Nodes-weighted-graph approach for rsfMRI data classification: Application to schizophrenia

摘要

We propose a new method to analyze rsfMRI data to identify people with Schizophrenia. The method can be extended to identify other mental diseases. Although many graph-based methods have been applied to achieve similar goals, most of them only focus on using graph with a single feature. Our method not merely uses existing network properties, but also takes advantage of the content of nodes to improve the classification process. Next, we apply selected features a machine learning technique, namely multi-kernel learning approach to perform the classification. We tested our method on rsfMRI dataset about Schizophrenia and Health Control. The results showed that our method does improve classification performances (classification accuracy of 93.1%. 7.39% outperform existing approach), and helps discovering disease-related brain regions.