A NOVEL APPROACH OF COMBINING IMAGE REGISTRATION AND SEGMENTATION FOR LESION DETECTION IN BREAST PHANTOM IMAGES OBTAINED FROM FUSED FULL FIELD DIGITAL MAMMOGRAPHY ULTRASOUND SYSTEM (FFDMUS)

摘要

Follow-up patient mammography analysis is necessary to avoid biopsies. Image registration can help the follow-up analysis. The combined effect of registration and segmentation in FFDMUS framework may improve sensitivity and specificity of lesion detection. Our methodology consists of CIRS breast phantom image acquisition in FFDMUS framework. The phantom was rotated and translated to acquire digital images. The system corrected the rotated images using optimized entropy-based strategy. Gradient vector flow (GVF)-based segmentation model was used to measure the registration error. For 29 pairs of breast phantom images having a rotation of up to 15° between source and target images, the mean registration error was 1.41 pixels and the standard deviation was 0.79 pixels. We tested our protocol synthetically. The software was developed in C/C++.