Segmentation of 3D ovarian ultrasound volumes using continuous wavelet transform

Abstract

A novel algorithm for the segmentation of 3D ultrasound images of ovary is presented in this paper. The algorithm is based on continuous wavelet transform (CWT) and consists of two consecutive steps. In the first step, the centers of follicles are determined by tracing the local maxima from higher to lower scale in the wavelet transform of input images. The center of follicle appears as local maximum near value 0 when the size of the follicle corresponds to the scale of CWT. In the second step, the shape of the follicle is outlined. This is done by casting the rays in different directions from the center of the follicle in order to find its border. The position of border is connected with the wavelet scale and the position of the first local minimum on each ray. The method was tested on a small set of real 3D ultrasound images. The results were evaluated visually, since we do not have manually annotated images.