An enhancing security in Alzheimer’s disease detection using attention-enhanced CNN with ResNet and channel-spatial attention blocks

In the quest for evolving Alzheimer's disease (AD) detection using medical imaging data, this paper has discovered the interaction between Convolutional Neural Networks (CNNs) and attention mechanisms. This study has presented an innovative framework that integrates Channel Attention Blocks (CAB) and Spatial Attention Blocks (SAB) into the ResNet-50 model to improve detection accuracy over feature representation in brain image datasets. The proposed CABs minimize data redundancy and emphasize key channel-wise features through adaptive recalibration and pooling. Meanwhile, the SABs enhance model balance by focusing on spatial regions of interest in the CAB-refined image. This work observed how well the ResNet deviations functioned and improved with the designed attention blocks. The results indicate that ResNet-50 outperforms other models. Further tests confirmed its selection over preliminary trials, evaluating different attention block configurations: parallel (P-ResNet-50), sequential channel-first (CS-ResNet-50), and spatial-first (SC-ResNet-50). The findings identified CS-ResNet-50 as the most effective configuration. In this research study, using the AD Neuroimaging Initiative (ADNI) dataset, CS-ResNet-50 attained an accuracy of 98.62%, which is improved over baseline ResNet-50 and other attention models like CBAM and SE blocks, with verified accuracies of 96.15% and 95.46%, respectively.