Image Inpainting to Improve the Registration Performance of Multiple Sclerosis (MS) Patient Brain with Brain Atlas

Abstract

Multiple sclerosis (MS) is an inflammatory autoimmune disease of the central nervous system which damage the myelin layer of White Matter (WM) and Grey Matter (GM). The loss of myelin layer (demyelination) exposes the WM and GM, which is viewed as lesions in the MRI brain scans. Loss of this layer will distort or interrupt the flow of signals from the brain to the parts of the body. To treat and monitor the progression of MS in standardized way, patient MRI brain scans are registered with brain atlas. Brain atlas is composed of serial sections along different anatomical planes of the healthy or diseased developing brain. However, in this registration step, the MS lesions create a strong distortion in the output transformation which creates a bias in registered image. In this thesis, we propose a novel image inpainting technique to reduce such bias. Image inpainting is used to reconstruct the lost or deteriorated parts of image data. We inpaint the MS lesions to make it appear like healthy tissue and register this inpainted MS brain with the brain atlas, and add the masked lesions afterwards. To evaluate the performance of our proposed inpainting algorithm, we employ a two-step evaluation process. Firstly, we inpaint distorted 2D images and artificial MS lesions in 3D MRI image data with our proposed and state-of-the-art methods. Secondly, we register the inpainted brain with an atlas and compare its performance with the ground truth. This two-step evaluation indicates that the proposed inpainted algorithm performs comparatively better than other state-of- the-art methods and it also increases the registration performance and significantly reduces the bias previously created by the MS lesions. The quantitative analysis has given the idea that comparative parameters known as dice and jaccard score are manifesting better performance with respect to the present ones.