Model Study Of Depressive Clinical Factor Based On Resting-state Functional Connectivity

The clinical diagnosis of suicidal ideation in depressive patients was with strong subjective perspectives. Machine learning algorithms combined with neuro-imaging techniques were approved to be able to assist in the objective diagnosis in recent works. But for individuals without typical suicide ideation, it is difficult to get a feasible interface in feature space to define the suicidal ideation of them due to the transient characteristic.Therefore this work applied semi-supervised learning as well as resting state fMRI functional connectivities to construct a suicidal ideation predicting model, which is able to assist the objective diagnosis of suicidal ideation in MDD patients. Semi-supervised learning algorithms were believed sensitive to subtle differences between the transition states. And it had successfully predicted mild cognitive impairment in different states, which disease distributes in the same way as suicidal ideation do.Firstly, the paper investigated an objective suicidal ideation diagnosing model using whole brain functional connectivities based on resting-state fMRI signals. By means of semi-supervised learning algorithm, the resulted functional connectivity features were then selected from high dimension features due to their weights calculated by feature cycling, kendall index and Fisher criterion. The transition states of depressive suicidal ideation were investigated by iterative self-organizing clustering algorithm(ISODATA). The study confirms that suicide ideation is a gradual process, while frontal-temporal circuit plays an important role in distinguishing the different stages of it. Moreover, by calculating the distance ratio of each sample to the centers of both extreme groups in feature space, the suicidal ideation of this sample can be estimated. This way the brain network impairment can be reflected better than using traditional ways.After that, dynamic modularity of brain regions was utilized to construct a spatiotemporal predicting model where the temporal correlation between modules was denoted as coupling index. Based on that, delltime and Flexibility were analysed as crucial characteristics to construct a suicidal ideation diagnosing model. The frontal and temporal lobes, as well as the amygdala brain area were found playing an important role in finding the biomarker of impulsive suicidal ideation. And the result of this study suggested that feature fusion is capable of improving clustering stability by giving up a little clustering quality.In conclusion, we explored an objective diagnosing model of suicidal ideation in depressive patients from multiple perspectives including static functional connectivity to dynamic modularity. These aberrant functional connectivities and dynamic modularity features were suggested to be related with dysfunction of cognitive and emotional disorders of depressive patients, who may help us to deepen the understanding of the neurophysiological mechanisms of depressive suicidal ideation.