Early detection of Alzheimer’s disease (AD) is crucial for prompt clinical intervention and enhanced patient outcomes. Here, we introduce a hybrid deep learning design that leverages EfficientNet-B6 spatial feature extraction capabilities and integrates the Long Short-Term Memory (LSTM) neural network to capture the sequential patterns in MRI scans for Alzheimer’s classification. The model is developed to categorize brain MRI scans into four diagnostic stages: non-demented, very mild demented, mild demented, and moderate demented, utilizing the publicly accessible augmented Alzheimer MRI datasets. A preprocessing step is applied, comprising scaling, normalization, and data augmentation, to ensure standardized and high-quality inputs to the network. The experimental findings for accuracy are presented, while we outline the conceptual basis of the hybrid model and its potential to address shortcomings of current methods, like inadequate spatial-temporal integration, scalability, and interpretability. The proposed architecture aims to enhance diagnostic results by leveraging EfficientNet-B6 multiscale spatial feature learning and LSTM temporal modeling. The proposed hybrid architecture evaluated on 80:20 training and validation dataset split, achieving a classification accuracy of 98.90% on a benchmark dataset, which shows the potential of the proposed hybrid architecture. This study lays the foundation for a comprehensive, interpretable, and scalable deep learning model for early AD detection from MRI scans.

Alzheimer’s Disease Early Detection Disease Classification Diagnosis EfficientNet LSTM Deep Learning