Functional Connectivity Analysis Of RfMRI For AD Patients Based On Improved EDC And NE-FastICA

The prevalence of alzheimer's disease in the elderly is increasing year by year,and the risk and proportion of the disease in women are twice that in men.The pathogenesis of alzheimer's disease in women may be different from that in men.The changes of brain function in AD patients are much earlier than the changes of brain structure.Resting-state functional magnetic resonance imaging is an important imaging method to study the pathogenesis of AD.In this study,we studied the functional connection of female AD patients in four different stages: normal,early mild cognitive impairment,late mild cognitive impairment and AD,to provide evidence for revealing the pathogenesis of female AD.The fast independent component analysis algorithm based on negative entropy has the advantages of fast convergence,high separation accuracy and high robustness.It is a commonly used functional connectivity analysis method at present.However,because NE-FastICA is sensitive to the initial value of randomization,the separation result is unstable,and it is difficult to select and determine the region of interest.At the same time,due to the existence of overestimation when determining the number of independent components,this paper proposes functional connection analysis method based on improved effective detection criteria and NE-FastICA.EDC has high estimation consistency and flexible penalty function.It converges relatively fast and has high robustness.It is often used to estimate the number of independent components.However,different penalty functions and different parameters lead to overestimation of the number of independent components.This paper proposes an improved strategy of introducing penalty functions into logarithmic functions and combining golden section method to determine the optimal value.NE-FastICA is a blind source separation method based on Newton's method which is sensitive to the initial value.It often leads to unstable separation results.In this paper,the steepest descent method and Newton's method are combined to improve the sensitivity of initial value.Armijo criterion is used to determine the step size of the steepest descent method,and the optimal relaxation factor is used to optimize the step size.Component selection is essentially a clustering problem.FuzzyC-means clustering algorithm based on Euclidean distance does not consider the correlation between components.In this paper,a FCM clustering method based on correlation distance is proposed.The correlation distance calculated after the absolute value of correlation coefficient is taken as a measure of non-similarity,which reduces the dependence on prior network template and the influence of subjective factors when selecting components of interest.The simulation results show that the proposed improved EDC avoids the problem of overestimation and improves the estimation performance significantly.The improved NE-FastICA algorithm reduces the dependence of the algorithm on initial values and improves the convergence speed.The results of the improved FCM clustering algorithm can be used to determine regions of interest and perform functional connectivity analysis.Statistical results showed that the functional connectivity between regions of interest such as visual syndesmosis cortex and angular gyrus and supramarginal gyrus was weakened when CN developed into EMCI.When EMCI developed into LMCI,the functional connectivity between somatosensory syndesmosis cortex and angular gyrus,visual syndesmosis cortex and middle temporal gyrus was strengthened,and the negative function between visual syndesmosis cortex and supramarginal gyrus began to appear.When LMCI developed into AD,functional connectivity between regions of interest such as visual synaptic cortex and supramarginal gyrus decreased significantly.