Research on Alzheimer's Disease MRI Image Classification Based on Spatial Attention Mechanism

Shanshan ZhaoHaolin ShiYanli Wan^*Changhao GuanQingjia ZengChongyang Zhang

• Chinese Academy of Medical Sciences & Peking Union Medical College Institute of Medical Information, Chaoyang, China

Early diagnosis of Alzheimer's Disease (AD) is crucial for improving patient quality of life and treatment outcomes. To enhance the accuracy of MRI image classification for Alzheimer's Disease, This study proposes a customized bidirectional spatial attention mechanism to enhance the model's focus on key lesion regions in AD MRI scans. Unlike conventional spatial attention mechanisms, the proposed ATT module generates spatial attention maps by integrating adaptively pooled features along both vertical and horizontal orientations, thereby enabling more refined adjustment of attention weights across different image regions. Additionally, to address issues of insufficient samples and class imbalance, we incorporated data augmentation and expansion strategies, providing a richer and more diverse set of samples to further improve the model's overall performance. Experimental results demonstrate that, on the augmented OASIS1 dataset, the improved Swin-Tiny+ATT model significantly boosts classification performance. Compared to the baseline Swin Transformer model, the accuracy of Swin-Tiny+ATT increased from 84.83% to 87.96%, recall from 89.82% to 91.92%, precision from 85.27% to 91.98%, and F1 score from 87.26% to 91.89%. These results indicate that the ATT module effectively enhances the model ' s ability to capture complex spatial features. The proposed deep learning approach based on the improved Swin Transformer demonstrates exceptional potential for MRI image classification in Alzheimer's Disease, providing strong support for early AD diagnosis.