The Analysis Of Dynamic Causal Modeling Of Brain Based On Cloud Platform

The rapid development of neuroscience,the generation of big data and the break-through of cloud computing have brought new opportunities for researchers? but at the same time,they have to face the challenge of resource integration.The emergence of the CBRAIN cloud platform not only offers remote access to data,distributed computing sites and a series of pipelines within control,but also provides a simple and easy-to-use graphical interface.Within CBRAIN,users can complete the data processing,analysis and sharing without any programming experience.However,the CBRAIN platform still suffers from a number of shortcomings: the data format is not sufficiently standardized?lack of brain network analysis module.Dynamic Causal Modeling(DCM)as a brain network approach to explore causal effects(effective connectivity)between brain regions,its advantages based on physiolog-ical models have attracted the attention of many scientists,but the disadvantage is that the calculation time of the model increases exponentially and needs strong computing power to back it up.Therefore,in order to speed up the data analysis,this paper rebuilds the existing platform and accomplishes the following work:(1)To better manage neuroimaging data,the standardized data format BIDS(Brain Imaging Data Structure)was used,preprocessing workflow fMRIPrep was introduced,and performance on the cloud platform was tested?(2)The source code and runtime envi-ronment of DCM was compiled and packaged using virtualization technology Singularity,and the feasibility of DCM on a large scale was deployed and validated?(3)Firstly,func-tional MRI(fMRI)data from Parkinson's Disease(PD)and Healthy Controls(HC)on PPMI(Parkinson's Progression Markers Initiative)were chosen,and the cloud platform was used to implement preprocessing acceleration and apply large-scale DCM for brain network analysis.Secondly,we chose the effective connectivity as a set of features and then feature selection method was applied to decrease the dimension and filter out the sub-set of features.Finally,an oversampling method was proposed to deal with the problem of unbalancing and oversampling,and several classifiers were employed to evaluate the AUC performance?(4)A statistical correlation method was used to analyze the correlation between the subset of features and the clinical scales of PD patients.Experimental results indicate that CBRAIN cloud platform can enhance the capabil-ity of data preprocessing to varying degrees.The test results of large-scale DCM analysis show that the higher the model complexity,the greater the platform's advantage in com-puting power.Further analysis of the subset of features obtained from large-scale DCM,the best AUC value was better than that of existing studies,and a subset of these features were significantly negatively correlated with the clinical scales,which indicates that the features have certain physiological significance and may provide assistance in the early di-agnosis of PD.Through large-scale DCM analysis,this paper demonstrates the advantages of cloud platform in terms of computing power and the feasibility and effectiveness of ap-plying complex models.By tackling the problem of insufficient performance of outdated computer hardware,CBRAIN provides technical support for more complex approaches in the future.