Segmentation of tissues in MR images using Modified Spatial Fuzzy C Means algorithm

MRI brain images are widely used in medical applications for research, diagnosis, treatment, surgical planning and image guided surgeries. These MR brain images are often corrupted with Intensity Inhomogeneity artifact cause unwanted intensity variation due to non- uniformity in RF coils and Rician noise, the dominant noise in MRI due to thermal vibrations of electrons and ions and movement of objects during acquisition which may affect the performance of image processing techniques used for brain image analysis. Due to this type of artifact and noises, sometimes one type of normal tissue in MRI may be misclassified as other type of normal tissue and it leads to error during diagnosis. In this work, a method is proposed which automatically segments normal tissues such as White Matter, Gray Matter and Cerebrospinal Fluid from MR images with Rician noise and Intensity Inhomogeneity artifact. The proposed method consists of preprocessing using wrapping based curvelet transform to remove noise and Modified Spatial Fuzzy C Means segments normal tissues by considering spatial information because neighboring pixels are highly correlated and also construct initial membership matrix using initial cluster center by incorporating spatial neighborhood information to improve strength of clustering. This combination reduces Intensity Inhomogeneity artifact by providing higher segmentation accuracy. In this proposed work, the accuracy, sensitivity and specificity are improved with better segmentation over other previous methods.

I. INTRODUCTION

Accurate diagnosis in medical procedures has become widely attainable by the advent of the different medical imaging modalities. Among those, Magnetic resonance imaging (MRI) is a noninvasive, flexible imaging tool and does not require ionizing radiation like x-rays. It reveals information about human soft tissue anatomy that is not externally visible. Besides all these good properties, MRI also suffers from two main obstacles: Noises and Intensity Inhomogeneity. Intensity Inhomogeneity [1] is also known as intensity non-uniformity (or INU artifact), manifests as a spatially slowly varying function, that makes pixels belonging to the same tissue be observed having different intensities due to the failure of RF coil. Noise can be introduced by acquisition and transmission errors. Medical image segmentation is a very important issue in medical imaging because of these obstacles. In order to produce a correct segmentation of normal tissues such as, WM, GM and CSF, the Noise and INU artifact needs to be modeled and compensated. Several approaches have been used for MR images in order to segment tissues in human brain. Dzung L. Pham et al. [2] develops algorithm that is formulated by modifying the objective function in the Fuzzy C Means algorithm to include a multiplier field. But in the presence of extreme noise it may perform poorly. Ardizzone et al. [3] employ the FCM segmentation of MRI by using Gullied filter for pre-processing to remove inhomogeneity in the images. Bianrgi, P. M. et al [4] proposed an MRI segmentation using neural network based FCM clustering algorithm. The method applied on one channel MR data, however, whereas MR images are multi-spectral and provides additional information; due to noise and inhomogeneity this algorithm fails to work. Wang et al. [5] proposed that the Modified Fuzzy C-Means (MFCM) algorithm. It changes original FCM by using both local and nonlocal information, and a dissimilarity index was used instead of the usual distance metric. K. Sikka et al. [6] develops a fully automated algorithm for Modified FCM framework. The FCM based techniques in [7] employ Gaussian smoothing to produce a more homogeneous and low-noise subject to work with. This approach is, however, limited by the equal feature weights of the standard FCM. Neelum Noreen et al. [8] proposed MRI segmentation through wavelet and Fuzzy c means to remove inhomogeneity but it does not enhance edge detail. Iraky khalifa et al. [9] proposed Fuzzy C Means (FCM) Clustering by using wavelet Decomposition for feature extraction and feature vector treat as input to FCM. This gives better segmentation than Fuzzy C Means. Anil A Patil et al, [10] proposed Image denoising using curvelet transform with Bayes Shrink Soft Thresholding to improve smoothness and edge preservation. Priti Naik et al. [11] develops denoising method using curvelet transform with two thresholding methods such as, hard thresholding and partial reconstruction. Hence this concludes that, curvelet transform based on wrapping using hard thresholding provides faster and better way to denoise the noisy image. Mary Sugantharathnam et al. [12] develops Curvelet Denoising in various Imaging Modalities using Different Shrinkage Rules. The result proves that Hard thresholding of curvelet coefficients using Wrapping are efficient in the removal of different noises in biomedical images.