Identifying Biomarker of Alzheimer's Disease Based on an Extensible Ensemble Learning Model

Using neural network to study the potential relationship between brain genetics and imaging has become an effective method to understand the neurodegenerative disease. At present, for neurodegenerative diseases such as Alzheimer’s disease, researchers are increasingly focusing on potential relationship between different modal data rather than just focusing on a single modal data. However, multi-modal data often has the characteristics of few samples and high dimensions. To solve this problem, the existing genetic evolution random neural network cluster (GERNNC) model uses Pearson correlation analysis to construct the fusion characteristics of brain regions and genes, and combines genetic evolution algorithm and neural network to classify AD patients and extract pathogenic factors. However, due to the limitation of early fusion methods to fuse multi-modal data in this model, it is difficult for the model to freely increase the input multi-modal data. In addition, because the model integrates a large number of random neural network, the stability of the model is inadequate. In order to solve the above two problems, this article improves the multi-modal fusion method, model framework and algorithm in the GERNNC model, and proposes an improved genetic evolution random neural network cluster (IGERNNC) model. The IGERNNC model uses a late fusion method, using multiple parallel GERNNCs to train multimodal data simultaneously, and fusing the training results to allow the model to freely add data from other modalities. In addition, in order to further improve the connection between different modalities, this article proposes a feedback mechanism and a ‘forced selection’ method for the IGERNNC model to enhance the stability. Through several independent experiments, IGERNNC model can more effectively identify AD patients and extract relevant pathogenic factors, which provides a theoretical basis for further understanding complex pathology of neurodegenerative diseases.