A Complex Diffusion Based Modified Fuzzy C- Means Approach for Segmentation of Ultrasound Image in Presence of Speckle Noise for Breast Cancer Detection Subodh Srivastava 1* , Guddu Kumar 1 , Ritesh K. Mishra 1 , Niharika Kulshrestha 2 1 Department of Electronics and Communication Engineering, National Institute of Technology, Patna 800006, India 2 Banaras Hindu University, Varanasi 221005, India Corresponding Author Email: subodh@nitp.ac.in https://doi.org/10.18280/ria.340406 ABSTRACT Received: 9 July 2019 Accepted: 20 June 2020 This paper proposes a single framework for segmentation of abnormalities for breast cancer detection from Ultrasound images in presence of Rayleigh noise i.e. noise removal and segmentation are embedded in single step. It accomplishes dual purpose in a single framework simultaneously for the preprocessing and segmentation. The proposed framework comprises of two terms, first term, is used for segmentation which is a modified fuzzy c-means segmentation (MFCM) approach while second term is an adaptive complex diffusion based non linear filter (ACDPDE) that performs as regularization function for removal of Rayleigh noise, enhancement, and edge preservation of ultrasound Image. The various existing segmentation methods viz. K-Means, Texture based, Fuzzy C-Means (FCM), total variation based FCM (TVFCM), Adaptive fourth order PDE based FCM (AFPDEFCM), and the proposed method are evaluated for 50 sample ultrasound images of breast cancer. The region of interest (ROI) segmented image of ultrasound breast tissue is compared with ground truth images. From the acquired results and its analysis, it is observed that the proposed method is more robust and provides better segmentation result for ultrasound images in terms of various performance measures such as Global Constancy error (GCE), Tanimoto coefficient, Variation of Information (VOI), Probability Random Index (PRI), Jaccard coefficient, accuracy, True Positive Rate (TPR), False Positive Rate (FPR), True Negative Rate (TNR), dice index, False Negative Rate (FNR), and Area under curve (AUC). The proposed approach is capable of handling segmentation problem of blocky artifacts while achieving good tradeoff between Rayleigh noise removal and edge preservation. The proposed method may be useful for finding additional 33% cases of breast cancer which is missed or not detected by mammography. Keywords: fuzzy C means, complex diffusion, ultrasound image, speckle noise, Rayleigh noise 1. INTRODUCTION In past few decades, there have been an increasing number of cancer affected patients. If the detection of cancer is not done at the proper time interval that could directly leads to the risk of patient’s life. So, in order to avoid those crucial cases, early detection of the cancer is very necessary. Several types of methods have been adopted for early detection [1-3]. In this work, we are focusing on the breast cancer’s detection. Breast cancer is widely prevalent form of cancer and is the reason of several deaths. In the India, out of 11 every woman is having breast cancer. It is second major cause of death after lung cancer. 53 percent of the cancers can be detected through mammograms, Ultrasound can pick up 33 percent additional cases of breast cancer that are missed or not detected by mammogram [4-7]. There is wide variation in breast tissues from person to person. If the breast tissues are highly dense, mammograms can be difficult to interpret for tumor diagnosis. For this reason, mammography is not that much efficient for detecting cancer and can vary over a wide range. In contrast to this, ultrasound is useful because it can differentiate between fluid-filled cysts and solid masses [8]. To take out cancerous cell for testing purpose, ultrasound image may guide biopsy needle for inserting in exact cancerous area. Ultrasound is easily available, and is not harmful for a human as well as it is very cost effective [8]. Ultrasound images are playing very important role to detect additional cases of breast cancer which are missed by Mammograms [6, 7, 9]. Various segmentation techniques have been used in order to find out various defects or abnormalities. The objective of image segmentation is to divide an image into various regions while removing objects that remain uncovered. Most important issue is image understanding due to the variety and complexity of images. There are various methods for image segmentation like clustering [10], region growth [11], watershed transform [12], active contour model [13], Graph Cut [14], etc. Among these methodologies, clustering is one of the most popular methods used for image segmentation as a result of its adequacy and speed. The ultrasound image has several segmentation challenges like accurate segmentation, high quality segmentation with low computation cost, for malign and benign tumor. Ultrasound imaging is the most outstanding strategies which have been utilized because of its financially low cost and versatility. It may undergo many issues such as artifacts, which makes hard to see to translate the image and acquire quantitative data from them. Ultrasound images may contain additive and multiplicative noises. Multiplicative noise is referred as the speckle noise. Additive noise mostly arises due to electronic, electrical or thermal effect where as Revue d'Intelligence Artificielle Vol. 34, No. 4, August, 2020, pp. 419-427 Journal homepage: http://iieta.org/journals/ria 419