Louvain clustering integration within density-based graph classification (Louvain dbGC) in Schizophrenia

Abstract Several brain disorders are characterized by their silent manifestations that do not display clinical symptoms and are usually diagnosed at advanced stages in which the brain disease may be irreversible. Common strategies to diagnose some brain disorders depend on self-reported symptoms and observed behavior during an extended period of time, and there are no quantitative tests to diagnose mental disorders. Mental disorders are the leading cause of disability in the US and are typically characterized by behavioral changes without clear signs of the structural changes often seen in brain diseases, such as those caused by tumors. With new diagnosis methods, more people are being diagnosed with mental disorders, and some research suggests the importance of early detection to improve patients’ prognoses in restoring the functionality of the brain. Therefore, the goal of this study is to identify biomarkers and underlying biological substrates that will lead to early diagnosis and improved treatment for schizophrenic patients. We combined clustering techniques and density-based graph classification to better predict abnormal functional networks in schizophrenics. The Louvain dbGC combines local and global graph measures with the mesoscale organization of brain networks. To evaluate the effectiveness of the Louvain dbGC, multiple feature selection and classification algorithms were applied. Comparison with state-of-the-art methods of (1) Seed-based Analysis, (2) Independent Component Analysis, and (3) WUD Graph analysis is conducted. The Louvain dbGC better classified and separated Schizophrenics from Healthy Controls with 99.3% accuracy, 98.80% sensitivity, and 100% specificity. The Louvain dbGC can be extended to other mental disorders to detect and monitor therapeutic interventions of such diseases.