Effective deep convolutional neural network with attention mechanism for Alzheimer disease classification

The reports from the Health Organizations indicates a sudden growth in neurocognitive disorders among middle-aged and elderly individuals. The accurate detection of Alzheimer's disease (AD) is essential for improving patient care, specifically during the early stages, where timely risk identification enables individuals to adopt preventive measures before irreversible brain damage occurs. Though, several studies have discovered about computerized approaches for AD, many existing techniques remain limited by inherent methodological constraints and insufficient clinical scrutiny. The current systems struggle to reliably predict the disorder in its initial stages. To reduce the need for frequent clinical visit and lower diagnostic costs, the machine learning and deep learning have emerged as powerful tools for AD detection.

This work reviews several research relevant on studies on AD and highlights how these computational techniques can support researchers in achieving more efficient and accurate early-stage detection. The Deep Convolutional Neural Network (Deep-CNN) with Attention mechanism is proposed to augment the spatial attention module and multi-class classification of Alzheimer disease stages. The model has trained and evaluated on the OASIS dataset using subject-level which satisfy statistical-validation and standard preprocessing.

The proposed Deep-CNN and attention model focuses the model capacity on diagnostically relevant regions. The proposed model achieved an accuracy of 97%, which is higher than existing methods like SVM with kernels (90.5%), SVM Gaussian radial basis kernel (85%), and traditional CNN (93.5%).

The visualizations of attention mechanism are used to increase the interpretability and demonstrate the attention maps which are align with known AD biomarkers. These results indicates that the attention-guided deep models can both improve multi-class MRI classification accuracy and provide clinically useful regional explanations.