Study On The Characteristics Of Brain Network In Depressive Patients Based On Functional And Structural Images

Depression is an important hidden danger that threatens mental health of people.Once it happens,it will bring great pressure on individuals and families.But now there is no objective biological evaluation index for the diagnosis of depression.At the same time,the existing medical resources are very scarce,which lead to the depression patients can not receive timely diagnosis and treatment.Therefore,it is very important to carry out computer-aided diagnosis of depression with machine learning algorithm.Among them,it has become a hot spot to explore depression by combining various MRI techniques.With the development of interdisciplinary research,graph theory is an important branch of neuroscience.In the previous studies,most of them used resting-state functional magnetic resonance imaging（rs-fMRI）or diffusion tensor imaging（DTI）to construct individual or group network for study.However,the study of only one modal data cannot fully show the alterations of brain,and there are some problems such as incomplete experimental results,the single feature and so on.Therefore,in order to overcome the limitations of previous single modal study,we will use rs-fMRI and DTI data in this study to construct a personal network for joint analysis.From the perspective of function and structure,it provides a new ideas to explore the potential neural mechanism of emotional disorders and cognitive impairment in patients with depression.Meanwhile,we also use two modal data networks to find effective biomarkers,which can provide reliable basis for computer-aided diagnosis of mental diseases.The study of this paper mainly includes the following three points:Firstly,the characteristics of the alterations of the whole brain functional network and structural network of depressive patients were studied systematically.In this study,the method of multimodal joint analysis is used to construct the functional network and structural network of individuals respectively by using rs-fMRI and DTI data from the perspective of graph theory.Through the difference analysis of network perproties,the influence of depression on brain function and structure was revealed.The results show that the small-world properties of the brain in depressive patients are lost,and there are also abnormal alterations in network connection and node efficiency.Meanwhile,the abnormal functional connections and morphological abnormalities of specific brain regions in depressive patients are related to the clinical score of depression.In the early stage of depression,the research on the robustness of structural network and functional network shows that structural network is more vulnerable to attack than functional network.Secondly,we explored the alterations of brain network in rectal cancer patients with depression tendency after surgery and chemotherapy.In this study,we used the functional and structural images of rectal cancer patients to construct brain network,and applied graph theory to analyze.The results showed that,comparing with the normal control group,rectal cancer patients showed abnormal small-world properties and global topology in the functional and structural networks.At the same time,when analyzing node efficiency,we found that the efficiency of some brain nodes in patients with rectal cancer increased.These results suggested that there was a compensatory mechanism in the brain network of rectal cancer patients to maintain normal physiological activities.Thirdly,we propose a research framework for the diagnosis of depression using multimodal brain network properties.In this study,we extract the features of multimodal brain network properties,and use machine learning method to diagnose depression.In the experiment,rs-fMRI and DTI images are used to construct individual functional network and structural network.Cluster coefficient()and network connection strength are extracted from the network as features.The minimum redundancy and maximum correlation（mRMR）and F-score algorithms are used for feature selection.Support vector machine（SVM）is used for classifier training,and 10fold cross validation is used to evaluate the effectiveness of features.The experimental results showed that the accuracy of depression classification reached 85.14%.This result showed that the properties value of multimodal brain network can be used as an effective biomarker to identify depression patients more accurately.