Automatic tumor detection in automated 3-D breast ultrasound using local phase information

Tumor localization in automated whole breast ultrasound (ABUS) images is an important and challenging task in a computer-aided detection (CAD) system. In this study, a completely automatic tumor detection method using the local phase information was proposed to accurately locate the tumor in ABUS images, thus accelerate the segmenting process. The phase in the max-energy orientation (PMO) image of the preprocessed ABUS image was acquired and binarized as well as the preprocessed image and the composed image obtained from the sum of the preprocessed image and the PMO image. Then morphologic postprocessing was performed to remove non-tumor regions and a region selection algorithm was used to get the tumor region from these three binary 3-D images and generate the volume of interests (VOI). The performance of the method was evaluated using a dataset of 40 cases and the sensitivity achieved 75% and 100% with false positives per volume of 0.25 and 2, respectively. Besides, the average precision rate and average recall rate of the detected VOI were 92.4% and 20.1%. The results proved that the proposed method could deal with various breast tumors and accurately locate the tumor.