Construction and application of bayesian network in early diagnosis of alzheimer disease's system

Most of methods in neuroimaging studies are usually based on voxel-wise statistical tests which are performed on functional or structural magnetic resonance(MR) images across or within subjects. Such methods( i.e.,t-test and paired t-test ) are univariate in nature and can only describe the linear association between the brain structure/function and other variables of interest. In this study, we propose a Bayesian network approach combining the data among MR images and clinical/cognitive function variables to assist the early prevention strategy and diagnose the mild cognitive impairment(MCI). The Bayesian network is a powerful knowledge representation and reasoning tool under conditions of uncertainty. Our application of such procedure describes nonlinear multivariate relationships between the human brain structures, function variables and other measurements of interest. Cross-validation approach was utilized in our study for the construction and evaluation of the proposed Bayesian network model. Twenty-five MCI subjects were selected from whom we acquired structural MR data. These MR images were then segmentated and registered to the MNI template and with automated structure volume calculated and made discrete. These volumes included hippocampus, thalamus and perirhinal, etc. With these imaging data and other clinical measurements, we used heuristic search method to construct our Bayesian model and evaluated its reliability using a scoring method. This construction procedure included the training of the structure and estimation of the parameters of the Bayesian network using MR data from 15 these subjects included in this study. To evaluate the use of the proposed model, inference was made with the masked variables using the MR data from the remaining subjects. It was found that MCI was mainly dependent on hippocampus, thalamus and entorhinal. Our method could be used to early diagnose the presence of the mild cognitive impairment from structural MR images and behavioral data.