Deep Learning Framework for Alzheimer’s Detection: Architecture Design and Comparative Analysis

Alzheimer's Disease (AD) is a progressive neurodegenerative disorder that significantly impacts cognitive function and quality of life. Deep learning has emerged as a promising approach for early AD diagnosis using neuroimaging modalities such as MRI. This paper presents a comprehensive review and comparative analysis of state-of-the-art deep learning models applied to MRI-based AD detection, including CNNs, hybrid architectures, and transformer-based networks. Key models are evaluated based on architecture design, dataset usage, accuracy, and ability to detect early-stage cognitive decline. In addition to literature synthesis, we propose a novel hybrid architecture that combines EfficientNet-B6 for multiscale spatial feature extraction with Long Short-Term Memory (LSTM) networks for capturing temporal dependencies. Although this paper does not include the empirical evaluation of the proposed model, it highlights its theoretical advantages and design motivation compared to existing approaches. The insights gained from this study serve to identify current limitations in the field, motivate architecture-level innovations, and guide future research towards clinically applicable Alzheimer's diagnostic tools.