Measuring The Similarity Of Dynamic Brain Functional Network

Purpose: Brain is a highly complex system.Its functional structure often changes along with a person's disease and aging.The traditional measurements for similarity of brain functional network are mostly based on static networks,which are not suitable for brain networks that are dynamically evolving along time.To address the particular need,we proposed for the first time two original indices,named as Dynamic Network Similarity(DNS)and Dynamic Network Evolution Similarity(DNES)respectively,for measuring the dynamic similarity of brain networks depending on different research preferences.Specifically,the DNS was proposed for measuring the comprehensive similarity of dynamic network,and the DNES was proposed for measuring the similarity of evolution feature of dynamic network.Meanwhile,a software package was developed for the calculation of DNS and DNES.Method: Both simulated data and real-world data were used to test the performance of DNS and DNES,the two original indices we proposed in this study,and the results were compared with those based on traditional similarity measurements.(1)Simulation data experiment: Correlation was analyzed between DNS / DNES and four similarity variables(??,?,?,?)of sublevel features of dynamic networks to examine sensitivity of these two indices.In addition,another set of simulation experiment was designed to study whether DNS and DNES were able to evaluate the general similarity changes between different networks that were dynamically evolving.(2)Real-world data experiment: We acquired resting-state functional magnetic resonance imaging data from 25 stroke patients in the real-world,whom were treated with different therapies(either Transcranial direct current stimulation(t DCS)or Sham)before and after the therapy treatment.We examined the t DCS group in advance using conventional statistical analysis methods and confirmed that their motor network evolved along with the treatment were highly consistent in the rehabilitation process.We also similarly examined the motor network in the Sham group and knew how it evolved during the rehabilitation process.The DNS and DNES indices were employed to evaluate the similarity of the rehabilitation course of this motor network between the patients who received either the same therapy or different therapies,and the results were checked against the aforementioned knowledge that we had acquired in advance.In addition,we compared DNS and DNES with the traditional methods to examine whether using the new indices were superior to using the traditional methods that were originally developed for dealing with static brain networks.Results: Simulation experiments showed that the DNS was significantly correlated with all the four variables of sublevel features ??,?,?,? of dynamic network,and the DNES index was correlated with the sub-features variables ?? and ?.In addition,they were sensitive to general similarity changes between the dynamic networks.Real-world data experiments showed that both the DNS and DNES indices were able to reveal the high similarity in the rehabilitation process of the motor network between patients who received the same therapy.The DNS and DNES values of the dynamic motor network between patients who were treated with the same therapies were significantly higher than the respective values between patients who were treated with different therapies.In contrast,such significances disappeared when traditional methods were used.In addition,even if it was only between the patients who treated with the same therapies,the measured similarities of motor network using traditional methods were significant different before and after the treatments.Conclusion: DNS and DNES,two original indices we proposed for the first time in this original study,provide a way for measuring the similarity of dynamic brain networks.Both the brain networks based on simulated data and real data have proved that DNS and DNES may accurately measure the features of dynamic networks with strong robustness.In addition,the new indices have overcome the instability and low repeatability,where are the main disadvantages of the traditional similarity methods.The traditional methods may only assess similarity at specific individual time points but cannot assess in general the evolving process along with the treatment.In contrast,our new indices comprehended the high similarity embedded in such a general developing process of the motor network among patients who received the same treatments and the inconsistency in this process across patients who received different treatments.As alternative approaches,The DNS and DNES may be adopted in longitudinal studies to quantitatively evaluate the similarity of dynamic brain network between individuals.