Segmentation of diffused type for ANA image based on edge model

This paper aimed to develop a nucleus segmentation technology in the context of Anti-Nuclear Antibodies (ANA) of diffused images. ANA images can be divided into two regions: the background and the nucleus. Then, an edge is taken between the background and the nucleus as the contour of the nucleus. This study proposes a two-group-method in order to enhance the image edges, and uses Sobel edge detection to calculate the edge gradient of the object. Determining the object's edge, by applying non-maximum suppression technology and hysteresis thresholding to the initial edge of the nucleus, enables us to obtain a complete closed curve or contour of the nucleus. This initial edge carries out three kinds of patches: mend approximate contour (MAC), mend contour for short-distance (MCSD), and mend contour for long-distance (MCLD), in order to complete optimal segmentation in regard to the nucleus. The medical images of ANA cells for this experimentation were provided by Dr. Xiezu Yi, MD, Department of Immunology and Rheumatology at Taichung Veterans General Hospital. The experimental results were separately analyzed and compared with Otsu's method, the Level Set method, and image segmentation software from ImageMaster.