Machine Learning-based Multi-classification for First-Episode Schizophrenics, Ultra-high risk Schizophrenics, and Healthy Controls

Abstract

Schizophrenia is a severe chronic disabling disease. Prompt treatment of ultra-high-risk individuals in the prodromal phase is of great significance for preventing the development of schizophrenia. The purpose of this study is to find a way to effectively distinguish ultra-high-risk individuals with schizophrenia, and to analyze important biomarkers of schizophrenia. There are 101 first-episode drug-naive schizophrenia patients, 49 ultra-high-risk individuals and 94 healthy people participated in our study. The cognition data, cortical thickness and the local gyrification index of these participants were collected for the identification of schizophrenia using various machine learning methods. Meanwhile, biological markers that indicate mental illness are identified by analyzing their relationship among different categories of individuals. Support vector machine performed best among the machine learning methods, with a classification accuracy of 86.4%. And the results indicate that the critical features for the identification of the three-type subject are executive function, the right cingulate gyrus, and the left temporal pole.