Integrating segment anything model with hierarchical enhanced linear attention for Alzheimer's disease detection

Alzheimer's disease (AD), a leading cause of dementia, poses substantial challenges due to its complex diagnostics and the subtle onset of symptoms. The emergence of machine learning, particularly deep learning, offers a new frontier in the accurate and early diagnosis of AD through advanced image analysis. Our study introduces the Hierarchical Enhanced Linear Attention Classification Model for AD (HELA-CAM), which incorporates novel attention mechanisms within a transformer architecture for analyzing 3 dimension (3D) magnetic resonance scans. Utilizing a backbone encoder based on the Segmentation Anything by Text Prompts (SAT), HELA-CAM addresses the key limitations of traditional convolutional neural networks by enhancing both the interpretability and accuracy of diagnostic predictions. The proposed HELA-CAM demonstrates superior performance against existing state-of-the-art methods on public datasets, showcasing its potential as a robust tool in clinical settings.