Influence of resting-state functional brain network’s time duration on recognizing major depressive disorder

Abstract

Machine learning has recently been applied into automatically recognizing major depressive disorder by taking functional connectivities as classification features. It opened the good clinical application of computer assisted major depressive disorder diagnosis. However, it is still unclear how much the neuropathology information of major depressive disorder can be captured by resting-state functional brain networks with different time durations, and it is unknown what the influence of resting-state functional brain network’s time duration on recognizing major depressive disorder is. The present research established the nonlinear models that describes the influence of functional brain network’s time duration on recognizing major depressive disorder by using the method of nonlinear regression, and illustrated the neuropathology information of major depressive disorder that is represented by functional brain networks with typical time durations. In the experiment, the resting-state functional magnetic resonance imaging (rs-fMRI) and Hamilton depression ratings were acquired, and the data of 64 clinical first-episode major depressive disorder patients and 53 control subjects were analyzed. The study constructed the resting-state large scale functional brain networks and the functional connectivity matrices under the anatomical automatic labeling (AAL) atlas and rs-fMRI data for each subject, and found the significant functional connectivities used as classification features by two sample t-test between the patient and control groups. Then, the sensitivity, specificity and accuracy of the support vector machine classifier were obtained by the leave one subject out cross validation. By modeling the relationship between the functional brain network’s time durations and classification performances, we obtained the nonlinear curve models of classification performances. The functional brain networks with about 46 TRs length have credible classification performance for the first time, and it show that the patient have enhanced functional connectivities between the posterior cingulate gyrus and the orbital frontal cortex, between the right orbital frontal middle gyrus and the left angular gyrus, and between the left gyrus rectus and the left hippocampus. The functional brain networks with about 114 TRs length have the best classification performance, and capture some more weakened functional connectivities between the right angular gyrus and the bilateral inferior parietal but supramarginal and angular gyri, middle frontal gyrus, between left amygdala and left supramarginal gyrus, right rolandic operculum. These abnormal functional connectivities support the hypothesis that major depressive disorder is the neurogenic disorder of distributed brain network with abnormal interactions. With the increase of time duration, functional brain networks capture certain functional connectivities that may not be directly related to the neuropathology mechanisms of major depressive disorder, and hence the classification performance decrease. Thus, the law of the influence of resting-state functional brain network’s time duration on recognizing major depressive disorder appears to be the inverted U-shape tendency. This may provide certain new references for further effectively investigating the neuropathology mechanisms and improving the intelligent recognition effects of major depressive disorder.