Alzheimer’s Classification Using Hybrid Deep Learning Models

Abstract Alzheimer’s disease (AD) is a progressive neurodegenerative disorder that re mains challenging to detect accurately at its early stages. Structural magnetic resonance imaging (MRI) is a valuable modality to identify early signs of AD;however, current diagnostic approaches often struggle to distinguish sub tle differences between subjects with AD, mild cognitive impairment (MCI) and normal control (NC). In particular, classification performance declines when dealing with fine structural variations in brain anatomy. This study ad dresses these limitations by developing six tailored convolutional neural net work (CNN) architectures for classifying two-dimensional axial MRI slices. The proposed CNN baseline model achieved an accuracy that exceeded 97%, while the integration of spatial and channel attention mechanisms further en hanced the accuracy to 99%. Furthermore, a novel multi-output convolu tional block attention module (CBAM) was designed to refine the classifica tion process, with notable improvements in MCI detection, a critical stage for timely intervention in AD progression. Experiments conducted using the Alzheimer’s Disease Neuroimaging Initiative (ADNI) data set demonstrate the superior performance of sequential channel and spatial attention strategies in improving diagnostic precision. These findings highlight the potential of attention-enhanced deep learning frameworks to advance reliable, clinically applicable tools for early AD detection.