Interpretable Multi-Attention Fusion Mechanisms for Early Detection of Transitional Phases in Alzheimer's Disease.

Alzheimer's disease (AD), a chronic neurodegenerative disorder, is characterized by progressive cognitive and memory decline, ultimately leading to physical deterioration and death. Early detection is essential for slowing the progression of symptoms and improving patient outcomes. Recent advancements in technology have led to the increased application of machine learning for AD detection. In this study, we aim to identify individuals at risk for future AD onset. To this end, we propose a Multi-Attention Gated Multimodal Unit model that integrates Electronic Health Records (EHRs) and Magnetic Resonance Imaging (MRI) data. We also present an evaluation framework designed to measure the model's effectiveness in early-stage AD detection. Experimental results demonstrate that our approach outperforms baseline models in recognizing early signs of AD. Furthermore, to enhance interpretability, we employ Gradient-Weighted Class Activation Mapping (Grad-Cam) heatmaps and attention maps, offering insights into the model's decision-making process and its ability to detect crucial AD-related features at an early stage.