Ultrasound Image Despeckling and detection of Breast Cancer using Deep CNN

Breast Cancer is a common type of cancer diagnosed and it is a leading cause of death amongst the female population worldwide. Ultrasound imaging is the preferred method used by hospitals for detection of breast cancer, due to the fact that it is much safer that other imaging modalities. However, Ultrasound images are contaminated with noise that is non-Gaussian and multiplicative referred to as speckles. Currently, medical technicians and physicians do diagnosis of breast cancer by manually inspecting the ultrasound images, which makes the process time consuming and costly. This may be considered as an issue which prevents the early detection of breast cancer. Hence, an early diagnosis of breast cancer can be beneficial in not only prescribing medical procedure that inhibits the cancer from spreading but also in minimizing the fatality rate. Due to the Speckles (noise) in ultrasound, automatic detection and diagnosis is an extremely difficult task. In this paper, a Convolutional Neural Network (CNN) has been proposed for Despeckling (Denoising) the ultrasound images and afterwards another CNN model is proposed for the classification of the ultrasound images into benign and malignant classes. The proposed models are tested on a Mendeley Breast Ultrasound dataset. Experimental results indicate that a classification accuracy of 99.89% is achieved through the proposed model and that the proposed model(s) outperform other methods in proposed in recent studies.

I. Introduction

Cancer is considered to be one of the most fatal ailments and a recent statistical study conducted in United States [1] has rated cancer as the second major cause of death. The cells of human body naturally age, die and get replaced by new cells. In some cases, the cells tend to grow abnormally and form a mass generally referred as tumor. The breast cancer tumors can be classified into two types; cancerous (malignant) and non-cancerous (benign). The non-cancerous tumors do not affect nearby organs and tissues, whereas the cancerous tumors are capable to spread out into various organs and tissues. The common organs for origination of cancer include breasts, lungs, skin, etc. A global statistical report of cancer affected cases from 185 countries has been presented by International Agency for Research [2]. It shows that among various types of cancers, female breast cancer is one of the commonly diagnosed cancers with 11.6% cases out of total observations and is also the leading cause of death in females. Hence, in order to reduce the mortality rate due to breast cancer, it is of prime importance to acquire an early and accurate diagnosis accompanied with immediate medical treatment.