Brain Functional Network Analysis With Applications Based On Adaptively Weighted FMRI Data

Brain functional network(BFN)analysis has been becoming a crucial way to reveal the inherent organized pattern of the brain and explore potential biomarkers for diagnosing neurological or psychological disorders,and has shown increasing potential in clinical applications.This makes constructing or learning a high-quality BFN from data become a very important and noteworthy issue.In recent years,researchers have proposed a series of methods to estimate BFN,such as Pearson's correlation and sparse representation.However,due to the fact that the observed data(such as fMRI)contain a lot of noises or artifacts,the estimated BFN is often unreliable.Therefore,a complex preprocessing pipeline is often used to improve the quality of the data prior to BFN estimation.One of popular preprocessing steps is data-scrubbing that aims at removing “bad” data from fMRI time series according to the amplitude of the head motion.Despite its helpfulness in general,the removal of some “bad” data not only reduce the statistical power,but also lead to different signal lengths across subjects,and thus introduce bias in BFN estimation.In addition,the traditional scrubbing scheme is fully independent to the subsequent BFN estimation task,which cannot guarantee that the removed data are necessarily unhelpful(accordingly,it cannot guarantee that the remaining signals are helpful).In order to address a series of issues caused by the traditional scrubbing scheme,this paper proposes a novel solution for weighting fMRI data,and carries out deep studies on it.First,a new method of fMRI signal weighting based on frame-wise displacement(FD)is proposed,which results in a new BFN estimation method.This method not only reduce the difficulty of threshold selection involved in the traditional scrubbing scheme,but also provide a more flexible framework that scrubs the data in the subsequent BFN estimation model.Second,aiming at the issues that the FD parameters being difficult(or unreliable)to estimate,we further develop a new learning framework.BFN estimation and fMRI data weighted are realized by alternating optimization algorithm,so that the framework achieves BFN estimation and weights data simultaneously.Finally,to verify the effectiveness of the proposed algorithm and framework,we apply the estimated BFN to the early diagnosis of Alzheimer's disease.The experimental results show that our newly estimated BFNs can significantly improve the final diagnosis accuracy and find potential biomarkers.