چكيده به لاتين
Multiple sclerosis (MS) is a chronic autoimmune disease in which the immune system mistakenly attacks the protective covering of nerve fibers, leading to communication problems between the brain and the rest of the body. Some MS patients have cognitive disorders such as working memory disorders. In this research, in order to investigate working memory, the n-back task was used with three different workloads. Past studies have achieved significant results in the diagnosis of various neurological diseases by examining the non-linear features of the electroencephalogram signal. The aim of this study is to investigate the non-linear features of the blood oxygen level dependence (BOLD) signal and use these features to separate healthy and MS groups. For this purpose, using the AAL atlas, the brain was divided into 116 regions and each voxel was labeled, then the BOLD signal of the corresponding voxels was averaged. In the next step, for each of these time signals, 6 non-linear features including Higuchi's fractal dimension, Katz's fractal dimension, Hurst exponent, Lyapunov exponent, Sample Entropy and Approximate Entropy were extracted. Wilcoxon rank-sum test was performed on the extracted features between the healthy and MS groups to determine the significant and distinguishing features between the two groups. The results showed that Higuchi's fractal dimension is significantly different in many areas in all workloads between MS and healthy groups. Next, in order to classify the two groups after determining the training and testing data sets using the Wilcoxon rank-sum test, the best features were selected from the training dataset and from 3 different classifications (Support Vector Machine, K-Nearest Neighbors and Naïve Bayes) were used for the classification task. After examining the results, it was found that the best result is obtained when the combination of all non-linear features, Support Vector Machine classifier and 1-back workload are used for classification; The result of accuracy, sensitivity and specificity of the classification reached 95%, 87.5% and 100% respectively. The results showed that the non-linear features of BOLD fMRI signal perform well in the diagnosis of MS.
