A novel diagnosis method based on multimodal-aware bi-objective competitive mechanism for Alzheimer's disease.

Alzheimer's disease (AD) is one of the most prevalent neurodegenerative disorders. Recent statistical surveys and studies indicate that AD is poised to become a major global health burden in the coming decades. Among current strategies, leveraging multimodal imaging data for early diagnosis and classification of AD has emerged as a promising approach. However, most existing methods primarily focus only on improving classification accuracy, while overlooking the importance of extracted imaging features and the underlying relationships among them. In this study, we propose a novel Multimodal-Aware Bi-Objective Competitive Mechanism Estimation of Distribution Algorithm (MABOC-EDA). The proposed MABOC-EDA incorporates both the individual significance of imaging features and their inter-modality interactions during the evolutionary process, two data structures were employed to represent them, which were updated according to a predefined update strategy. These structures were then used to construct a probabilistic model, from which the optimal feature subset was derived. The proposed method achieved classification accuracies and area under the ROC curve of 95.74%/97.25%,87.23%/89.93%, and 86.54%/88.76% for the AD vs. CN, AD vs. MCI, and CN vs. MCI classification tasks, respectively. Moreover, the brain regions that contribute most significantly to classification and those exhibiting strong correlations in the context of AD diagnosis. The correlated brain regions identified by proposed method provide insights that may facilitate the discovery of complementary biomarkers across different imaging modalities.