Machine learning technique has long been utilized to assist disease diagnosis, increasing clinical physicians’ confidence in their decision and expediting the process of diagnosis. In this case, machine learning technique serves as a tool for distinguishing patients from healthy people. Additionally, it can also serve as an exploratory method to reveal intrinsic characteristics of a disease based on discriminative features, which was demonstrated in this study. Resting-state functional magnetic resonance imaging (fMRI) data were obtained from 148 participants (including patients with schizophrenia and healthy controls). Connective strengths were estimated by Pearson correlation for each pair of brain regions partitioned according to automated anatomical labelling atlas. Subsequently, consensus connections with high discriminative power were extracted under the circumstance of the best classification accuracy. Investigating these consensus connections, we found that schizophrenia group predominately exhibited weaker strengths of inter-regional connectivity compared to healthy group. Aberrant connectivities in both intra- and inter-hemispherical connections were observed. Within intra-hemispherical connections, the number of aberrant connections in the right hemisphere was more than that of the left hemisphere. In the exploration of large regions, we revealed that the serious dysconnectivities mainly appeared on temporal and occipital regions for the within-large-region connections; while connectivity disruption was observed on the connections from temporal region to occipital, insula and limbic regions for the between-large-region connections. The findings of this study corroborate previous conclusion of dysconnectivity in schizophrenia and further shed light on distribution patterns of dysconnectivity, which deepens the understanding of pathological mechanism of schizophrenia.
- Functional connectivity
- Hemispherical distribution of connections
- Large-region connectivity
- Resting-state fMRI