Attention-Enhanced Dual-Path CNN for Early Alzheimer’s Detection from Multi-Planar T1-Weighted MRI

Alzheimer’s disease (AD) is a progressive neurodegenerative disorder, and early detection is essential for effective clinical intervention. Previous deep learning approaches have achieved promising results using MRI data; however, most rely on full 3D volumes or standard 2D axial and coronal slices, often overlooking the diagnostic value of parasagittal views. In addition, many existing models lack attention mechanisms and multi-branch architectures, limiting their ability to capture both localized and contextual features. To overcome these limitations, we propose an attention-guided dual-path CNN that integrates sagittal, coronal, and parasagittal slices extracted at a 6.257° off-midline angle. The architecture combines a focused SNeurodCNN branch with an Inception-v4 path enhanced by CBAM for multi-scale feature extraction. Using T1-weighted MRI scans from the ADNI database, the proposed model achieved an accuracy of 98.9% and an AUC of 0.992, highlighting its potential for accurate and early AD classification.