Accurate and early diagnosis of Alzheimer’s disease (AD), particularly in its mild cognitive impairment (MCI) stage, remains a major challenge in clinical neurology because of the complexity of neuroimaging data and limitations in current diagnostic methods. This paper presents a novel hybrid framework that significantly advances the automated classification of AD, MCI, and cognitively normal (CN) individuals by integrating structural magnetic resonance imaging features with key clinical data. Leveraging the FreeSurfer tool for robust preprocessing, our approach combines texture-based descriptors—such as Gray Level Co-occurrence Matrix, scale-invariant feature transform, local binary pattern, and histogram of oriented gradient—with patient-reported clinical indicators to construct a comprehensive hybrid feature vector. The classification process is optimized through a transfer learning model based on Visual Geometry Group 16-layer network (VGG16) architecture, further enhanced by the butterfly optimization algorithm to achieve superior model convergence and classification accuracy. The significance of this research lies in its unique integration of imaging biomarkers and clinical assessments, addressing the shortcomings of models that rely exclusively on neuroimaging. Extensive validation on the Alzheimer’s Disease Neuroimaging Initiative data set demonstrates that the proposed method achieves an impressive 98.4 % accuracy for binary AD versus CN classification and 89.8 % for multiclass (AD, MCI, CN) scenarios—substantially outperforming state-of-the-art conventional convolutional neural networks and ensemble models. The impact of this work is underscored by its potential to improve clinical decision-making, support early intervention strategies, and facilitate more effective management of AD in real-world healthcare settings.