Neural network analysis applied to tumor segmentation on 3D breast ultrasound images

Our study presents a fully automatic tumor segmentation method using three-dimensional (3D) breast ultrasound (US) images. The proposed method is an approach based on 2D image processing techniques, which considers the variations of contours between two adjacent planes in a 3D dataset. In this approach, a reference image obtained in the previous plane was used to facilitate the segmentation in the next plane. To determine the initial reference image, we extracted five features from regions in each 2D slice and applied neural network analysis to discriminate the tumor from the background. Finally, three area error metrics were calculated to measure the overall performance of the system.