融合先验知识的连边功能连接震后PTSD分类方法OA北大核心CSTPCD

It is of great clinical significance to extract the specific neuroimaging features of patients diagnosed with seismic post-traumatic stress disorder(PTSD)and design its categorical model.Compared with the traditional neuroimaging classification model,this study used prior knowledge to extract ROIs to reduce a large number of noisy signals.Meanwhile,the research also changed the functional connection mode and proposed a classification model named RFE-MLP,which effectively improved the classification accuracy of post-earthquake PTSD and post-earthquake non-PTSD control groups.Firstly,in view of a large number of noisy signals in the whole brain,the PTSD-related brain regions were extracted by using prior knowledge,focusing on the feature construction of PTSD-related brain regions after the earthquake.Secondly,the traditional node function connection was replaced by the edge-centric function connection to obtain the high-order feature information of the brain function connection.Finally,the RFE-MLP classification model was proposed,which learned the model weight information adaptively and improved the accuracy of PTSD after earthquake.The results showed that the classification accuracy(ACC),sensitivity(SEN),specificity(SPE)and area under curve(AUC)values of post-earthquake PTSD patients and post-earthquake health control groups reached 92.5％,93.5％,92.1％and 94.8％,and it was found that the left middle frontal gyrus(Frontal_Mid-L)had the highest correlation with post-earthquake PTSD severity.The experimental results show that this method extracts more key feature information from fMRI,improves the accuracy of classification and diagnosis of PTSD after earthquake,and facilitates the precise localization of brain regions related to PTSD after earthquake and other trauma types.