Multiple Sclerosis Detection Based on MR Images Using Features Pseudo Zernike Moments, Gray-Level Co-Occurrence Matrix, and Imperialist Competitive Algorithm

Multiple Sclerosis (MS) is an inflammatory disease in which myelin sheaths in nerve cells of the brain and spinal cord are damaged. The symptoms of this disease do not appear until its advanced level and accurate diagnosis can be made only after the onset of disease. MR images are a diagnostic tool that is currently used for diagnosing MS. It enables specialists to detect lesions caused by this disease in the brain. However, the analysis of these images requires experience and expertise of the physician and different interpretations could be reached during diagnosis. Sometimes the dissemination of lesions caused by diseases such as Alzheimer s, MS, and other diseases involved in the brain is not possible because of the nature of images of MR images. Thus methods based on artificial intelligence are common techniques that can distinct patients with MSs and healthy ones via MR images. In the present paper, a set of features are extracted from brain MRI by descriptors such as Pseudo-Zernike polynomials and co-occurrence matrices. Then in the next phase, the best features are selected via colonial competitive algorithm (CCA). Finally, relying on a support vector machines (SVM) whose parameters have been optimized by the genetic algorithm, the existence or absence of disease is diagnosed via the sample MR images. The implementation was carried out using by laboratory database at the university of Cyprus as well as pictures of clinical center of Ghaem hospital in Mashhad. Results show the proposed model has the best response to MS diagnosis from brain images with the highest level of certainty. Algorithm repeatability testing done to solve the uncertainty issue. Eventually, after k-fold cross validation, experimental accuracy in both sets was equal to %95 and %97. To solve the issue of uncertainty, testing the null hypothesis and repeatability of the algorithm were conducted. In general, after k-fold cross-validation, variables of accuracy, specificity, and sensitivity were obtained as 96.28%, 97.41%, and 96.53% respectively

Keywords:

Multiple Sclerosis (MS) , Pseudo Zernike Moment , Gray-level co-occurrence matrix (GLCM) , Support vector machine (SVM) , Genetic algorithm (GA) , Uncertainty