Accurate Segmentation of Breast Tumors in Ultrasound Images Using a Custom-Made Active Contour Model and Signal-to-Noise Ratio Variations

Outlining tumors in ultrasound B-mode images is an important process in many diagnostic procedures, but manual tumor segmentation is usually a time-consuming and challenging task. This paper presents a custom-made active contour model that is specifically designed for segmenting tumors in ultrasound images. The algorithm starts by drawing a circular contour around a manually selected point inside the tumor. The vertices of the circular contour are iteratively moved from the interior of the tumor to the tumor boundary. The motion of the vertices is controlled using an ultrasound-based statistical parameter, called the envelope signal-to-noise ratio (SNR), that is sensitive to variations in tissue anatomy. The proposed algorithm has been used to outline breast tumors in 10 ultrasound B-mode images. When compared to tumor outlines delineated by a human expert, the outlines obtained using the proposed algorithm achieved sensitivity values as high as 96.98%. The proposed algorithm provides a simple and accurate method for tumor segmentation in ultrasound images.

I. Introduction

The American Cancer Society estimated that for American women, breast cancer is the most common cancer when cancers of skin are excluded and is the second cause of cancer death [1]. Early detection of breast cancer is crucial for successful treatment and increasing the survival rate.