for clinical decision support, encouraging further exploration and application in real-world diagnostic settings.