英文摘要 | Schizophrenia is a complex mental disorder characterized by abnormalities of behavior, thinking, emotion, etc. The cause of the disorder is unclear, and both genetic and environmental factors were reported to work together. The onset of the disorder is mostly found at teenagerhood, and thereafter patients are hard to be cured, leading to heavy economic burden to family and security burden to society. Therefore, to elucidate the pathogenesis and to find biological markers for diagnosis and treatment, become urgent global-wide tasks for researchers of the disorder.
In the current paper, we use resting-state functional magnetic resonance imaging (rs-fMRI) to search for objective biological markers of the disorder with sample-wise comparisons. Later, we tested classifiers with rs-fMRI measurements as feature to perform automatic classification. The classification performance of the existing schizophrenia classification models was evaluated and the common attributes among the misclassified schizophrenia patients were analyzed. During the period processing rs-fMRI data, we developed a one-stop fMRI data processing software package based on MATLAB, called BRANT, with functions covering almost all processing steps.
The main contributions are summarized as follows:
- Study of abnormal functional connectivity in schizophrenia using a multi-center schizophrenia dataset. In the current study, we used a large sample of rs-fMRI data from seven datasets to investigate group-wise difference of functional connectivity between schizophrenia patients and normal controls. We have found abnormal functional connectivity in among subareas of frontal-parietal lobe, frontotemporal lobe, thalamus, and default mode network. Besides, we found the reduced functional connectivity of left superior temporal gyrus (A22c) and right inferior occipital gyrus (iOccG) in schizophrenia was negatively correlated with the negative PANSS score, which was associated with the impaired multisensory integration function in schizophrenia suggested by previous studies.
- Classification of schizophrenia patients and normal controls. In previous classification studies of schizophrenia, a proportion of schizophrenia patients ranging from 10% to 30% is consistently misclassified but rarely discussed. Therefore, it is interesting for us to find out common characteristics shared by misclassified schizophrenia patients. In the current study, we performed binary classification on a large dataset of 1082 subjects using four popular models. The overall classification accuracy reached a range between 81.96% and 83.54%. Then, we collected schizophrenia patients that are either misclassified or correctly classified across all classifiers and conducted sample-wise comparisons between the two groups. In the results 148 significantly increased functional connectivity in the misclassed group were found among subareas of inferior frontal gyrus, orbital frontal gyrus, insula, insular-opercular, medial prefrontal cortex and medial frontal gyrus. Among the increased functional connectivities, the left inferior frontal gyrus (A45r) – left insula (vId/vIg) connectivity is found having significant negative correlation with positive PANSS score (R=-0.36, P=0.006) in the misclassified group, but not in the correctly classified group (R=0.05, P=0.26). Our results suggest the relationship between inferior frontal gyrus-insular functional connectivity and patients’ positive symptoms is different in the misclassified group, leading to failure of functional connectivity-based classification.
- One-stop fMRI batch processing software. Existing data processing toolboxes for resting-state functional MRI (rs-fMRI) have provided us powerful tools and friendly graphic user interfaces (GUIs). However, many toolboxes only cover a certain range of functions, and use exclusively designed GUIs. To facilitate data processing and alleviate the burden of manually drawing GUIs for new functions, we have developed a versatile and extendable MATLAB-based software, BRANT (BRAinNetome fMRI Toolkit), with optimized calculations, efficient file handling method, and code-generated GUIs. Functions of BRANT cover a wide range of fMRI data processing, including batch preprocessing pipeline, brain spontaneous activity analysis, functional connectivity analysis, complex network analysis, statistical analysis, and results visualization. With BRANT, users can find efficient functions for batch processing, while developers can quickly publish scripts with code-generated GUIs.
|
修改评论