A Novel Classification Scheme for Early Alzheimer’s Disease (AD) Severity Diagnosis Using Deep Features of a Hybrid Cascade Attention Architecture

Alzheimer’s Disease (AD) is a progressive neuro degenerative disorder requiring early and accurate diagnosis for effective management. This study introduces a novel approach for AD classification by integrating a CNN with a cascade attention model (CAM-CNN). The CAM-CNN model enhances classification accuracy and reduces processing complexity compared to traditional CNNs by incorporating advanced attention mechanisms and novel cost functions. The dataset comprises 33,984 MRI scans categorized into four cognitive impairment levels: Non-Demented, Very Mild Demented, Mild Demented, and Moderate Demented, with each image resized to 128×128 pixels. Data augmentation techniques, including shifts, shears, random rotation, zooming, rescaling, and flipping, were applied to increase the training set’s diversity. The model employs constraint cost functions, cross-network pre-trained parameters, and an entropy addition method to improve feature extraction and classification performance. Experimental results demonstrate the proposed model’s superiority, achieving the highest sensitivity, specificity, and accuracy scores compared to VGG16, VGG19, ResNet, and CNN+Attention architectures. The CAM-CNN model’s advanced architecture and robust training methodology provide a highly effective tool for early diagnosis and management of Alzheimer’s Disease, significantly enhancing classification performance.