A Multi-Label Deep Learning Model for Detailed Classification of Alzheimer's Disease

Abstract

Background: Accurate diagnosis and classification of Alzheimer's disease (AD) are crucial for effective treatment and management. Traditional diagnostic models, largely based on binary classification systems, fail to adequately capture the complexities and variations across different stages and subtypes of AD, limiting their clinical utility. 

Methods: We developed a deep learning model integrating a dot-product attention mechanism and an innovative labeling system to enhance the diagnosis and classification of AD subtypes and severity levels. This model processed various clinical and demographic data, emphasizing the most relevant features for AD diagnosis. The approach emphasized precision in identifying disease subtypes and predicting their severity through advanced computational techniques that mimic expert clinical decision-making. 

Results: Comparative tests against a baseline fully connected neural network demonstrated that our proposed model significantly improved diagnostic accuracy. Our model achieved an accuracy of 83.1% for identifying AD subtypes, compared to 72.9% by the baseline. In severity prediction, our model reached an accuracy of 83.3%, outperforming the baseline (73.5%). 

Conclusions: The incorporation of a dot-product attention mechanism and a tailored labeling system in our model significantly enhances the accuracy of diagnosing and classifying AD. This improvement highlights the potential of the model to support personalized treatment strategies and advance precision medicine in neurodegenerative diseases.