The distinguishing intrinsic brain circuitry in treatment-naive first-episode schizophrenia: Ensemble learning classification

Schizophrenia is frequently characterized as a prototypical disorder of integration of brain function involving almost all intrinsic connectivity networks. However, a consistent conclusion regarding the most distinguishing brain circuitry in schizophrenia has not yet been reached. In this study, we used a novel network-based ensemble method to explore the most distinguishing brain circuitry in treatment-naive first-episode schizophrenia (n = 41) and healthy controls (n = 38) who underwent the task-free functional MRI scanning. Ensemble method showed commendable discrimination ability (84.7% for classification accuracy, 91.9% for sensitivity, 74.5% for specificity, all p < 0.05 for permuted test). The most distinguishing connections were located in the right paralimbic system and bilateral default mode network. Notably, distinguishing aberrations were significantly correlated with symptom severity (negative score: R-2 =0.58, P < 0.05, Bonferroni corrected; positive score: R-2 =0.74, P < 0.05, Bonferroni corrected) in schizophrenia patients. These most distinguishing aberrations present good potential for the underlying symptoms, and provide great insight into the mechanism of schizophrenia. Our results suggested that the ensemble method was a powerful tool to help with clinical diagnosis of schizophrenia and to explore the mechanism of schizophrenia. (C) 2019 Elsevier B.V. All rights reserved.