Research On Brain Age Estimation Algorithm Based On Magnetic Resonance Image Data Mining For Dementia

In recent years,neuroimaging technologybecomes more and more popular in the early diagnosis of dementia.Among neuroimaging technologies,the Magnetic resonance imaging(MRI),which has advantages of high resolution,no contrast media,no radiation,low price,and can quantatively reflectthe changesof structure and function of brain as well as metabolite concentration,has achieved good results in the early diagnosis of dementia.However,the application of MRI is mainly based on the visible information for diagnosis.In addition to the visible changes,MR image also contains some invisible changes which are often characterized as the evolution of the nature of the disease,provides more substantial and early information for disease diagnosis,and can obtain higher quality early MR imaging marker.Brain age is one of the representative imaging marker.At present,the real age is used as the label of the age estimation model,and the model is trained by minimizing the difference between the estimated age and the actual age,which is different from the hypothesis that the AD process is a form of accelerated aging.Since there is a deviation between real age and brain age in different stages of AD,the real age as label will lead to error without question.Based on this,this paper makes a further study on brain age estimation for diagnosis of dementia,and aims to improve the performance of brain age estimation.In this paper,not only the existing algorithm is used for brain age estimation of AD,but also a new algorithm is proposed for brain pathological age estimation based on AD.Besides,the two algorithms are applied into the brain age estimation of Subcortical ischemic vascular dementia(SIVD).The main work of this paper is as follows:(1)A brain age estimationestimation algorithm is studied and implemented based on MR image of AD.(2)A brain pathological age estimation algorithm based onseparability distance criterion and SVR is proposed.In this algorithm,real age plus age deviation is usedas training label to replace the real age;the ratio of interclass variance to intraclass variance is designed as a new fitness function,replacing the difference between brain age and real agein the existing algorithm.(3)The brain age estimation algorithm of SIVD is proposed based on PCA and Fisher LDA,and the effects of feature reduction and different kernel functions on brain age estimation are studied and analyzed.(4)The brain pathological age estimation algorithm of SIVD based onseparability distance criterion and SVRis put forward and researched,where real age plus age deviation is used as the training label to replace the real age in the existing algorithm;a new fitness function is designed;the ratio of interclass variance to intraclass variance is designed as a new fitness function,replacing the difference between brain age and real age in the existing algorithm.The brain age estimation of AD and VD based on MR imagesare respectively conducted in the paper;the effect of the estimated brain age on diagnosis of dementia is studied.Not only the existing brain age estimation algorithmis realized for AD and SIVD,but also a kind of brain pathological age algorithm isproposed for AD and SIVD.This study will lay foundation and provide new idea for subsequent research of brain age estimation and early diagnosis of dementia.