With structural magnetic resonance imaging (MRI) images, conventional methods for the classification of schizophrenia (SCZ) and healthy control (HC) extract cortical thickness independently at different regions of interest (ROIs) without considering the correlation between these regions. In this paper, we proposed an improved method for the classification of SCZ and HC based on individual hierarchical brain networks constructed from structural MRI images. Our method involves constructing individual hierarchical networks where each node and each edge in these networks represents a ROI and the correlation between a pair of ROIs, respectively. We demonstrate that edge features make significant improvement in performance of SCZ/HC classification, when compared with only node features. Classification performance is further investigated by combining edge features with node features via a multiple kernel learning framework. The experimental results show that our proposed method achieves an accuracy of 88.72% and an area under the receiver operating characteristic (ROC) curve (AUC) of 0.9521 for SCZ/HC classification, which demonstrate that our proposed method is efficient and promising for clinical applications for the diagnosis of SCZ via structural MRI images. Therefore, this paper provides an alternative method for extracting high-order cortical thickness features from structural MRI images for classification of neurodegenerative diseases such as SCZ.
Sign-in or become an IEEE member to discover the full contents of the paper.