The Complexity Of Resting-state FMRI Signals Was Examined Using Different Variations Of Entropy

Resting-state functional magnetic resonance imaging（rs-fMRI）can reflect the spontaneous neural activity of the human brain,and entropy,as a complexity measure,can be used to analyze the complexity of the human brain,which is a nonlinear system,to explain the activity of neurons.In recent years,more and more studies have started to use entropy to analyze rs-fMRI signals.These studies have used a variety of approximate measures of entropy,but few studies have been conducted to analyze more systematically the reliability and validity of different entropy methods when applied to rs-fMRI signals.In this paper,brain entropy was calculated based on rs-fMRI signals using seven entropy methods（approximate entropy,sample entropy,fuzzy entropy,permutation entropy,range entropy,dispersion entropy,and differential entropy）with the aim of exploring the test-retest reliability and validity of multiple entropies.Also,brain age prediction and brain aging study were conducted using seven brain entropy to further verify the validity of entropy.The specific methods and results are as follows:Firstly,the test-retest reliability study.We calculated and analyzed the intraclass correlation coefficient and their fluctuation changes corresponding to seven BENs at different number of repetitions using ten repetitions of the Midnight Scan Club dataset,where the differential entropy showed the lowest fluctuations and the highest average intraclass correlation coefficients（0.49,0.5,0.5）when k varied（3,6,10）,and the results were validated on three brain atlases.Subsequently,the simulation experiments showed that the entropy curves of differential entropy and approximate entropy are the smoothest.Conclusion:Differential entropy has excellent test-retest reliability.Secondly,the effectiveness study.We systematically evaluates the effectiveness of seven entropy methods by verifying the classification performance of different BENs in binary classification tasks on three mental disorders（attention deficit and hyperactivity disorder,bipolar disorder and schizophrenia）and healthy control,and reports the critical brain regions affecting the classification by the weights of SVM.Conclusion: The classification results showed that differential entropy was more applicable to the attention deficit and hyperactivity disorder classification task（accuracy = 0.8548）,sample entropy was more applicable to the bipolar disorder classification task（accuracy = 0.8824）,and range entropy was more applicable to the schizophrenia classification task（accuracy = 0.7841）.Thirdly,the brain age prediction and brain aging study.Brain age prediction was successively performed at the whole brain and network levels on lifepan（6-85 years,N=514）data of healthy people using seven entropy methods,with differential entropy performing best with a coefficient of determination of 0.6175 and a correlation coefficient of 0.7870.The network-level prediction results suggested that some networks（such as the cinguloopercular network）can better reflect brain aging process.Next,differential entropy was chosen to conduct brain aging study:（1）The results of the correlation analysis showed that at the network level,changes in brain entropy on the cingulo-opercular network best reflected the brain aging process.（2）The results of the one-way ANOVA revealed the ten brain regions that may play the most critical role in brain aging such as left central opercular cortex,left postcentral gyrus.Conclusion: Among the seven entropy methods,differential entropy may be the best entropy method to analyze brain aging and can be used for brain age prediction tasks.