Edge detection of medical image processing using vector field analysis

Ultrasound (US) breast cancer images is one of the most complicated medical images to extract the desired area of interest. It is often difficult to separate the tumor region from the background tissues. Therefore, tumor segmentation is the challenging problems in the computed aided diagnosis. Among many image segmentation techniques, a generalized gradient vector flow (GGVF) method is one of the popular techniques. It is based on vector transformation of the edge map of the gray scale image. GGVF introduces a non-uniform diffusion to preserve the large gradient of the boundary area and smooth the gradients caused by noise and speckles. However, the improper numerical iteration of GGVF may lead the false contours or the existing noise and finally the snake could not reach the true boundary. In this paper, the new vector field analysis for breast tumor US image segmentation is proposed. The GGVF vector field will be derived from the edge map of the original image. The algorithm analyzes the GGVF vectors in terms of the entropy of the angle of vectors in the corresponding window. The windows will be vertically and horizontally flipped, then the entropy will be evaluated again. Next, the ratio of the entropy before and after flip will be determined to be the classifier of the boundary and non-boundary. The algorithm has been tested on the real US breast tumor images with a set of ground truth images hand-drawn by radiologists. The proposed algorithm is compared with conventional edge detectors such as Sobel and Canny operator. The numerical experiments show that the proposed techniques lead to a better segmentation accuracy with the reference to the conventional edge detection.