Analysis of Brain MRI Images Using Various Image Segmentation Techniques

Abstract

Tumor detection from brain MRI (Magnetic Resonance Imaging) is always a challenge newlinefor radiologists and researchers. Image segmentation is considered to be an efficient way newlineto extract out the tumor region in brain from the Magnetic Resonance Imaging (MRI) newlineimages. These images have severe irregularities and contrast issues, due to which the newlineradiologists and surgeons may get confused and prescribe unappropriated medical advice. newlineThe segmentation is proved to be a best solution for tumor detection, but a MRI machine newlinehandles lots of images in a day and expected to be efficient and effective every time. To newlineaddress such challenges, this thesis aims to understand, analyse and propose the solutions newlineof problems related to Brain MRI image segmentation. newlineThe recent literature on image segmentation has been studied thoroughly and presented newlinein the thesis to provide readers with latest and new perspectives related to the subject. The newlineresearch work is focussed on the analysis of brain MRI image segmentation with the newlineapplication of morphological operators. Manual segmentation is treated as the first step newlineof analysis to obtain gold standard for comparing the other semi- automatic and fully newlineautomatic segmentation algorithms. The morphological operations along with semi- newlineautomatic segmentation technique in brain MRI images are applied and analysed to newlinepropose Structural Element based Morphological Segmentation (SEMS) algorithm. The newlineimpact of different shaped structural elements of morphological operators are analysed newlineby applying it on segmented images in the proposed algorithm. Further, a CNN model newlinecapable of performing morphological segmentation on multiple images with high newlineaccuracy has also been proposed. The proposed Improved Fully Automatic Segmentation newline(IFAS) model has been trained and tested on multiple dataset and types of enhanced newlineimages, which in turn make it more reliable and robust. The experimentation and analysis has been conducted on the datasets extracted from online available open source Kaggle.