Multiple Sclerosis (MS) is a chronic autoimmune disease that affects the central nervous system. It is characterized by the appearance of inflamed lesions. The diagnosis, as well as the prognosis of these lesions, are considered as tedious tasks and time consuming. The objective of this paper is to propose an automated method allowing to assist and detect MS lesions in 3D MRI acquisition. A deep learning network based on the Detectron-2 model has been used. The proposed model involves three main steps: preprocessing, feature extraction and MS detection. A database of 4500 images was acquired from the National Institute of Neurology Mongi Ben Hmida of Tunisia and was preprocessed in order to train the model. In the prognosis stage, chaotic features extracted from the segmented lesions are used in order to define pertinent clinical characteristics that will help neurologists to predict lesions progression in the majority of complicated cases. In the diagnosis stage, the proposed deep model achieves an average detection accuracy of 98% by evaluating the result on healthy and pathological images. Automatic segmentation results show that our proposed method has the ability to segment MS lesions with an average accuracy of 96.4%.