Alzheimers-Data-Fusion

Alzheimer's disease (AD) is a debilitating neurodegenerative disorder affecting a large and growing number of individuals worldwide. With the projected increase in AD cases, there is a pressing need for accurate and early diagnosis to maximize treatment effectiveness. This study aims to achieve this goal by developing an optimal data fusion strategy. We believe that simple concatenation of various data modalities is not an efficient approach since the modalities have different signal-to-noise ratios and vary in their contribution to the overall predictive power. We hypothesize that a better strategy would involve reducing the dimensionality of each data modality first (hence, removing the noise as much as possible) before combining them within a classification model. To reduce dimensionality, we explore both unsupervised and supervised techniques, and evaluate the resulting alternative methods based on data from the National Alzheimer's Coordinating Center. Our findings demonstrate that effective dimensionality reduction for each modality before data integration can significantly enhance the accuracy of AD prediction. Furthermore, our study reveals that our suggested supervised encoder, which involves slight modifications to standard deep neural networks, outperforms other dimensionality reduction strategies, thereby offering additional advancements in AD prediction.