Integrating enhanced Crested Porcupine Optimizer with contrastive learning for multimodal Alzheimer’s disease diagnosis

By combining brain scanning and genetic data, imaging genetics provides multidimensional insights essential for decoding Alzheimer’s disease (AD) and improving its detection. Its data, however, are highly heterogeneous and difficult to model relationships. To overcome these limitations, we introduce an enhanced Crested Porcupine Optimizer (CPO) integrated within a contrastive learning framework to improve the identification of risk features. In the optimization phase, the Sobel sequence and an echelon-based defense mechanism in dynamic population stratification were applied to improve the search mechanism. These changes can reduce premature convergence, allowing the algorithm to more fully traverse the complex, high-dimensional search space, thereby improving overall optimization efficiency. Next, contrastive learning was used to jointly model image and genetic features. By constructing positive and negative sample pairs, it can learn the potential association structure. Comprehensive evaluations show that our method gets superior performance over other methods.