Research On Removing Site Effects From Multisite MRI Data

Pooling magnetic resonance imaging(MRI)data across research studies to sharing larger sample data,presents exceptional opportunities to advance of neuroscience research.Merging MRI data from multi-site studies has great potential for obviating some of the shortcomings of single-site study,and will significantly enhance the reliability and reproducibility of the multisite results.However,scanner/site confounds collected on different sites or studies can lead to spurious findings,which will hinder discovering the mechanism of psychiatry brain disease based on multi-sites larger sample MRI data.Unfortunately,the most widely used methods to removing scanner/site noise-parametric model method and data-driven method,cannot effectively remove the impact of site noise from multisite MRI data.In real applications,the scanner/site noise and the variables of interest are often correlated at different levels.In order to avoid destroying the information of signal variables,the noise components related to signals will be retained during scanner effects denoising,which will lead to incompletely noise elimination.In this dissertation,we propose an effective dual projection regression model to remove the sits noise effectively without losing the information related to the signals of interest.This method can combine with both parametric regression method and the data-driven method to achieve the best denoising performance.This dual projection proposal could identify the effects of site noise accurately and comprehensively,and shows high sensitivity,high specificity and high efficiency on removing the scanner/site effects.Simulated multisite MRI data and the real multisite MRI data were used to test the effectiveness and accuracy of the proposed denoising modal.The dual-projection denoising modal shows better performance than original parametric modal and data-based denoising modal.It can effectively remove the scanner/site effects without weakening the signal information.It can also combine with tensor decomposition methods to denoising the scanner/site effects from multi-modal MRI data.This approach will ha great promise for getting accurate and reliable results of multi-sites MRI data,avoiding the confusion neuroscience results,and promoting neuroscience researchers perform deeper and systematic psychiatry brain disease study.