In this study, we propose a hybrid approach involving feature extraction, feature selection, and optimized learning for the diagnosis of multiple sclerosis (MS), which can detect the lesion caused by MS plaques in the brain using magnetic resonance imaging analysis. A major challenge associated with lesion diagnosis by neurologists is that it is a timeconsuming process and demands high expertise; therefore, researchers have been stimulated to find an auto-diagnose method of the disease. Given the high resemblance of MS plaque-induced lesions and other lesions such as Alzheimer's or dementia, scant research has explored the diagnosis of MS-induced lesions, most of which suffering from the lack of an efficient and accurate method. Informed by the need for a precise hybrid model for the classification of MS plaques and other comparable lesions, a solution is proposed that utilizes an efficient model. In this method, after image preprocessing, the feature vector is created by applying fractal and Pseudo-Zernike Moments descriptors. Feature selection using the Difference Evolution) algorithm to select the minimum subset of features will reduce the number of Extreme Learning Machine (ELM) inputs for classification. To improve the classification effect, the ELM wavelet kernel parameters are also regulated by the Shuffled Frog-Leaping Algorithm. By applying the proposed model to a set of brain MR images obtained from healthy subjects and MS patients during different experimental iterations, an average accuracy of 97% was obtained. The results of the method were estimated under specific conditions, and finally the proposed model yielded desirable outputs compared to similar methods.