Study of a breast tumor extraction algorithm for three-dimensional ultrasound images, using multiple space differentiation filters: Automated breast tumor extraction algorithm with improved accuracy

Abstract

The system to be developed in the present study will enable us to extract the tumor region automatically from three-dimensional ultrasonic images of the breast, and differentiate benign from malignant tumors by using the characteristics of their surface form. In application of such a system, the accuracy of diagnosis greatly depends on its ability to extract tumor automatically. We developed an algorithm for determination of the tumor region using fuzzy reasoning, that is, we classified each voxel of three-dimensional images as “tumor,” “normal tissue” and “boundary,” and, using relaxation techniques to resolve regional contradictions, made final decisions as to the tumor region. It must be noted that, according to this algorithm, a three-dimensional the space differentiation filtering automatically generates a membership function for the fuzzy reasoning. Previous attempts of extracting tumor used a single LoG filter as the space differentiation filter. We recently developed a method which can cope with diverse ultrasound images more flexibly. With this new method, multiple DoG filters with varying characteristics are prepared in addition to the LoG filter, and the optimum one is selected from multiple extraction results. The introduction of this method improved the accuracy of extraction.