An Automatic Energy-based Segmentation Method For Ultrasound Images

UltrasonograPh is widely used in the clinical imaging diagnosis, due to its non-invasive, radiation-free, real time, convenience, good reproducibility, low cost and so on. With the development of medical image processing technology, computer aided diagnosis(Computer-Aided Diagnosis, CAD) becomes very useful for the objective analysis of ultrasound image. Doctors to diagnose disease, not only observe the lesion region texture and analysis the lesion’s shape, but also contrast the lesion region and background. An algorithm which can achieve the exact segmentation of ultrasound images contributes doctors to obtaining the accurate outline of lesions from the complex ultrasound images.It also lays the foundation for the quantitative of target shape and outline and the assessment of benign and malignant tumors for the next step. So, it is an important step in CAD. Therefore, the research of ultrasound image automatic segmentation to realize exactly getting lesion area is important for CAD.This paper presents an automatic energy-based region growing method to automatically segment the tumor region in ultrasound images. First, using the superpixels located the image boundaries to construct a background dictionary. Through sparse reconstruction algorithm, it automatically selects seed points. Then, by constructing an energy function, it effectively controls the region growing process to make sure the segmentation region converge to the tumor edge position and keep the internal energy consistency to get the accurate segmentation results. In order to verify the accuracy of this algorithm, this paper adopts liver ultrasound images and breast ultrasound images for testing, and compares it with four popular segmentation methods, i.e. K-means, FCM, RGB and GVF. The experimental results show that the proposed algorithm in this paper can automatically segment the tumor region of ultrasound images and get a more accurate result.The main innovation points in this paper are shown as follow:1. In the ultrasound image segmentation, we use the superpixels as the basic operating unit to improve the efficiency of the algorithm. At the same time, it also improves ability of defending noise.2. We propose an algorithm for automatica seed selection by sparse reconstruction.3. We propose an automatic energy-based region growing method for ultrasound image segmentation. Based on the regional growth algorithm, we create a novel energy function to control the region growing process to get the more accurate segmentation results...