VisionMamba-Based Multimodal Learning for Alzheimer's Disease and Mild Cognitive Impairment Diagnosis

Early diagnosis of Alzheimer's Disease (AD) and Mild Cognitive Impairment (MCI) is critical for delaying disease progression and improving patient outcomes. This study proposes a multimodal deep learning approach based on the VisionMamba framework, leveraging 2D MRI slice images as the primary imaging data source. By integrating cognitive scores (e.g., MMSE, FAQ) and demographic information (e.g., gender, age), the method achieves efficient and accurate classification of AD and MCI. In this approach, the VisionMamba structure is utilized to extract features from individual 2D MRI slices. These features are then combined with cognitive scores and demographic attributes through a multimodal fusion technique. Specifically, a late fusion strategy is adopted to integrate imaging and non-imaging features at the classification stage, enhancing the overall model performance. Experimental results demonstrate that the VisionMamba-based 2D imaging classification method outperforms traditional 3D image processing approaches in terms of accuracy, recall, and F1 score, while significantly reducing computational costs. This study underscores the potential of VisionMamba in leveraging 2D MRI slices and multimodal fusion techniques for efficient and scalable AD and MCI diagnosis. The findings highlight its capability to provide an accurate and computationally efficient solution for early detection.