Research On Pattern Recognition Of Brain Diseases Based On Multi-dimensional Network Spatial Feature

The pathogenesis of brain diseases is complex,which makes the clinical diagnosis of brain diseases very difficult.In recent years,with the development of electrophysiological signal acquisition and other medical imaging technologies,the datadriven based diagnosis of brain disease has become a new research hotspot.By exploring the differences in latent features in electrophysiological signal and imaging data of the brain,it is also beneficial to explore the pathogenesis of brain diseases.However,most of the existing studies are usually based on low-dimensional features such as global attributes when characterizing the features of the data.Such features often lack effective discrimination ability,especially when characterizing the complex brain networks related to brain diseases.To this end,we attempt to characterize the spatial patterns of the brain network,and achieve high-precision identification of brain diseases based on the multidimensional network spatial features.In order to explore the generality of this method,we selected three representative brain disease-related groups including epilepsy,schizophrenia,and the aging population,and carried out this study on the corresponding data sets.Aiming at the problem that patients with refractory epilepsy without structural abnormalities are easily misdiagnosed with the medically controlled epilepsy patients,we first attempt to realize the identification of the two types of patients based on resting-state scalp EEG signals.By the comparison of the network at the group level and the analysis of the network spatial pattern,we verified the importance of the beta-band brain activity of some brain regions like the left temporal lobe in characterizing the differences between the two types of patients.Based on the spatial pattern analysis,the highest recognition accuracy of 90.00% was achieved for differentiating the two types of patients.Further,we evaluated the classification performance based on the fused spatial features extracted from the functional networks and the effective networks,so as to explore the effect of the fused spatial features for the identification of epilepsy brain networks.The results show that the recognition accuracy is up to 96.67% based on the fused spatial features,which is significantly improved compared to the single-network.In order to explore the generality of the above result,we further conducted the evaluation based on the functional and structural networks of the MRI datasets referring to the research of aging and schizophrenia.By introducing diffusion map embedding,we find that different ways to fuse the spatial feature fusion could lead to different effects on the recognition effect.Herein,the recognition accuracy of aging datasets can be increased from 90.00% to 91.15% based on the fused spatial feature,and the recognition accuracy was also increased from 77.50% up to 87.50% for the schizophrenia datasets.In summary,based on the analysis of brain networks and different network features,we evaluated the classification performance based on the fused network spatial features on the dataset of epilepsy,schizophrenia,and aging humans respectively.The results show that the fused network spatial features are more effective in representing complex networks,and could be well generalized to other brain diseases.This method is not only conducive to the high-precision identification of brain diseases,but also may provide a new perspective for exploring the pathological mechanisms of brain diseases.