Dynamic Network Communities Interaction Assessing Depression Treatment Via Resting-state FMRI

Depression is a usual kind of psychogeny with a high disability rate and can causes serious harm to patients' body and mind.However,the pathological mechanism of depression is unknown and antidepressant drugs are not very effective for patients.In previous studies,most of the research assumed that the signal was approximately stationary when they assessed depression treatment via resting-state functional magnetic resonance imaging(rsfMRI).However,few people pay attention to the role of rsfMRI's dynamic attribute in treatment assessment of depression.So the dynamic characteristics based on the community detection algorithm were calculated and features of dynamic communities were designed to assess depression treatment in this paper.The features of dynamic communities express the dynamic properties of the single brain region and dynamic interaction between brain regions separately.At the same time,the two kinds of features of dynamic communities under special time periods were used to assess depression treatment further.This dynamic research is aim to provide more effective supplementary evidence for the clinical evaluation in depression treatment.The details of research are as follows:(1)Dynamic community features were used to assess the brain networks changes between pre-therapy and post-treatment depressive patients.Dynamic communities based on the community detection algorithm were calculated and the dynamic communities feature flexibility was analyze to find the differences between the pre-therapy,post-treatment patients and the health via permutation test and pair permutation test.Flexibility is one of traits of dynamics communities and means the number of times a region changes.The result showed that the distribution of the flexibility values is significant between the pre-therapy patients and the health in the default network and cognitive control network but isn't significant between the post-treatment patients and the health in two rsfMRI databases in China.At the same time,there were significant flexibility values differences between the pre-therapy and the post-treatment patients in the default network using the rsfMRI database with 8-week treatment and significant differences in the default network and the control network using the rsfMRI database with 3-week treatment.Then,a significant linear correlation was found between the flexibility values and the 17 Hamilton Depression scale scores in the default network and the control network of the pre-therapy and post-treatment patients.In a word,flexibility shows the dynamic properties of rsfMRI recordings and can illustrates the differences in brain network changes before and after treatment of depression clearly.(2)Dynamic communities features were used to predict the effect of short time escitalopram treatment via rsfMRI of depressive patients who were hospitalized before treatment.Module allegiance matrix expresses the correlation between regions in dynamic communities.Jump matrix expresses the dynamics of correlation between regions in dynamic communities.The method of jump matrix highlights the unbound relationship between adjacent time windows which can predict the effect of short time escitalopram treatment with flexibility and moduleallegiance matrix.Support vector machine was used to predict the effect of short time escitalopram treatment via the above dynamic communities features.The results showed that the accuracy rate of predicting effect of escitalopram treatment by flexibility and jump matrix were both 79.41% and the results were robust.Support vector machine found that the flexibility differences between response and non-response patients focus on brain region level but the jump matrix differences focus on brain network level.At the same time,module allegiance matrix further analyzed the different brain regions found by flexibility and found a core brain region distinguishing escitalopram treatment effect.All above features provided an effective supplementary basis in distinguishing patient who is suited to use escitalopram.(3)Dynamic communities features during the specific time periods were further used to predict the effect of escitalopram treatment.The values of flexibility and jump matrix were mainly concentrated in a few time windows according to analyze the distribution of flexibility and jump matrix values during the whole scanning time.So,the result showed that the active of brain area itself and brain areas interaction is not stable during the whole scanning time but concentrated in a few time periods.Then,response group and non-response group were found with the similar unstable patterns via comparing instability statistics.What's more,several largest flexibility value time windows and jump matrix time windows was selected to distinguish the response group and non-response group.The results showed that the classification accuracy was over 80% when predicting effect of short time escitalopram treatment with the largest 2 to 5 flexibility value time windows and the highest robust classification accuracy result was 88.24%.Two test databases with same condition were collected synchronously and their test accuracy were 86.49% and 85.29%,respectively.Moreover,highest robust classification accuracy was 94.12% with several largest jump matrix time windows.Test accuracy of the two test databases were 81.08% and 82.35%,respectively.In short,dynamic communities features during the specific time periods have more effective forecast accuracy on short time effect of treatment and can provide more effective supplementary basis for taking escitalopram.