The Identification research of bipolar disorder based on CNN

Abstract

Bipolar disorder (BD), as a type of mood disorder, refers to the disease that has both manic episodes and depressive episodes. At present, the clinical diagnosis and treatment of bipolar disorder is often unsatisfactory, the disease always has a high rate of misdiagnosis and recurrence. The specific pathogenesis of bipolar disorder is still unclear, but related studies have shown that bipolar disorder is highly heritable, so a large number of studies have been carried out around the genomics of the disease, including some research combined with machine learning. Motivated by these research, we use the Single Nucleotide Polymorphism (SNP) which through the Genome-wide association analysis (GWAS) obtained as molecular genetic markers, combined with Convolutional Neural Network (CNN) constructing a recognition model for bipolar disorder. The experimental results show that the overall recognition rate can up to 79%, compared to similar studies, our model has achieved excellent performance. In addition, unlike traditional GWAS for finding pathogenic sites, our study first attempted to use it as a feature selection method of data preprocessing in machine learning, the experimental results prove the feasibility of this method. We believe that our research will provide some new ideas for the precision diagnosis and treatment of bipolar disorder in the future.