A systematic approach for noise removal and image segmentation for brain tumor analysis

Abstract

An automated restoration and classification methods for detection of tumors in different medical images demands high accuracy since it deals with human life. In automated medical diagnostic systems, magnetic resonance imaging (MRI) is a significant powerful imaging techniques and successful diagnostic tool developed to investigate the anatomical and physiological information of internal body parts gives better results than any other imaging modalities. It has the potential to determine the possible abnormalities in several parts of human body and therefore it is often used in clinical practice. MR image has noise during acquisition and leads to serious inaccuracies during classification. So there is a need of developing the restoration and segmentation algorithms for detecting tumor. The use of artificial intelligence techniques such as neural networks and fuzzy logic shows high potential in this field. Hence, in this thesis the neurofuzzy system has been applied for classification and detection purposes. MR image restoration and segmentation methods play a vital role in medical applications which represent the reasonable algorithms used for detecting the tumor. This thesis focuses on developing MRI restoration and segmentation to distinguish the presence of brain tumour or not. To determine the brain tumor, MR image restoration has been performed as a pre-processing step to obtain better restored image by removing the rician noise in MRI. In this thesis, two MR image restoration methods have been proposed by using different filters and hexagonal fuzzy membership function. The results of these methods are compared and better restored MR image has been taken for segmentation process to detect tumor. MRI segmentation has been implemented in two different methods using fuzzy, Rough Set Theory (RST) and Deep Convolutional Neural Network (DCNN). MR image restoration has been done using adaptive fuzzy hexagonal function with local order filter and Non Local Mean (NLM) filter. The fuzzyhexagonal membership function has