A Multimodal Fusion Model for Breast Tumor Segmentation in Ultrasound Images

Breast cancer is a leading cause of death among women worldwide. Accurate segmentation of breast tumors in Bmode ultrasound (BUS) images is crucial for early detection and diagnosis. However, this task remains challenging due to indistinct boundaries and artifacts. Existing methods using only BUS images often mistake artifacts in BUS images for tumors, posing challenges in accurately delineating tumor contours. To overcome the interference of artifacts and achieve more accurate segmentation, a multimodal method combining B-mode and elastography ultrasound images for breast tumor segmentation is proposed in this study. The proposed method utilizes BUS images and elastography as dual inputs of the network because elastography can depict the stiffness of breast tissue. The attention-separation-and-aggregation gate (ASA-Gate) module is implemented to achieve cross-modal feature fusion and extraction. Experimental results on an in-house dataset demonstrated the effectiveness of the method in reducing the interference of artifacts and improving the segmentation accuracy in BUS images.