Research On Extraction Of Brain Functional Network Features For Individualized Brain Age Estimations

The resting state functional magnetic resonance imaging(f MRI)technique has been widely used in the study of human brain cognitive function and the evaluation of individual cognitive parameters.Extracting effective brain function features is essential for evaluating individual cognitive parameters based on resting state f MRI,because the effectiveness of brain function features directly relates to the accuracy of subsequent individual cognitive parameter estimations.Recent studies have shown that human brain age(rather than physiological age)is an important aspect that reflects the cognitive function of the human brain.The deviation of human brain age and physiological age is closely related to mental illness.Therefore,human brain age estimation has received widespread attention from researchers.In this study,we will use the human brain age assessment to study how to effectively extract hidden brain function information in the resting state f MRI data.The specific work of this article is as follows:(1)Feature extraction based on earth mover's distance(EMD).EMD can evaluate the similarity between two time series in the presence of a time difference,which allows EMD to break through the limitations of traditional brain function extraction that correlation based on two time series.In this study,EMD was introduced to measure the strength of the functional connection between brain regions.then the three methods of support vector regression,elastic net,and random forest were used to build a human brain age estimation model to verify the effectiveness of EMD-based strategies.The experimental results show that the effect of human brain age assessment based on EMD features is slightly worse than that of traditional correlation coefficient features(the correlation coefficient between the evaluation value and the real value R = 0.66 vs R = 0.78),but the error is within the acceptable range.(2)Feature extraction based on dynamic time warping(DTW)algorithm.The DTW algorithm supports the dynamic selection of the minimum distance between two time series to solve the problem of asynchronous correlation between signals.This feature of the DTW algorithm can still effectively solve the limitation that traditional brain function feature extraction only considers fully synchronized signals.In this study,the DTW algorithm was introduced to extract brain function features,and four different objective functions were used to measure the asynchronous correlation of time series.The experimental results show that high-precision human brain age assessment can be achieved based on the brain function features obtained by two of the target functions(the correlation coefficient R=0.78 between the evaluation value and the real value is equivalent to the feature based on the correlation coefficient).(3)Feature extraction based on cross recurrence plots(CRP)and recurrence quantification analysis(RQA).Using traditional feature extraction methods based on correlation coefficients,the dynamic correlation between time series cannot be found.However,CRP can obtain the data correlation of any two time points in two time series,so as to mine dynamic features.This study uses CRP to obtain the dynamic correlation of two time series,and then uses the RQA method to explore the multi-angle correlation between time series.The experimental results show that the brain function features extracted based on CRP and RQA also can achieve a estimation of the age of the human brain(the correlation coefficient between the evaluation value and the real value R = 0.72).The innovation and significance of this research lies in: through the introduction of three feature extraction algorithms,it provides a new idea for carrying out brain function research and individualized evaluation based on resting state f MRI.Although the individual age predictions based on EMD(R = 0.66)and CRP + RQA(R = 0.72)is not so good as compared to predictions based on functional connectivities calculated using correlation coefficients between f MRI times series(R = 0.78),these features are expected to be useful supplements to traditional features.