毕业生知识库

Schizophrenia is one of the most complex psychiatric disorder, although the underlying neuropathology is not clear. Researchers have proposed several hypothesis trying to explain the pathogenesis of schizophrenia, including neural development hypothesis and dopamine hypothesis. The exact mechanisms of schizophrenia still remain unresolved and controversially discussed due to the diversity of schizophrenia symptoms. The clinical diagnosis of schizophrenia mainly base on the behavior of patients and rating scales. On the one hand, this kind of diagnosis is labour-intensive and time-consuming; on the other hand, it’s too dependent on descriptions of patients or their relatives and experience of doctors. These problems could disappear if we can build an automatic diagnosis system based on MRI data, making diagnosis more convenient and objective. Finding the potential biomakers of schizophrenia is the key to achieve high accuracy and automatic diagnosis of schizophrenia. Previous studies have revealed widespread abnormalities associated with schizophrenia, especially the reduction of brain gray matter volume and surface thickness. However, these studies showed some heterogeneity due to various factors such as heterogeneity of schizophrenia. In order to reduce the impact of other factors, we conducted a series of experiments on multi-center data trying to find biomarkers of schizophrenia and build automatic diagnosis system.

First, we conducted Voxel-based morphometry (VBM) analysis on multi-center data, in order to reveal the consistent abnormalities in schizophrenia. The results showed that schizophrenia group has significant gray matter reduction and the reduction in superior temporal lobe (STG) and insular are consistent across multi-center data study.

Furthermore, we examined structural abnormalities between schizophrenia patients with auditory verbal hallucinations (AVH) and schizophrenia patients without AVHs, considering AVHs is one of the most devastating symptoms of schizophrenia. Results showed that schizophrenia patients with AVHs has significant thickness reduction in left middle temporal gyrus compared to patients without AVHs.

At last, we trained a model to classify schizophrenia patient and healthy control using morphological features (GM, WM, CSF) extracted from MRI data. We built the model from 640 subjects scanned with Siemens scanner, and tested the unaltered model on a completely independent sample of 635 subjects scanned with different scanners. The classification rate of the independent sample was 75.5%. Feature importance analysis results showed that features extracted from STG and insula have higher importance scores compare to features extracted from other regions, which was consistent with VBM analysis results. These findings suggests that anatomical deficits in STG and insula lobe may modulate the pathogenesis of schizophrenia.