الفهرس 
Research BackgroundOnly 14 pages are availabe for public view
 |  | from | 123 |  |  |
| | | from | 123 | | |
Abstract
Multiple sclerosis (MS) is a permanent harmful disease that destroys brain cells. It destroys the control ability of vision, balance, muscle control, and other basic body functions. Recently, Magnetic Resonance Imaging (MRI) is increasingly used nowadays in many medical applications. Manual segmentation of MS lesions in MR images by human specialists is time-consuming, subjective, and inclined to inter-expert inconstancy. In this manner, automatic segmentation is required as an elective to manual segmentation. In any case, utilizing automatic segmentation to distinguish MS lesion is exceptionally challenging. It must be amazingly precise since it influences individuals’ lives. A huge research effort has been carried out to automate the detection and diagnosis process of MS lesions using machine learning techniques. Research work in this area especially in large data faces many challenges like time consumption, data acquisition, memory insufficiency and low accuracy. Another important challenge is related to the training data set which is, in most cases, is unbalanced. Imbalanced data refers to a classification problem where the number of observations per class is not equally distributed, often have a large number of observations for one class. This problem exerts a major impact on the value and meaning of accuracy and on certain other well-known performance metrics such as dice similarity and sensitivity. In our study we consider these problem by introducing a new methodology that uses a hybrid machine learning model. This hybrid model combines Ensemble algorithm with a feature-based method to solve the problem of imbalanced classification data without loses. Two-dimension discrete wavelet transform (2D-DWT) and textural features are used to extract local information from MR image analysis. We propose two different hybrids
model: Ensemble Support Vector Machine (ESVM) and Ensemble Decision Tree (EDT). We detect MS twice using both two different models and compare the results. Results shows that the two different model gives the same accuracy. Considering the imbalance data challenge, our method is amongst the top performing solutions. The two proposed methods show high improvements in MS lesion detection compared with other studies provided.