Synopsis of multiple segmentation results for breast lesion characterization

Abstract

When characterizing a tumor or lesion as malignant or benign, a system () receives an image of the lesion volume (), employs a processor () to perform a raw segmentation of the image, the results of which are stored to memory (). Then processor then executes a hole-filling procedure to fill in dark areas in the image of the lesion representing necrotic tissue that absorbed little or no contrast agent, and optionally a leakage removal procedure to remove image voxels associated with non-lesion tissue, e.g., blood vessels, in which the contrast agent was present during imaging, to generate a complete lesion volume. A voxel analyzer () assesses a number of voxels included in the raw segmentation of the lesion image, and the final segmentation (e.g., after filling and optional leakage removal). A segmentation comparator () computes a ratio of dark area voxels related to necrotic tissue detected after the raw segmentation to total voxels detected in the final image segmentation. The ratio is then used to determine a likelihood of malignancy, with a higher ratio indicating a higher likelihood.